Making stemming optional. (#2791)

Fixed code and CI to run on no default features.

Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>

.github/workflows/coverage.yml

@@ -15,11 +15,11 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - name: Install Rust
-        run: rustup toolchain install nightly-2024-07-01 --profile minimal --component llvm-tools-preview
+        run: rustup toolchain install nightly-2025-12-01 --profile minimal --component llvm-tools-preview
       - uses: Swatinem/rust-cache@v2
       - uses: taiki-e/install-action@cargo-llvm-cov
       - name: Generate code coverage
-        run: cargo +nightly-2024-07-01 llvm-cov --all-features --workspace --doctests --lcov --output-path lcov.info
+        run: cargo +nightly-2025-12-01 llvm-cov --all-features --workspace --doctests --lcov --output-path lcov.info
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v3
         continue-on-error: true

.github/workflows/test.yml

@@ -39,11 +39,11 @@ jobs:
 
     - name: Check Formatting
       run: cargo +nightly fmt --all -- --check
-    
+
     - name: Check Stable Compilation
       run: cargo build --all-features
 
-    
+
     - name: Check Bench Compilation
       run: cargo +nightly bench --no-run --profile=dev --all-features
 
@@ -59,10 +59,10 @@ jobs:
 
     strategy:
       matrix:
-        features: [
-            { label: "all", flags: "mmap,stopwords,lz4-compression,zstd-compression,failpoints" },
-            { label: "quickwit", flags: "mmap,quickwit,failpoints" }
-        ]
+        features:
+          - { label: "all", flags: "mmap,stopwords,lz4-compression,zstd-compression,failpoints,stemmer" }
+          - { label: "quickwit", flags: "mmap,quickwit,failpoints" }
+          - { label: "none", flags: "" }
 
     name: test-${{ matrix.features.label}}
 
@@ -80,7 +80,21 @@ jobs:
     - uses: Swatinem/rust-cache@v2
 
     - name: Run tests
-      run: cargo +stable nextest run --features ${{ matrix.features.flags }} --verbose --workspace
+      run: |
+        # if matrix.feature.flags is empty then run on --lib to avoid compiling examples
+        # (as most of them rely on mmap) otherwise run all
+        if [ -z "${{ matrix.features.flags }}" ]; then
+          cargo +stable nextest run --lib --no-default-features --verbose --workspace
+        else
+          cargo +stable nextest run --features ${{ matrix.features.flags }} --no-default-features --verbose --workspace
+        fi
 
     - name: Run doctests
-      run: cargo +stable test --doc --features ${{ matrix.features.flags }} --verbose --workspace
+      run: |
+        # if matrix.feature.flags is empty then run on --lib to avoid compiling examples
+        # (as most of them rely on mmap) otherwise run all
+        if [ -z "${{ matrix.features.flags }}" ]; then
+          echo "no doctest for no feature flag"
+        else
+          cargo +stable test --doc --features ${{ matrix.features.flags }} --verbose --workspace
+        fi

Cargo.toml

@@ -37,7 +37,7 @@ fs4 = { version = "0.13.1", optional = true }
 levenshtein_automata = "0.2.1"
 uuid = { version = "1.0.0", features = ["v4", "serde"] }
 crossbeam-channel = "0.5.4"
-rust-stemmers = "1.2.0"
+rust-stemmers = { version = "1.2.0", optional = true }
 downcast-rs = "2.0.1"
 bitpacking = { version = "0.9.2", default-features = false, features = [
     "bitpacker4x",
@@ -56,7 +56,6 @@ itertools = "0.14.0"
 measure_time = "0.9.0"
 arc-swap = "1.5.0"
 bon = "3.3.1"
-i_triangle = "0.38.0"
 
 columnar = { version = "0.6", path = "./columnar", package = "tantivy-columnar" }
 sstable = { version = "0.6", path = "./sstable", package = "tantivy-sstable", optional = true }
@@ -71,18 +70,17 @@ futures-util = { version = "0.3.28", optional = true }
 futures-channel = { version = "0.3.28", optional = true }
 fnv = "1.0.7"
 typetag = "0.2.21"
-geo-types = "0.7.17"
 
 [target.'cfg(windows)'.dependencies]
 winapi = "0.3.9"
 
 [dev-dependencies]
-binggan = "0.14.0"
+binggan = "0.14.2"
 rand = "0.8.5"
 maplit = "1.0.2"
 matches = "0.1.9"
 pretty_assertions = "1.2.1"
-proptest = "1.0.0"
+proptest = "1.7.0"
 test-log = "0.2.10"
 futures = "0.3.21"
 paste = "1.0.11"
@@ -115,7 +113,8 @@ debug-assertions = true
 overflow-checks = true
 
 [features]
-default = ["mmap", "stopwords", "lz4-compression", "columnar-zstd-compression"]
+default = ["mmap", "stopwords", "lz4-compression", "columnar-zstd-compression", "stemmer"]
+stemmer = ["rust-stemmers"]
 mmap = ["fs4", "tempfile", "memmap2"]
 stopwords = []
 
@@ -175,6 +174,18 @@ harness = false
 name = "exists_json"
 harness = false
 
+[[bench]]
+name = "range_query"
+harness = false
+
 [[bench]]
 name = "and_or_queries"
 harness = false
+
+[[bench]]
+name = "range_queries"
+harness = false
+
+[[bench]]
+name = "bool_queries_with_range"
+harness = false

README.md

@@ -123,6 +123,7 @@ You can also find other bindings on [GitHub](https://github.com/search?q=tantivy
 - [seshat](https://github.com/matrix-org/seshat/): A matrix message database/indexer
 - [tantiny](https://github.com/baygeldin/tantiny): Tiny full-text search for Ruby
 - [lnx](https://github.com/lnx-search/lnx): adaptable, typo tolerant search engine with a REST API
+- [Bichon](https://github.com/rustmailer/bichon): A lightweight, high-performance Rust email archiver with WebUI
 - and [more](https://github.com/search?q=tantivy)!
 
 ### On average, how much faster is Tantivy compared to Lucene?

benches/agg_bench.rs

@@ -1,5 +1,6 @@
 use binggan::plugins::PeakMemAllocPlugin;
 use binggan::{black_box, InputGroup, PeakMemAlloc, INSTRUMENTED_SYSTEM};
+use rand::distributions::WeightedIndex;
 use rand::prelude::SliceRandom;
 use rand::rngs::StdRng;
 use rand::{Rng, SeedableRng};
@@ -54,12 +55,18 @@ fn bench_agg(mut group: InputGroup<Index>) {
     register!(group, extendedstats_f64);
     register!(group, percentiles_f64);
     register!(group, terms_few);
+    register!(group, terms_all_unique);
     register!(group, terms_many);
     register!(group, terms_many_top_1000);
     register!(group, terms_many_order_by_term);
     register!(group, terms_many_with_top_hits);
+    register!(group, terms_all_unique_with_avg_sub_agg);
     register!(group, terms_many_with_avg_sub_agg);
     register!(group, terms_few_with_avg_sub_agg);
+    register!(group, terms_status_with_avg_sub_agg);
+    register!(group, terms_status);
+    register!(group, terms_few_with_histogram);
+    register!(group, terms_status_with_histogram);
 
     register!(group, terms_many_json_mixed_type_with_avg_sub_agg);
 
@@ -132,12 +139,12 @@ fn extendedstats_f64(index: &Index) {
 }
 fn percentiles_f64(index: &Index) {
     let agg_req = json!({
-      "mypercentiles": {
-        "percentiles": {
-          "field": "score_f64",
-          "percents": [ 95, 99, 99.9 ]
+        "mypercentiles": {
+            "percentiles": {
+                "field": "score_f64",
+                "percents": [ 95, 99, 99.9 ]
+            }
         }
-      }
     });
     execute_agg(index, agg_req);
 }
@@ -174,6 +181,19 @@ fn terms_few(index: &Index) {
     });
     execute_agg(index, agg_req);
 }
+fn terms_status(index: &Index) {
+    let agg_req = json!({
+        "my_texts": { "terms": { "field": "text_few_terms_status" } },
+    });
+    execute_agg(index, agg_req);
+}
+fn terms_all_unique(index: &Index) {
+    let agg_req = json!({
+        "my_texts": { "terms": { "field": "text_all_unique_terms" } },
+    });
+    execute_agg(index, agg_req);
+}
+
 fn terms_many(index: &Index) {
     let agg_req = json!({
         "my_texts": { "terms": { "field": "text_many_terms" } },
@@ -222,6 +242,39 @@ fn terms_many_with_avg_sub_agg(index: &Index) {
     });
     execute_agg(index, agg_req);
 }
+fn terms_all_unique_with_avg_sub_agg(index: &Index) {
+    let agg_req = json!({
+        "my_texts": {
+            "terms": { "field": "text_all_unique_terms" },
+            "aggs": {
+                "average_f64": { "avg": { "field": "score_f64" } }
+            }
+        },
+    });
+    execute_agg(index, agg_req);
+}
+fn terms_few_with_histogram(index: &Index) {
+    let agg_req = json!({
+        "my_texts": {
+            "terms": { "field": "text_few_terms" },
+            "aggs": {
+                "histo": {"histogram": { "field": "score_f64", "interval": 10 }}
+            }
+        }
+    });
+    execute_agg(index, agg_req);
+}
+fn terms_status_with_histogram(index: &Index) {
+    let agg_req = json!({
+        "my_texts": {
+            "terms": { "field": "text_few_terms_status" },
+            "aggs": {
+                "histo": {"histogram": { "field": "score_f64", "interval": 10 }}
+            }
+        }
+    });
+    execute_agg(index, agg_req);
+}
 
 fn terms_few_with_avg_sub_agg(index: &Index) {
     let agg_req = json!({
@@ -234,6 +287,17 @@ fn terms_few_with_avg_sub_agg(index: &Index) {
     });
     execute_agg(index, agg_req);
 }
+fn terms_status_with_avg_sub_agg(index: &Index) {
+    let agg_req = json!({
+        "my_texts": {
+            "terms": { "field": "text_few_terms_status" },
+            "aggs": {
+                "average_f64": { "avg": { "field": "score_f64" } }
+            }
+        },
+    });
+    execute_agg(index, agg_req);
+}
 
 fn terms_many_json_mixed_type_with_avg_sub_agg(index: &Index) {
     let agg_req = json!({
@@ -419,14 +483,21 @@ fn get_test_index_bench(cardinality: Cardinality) -> tantivy::Result<Index> {
         .set_stored();
     let text_field = schema_builder.add_text_field("text", text_fieldtype);
     let json_field = schema_builder.add_json_field("json", FAST);
+    let text_field_all_unique_terms =
+        schema_builder.add_text_field("text_all_unique_terms", STRING | FAST);
+    let text_field_many_terms = schema_builder.add_text_field("text_many_terms", STRING | FAST);
     let text_field_many_terms = schema_builder.add_text_field("text_many_terms", STRING | FAST);
     let text_field_few_terms = schema_builder.add_text_field("text_few_terms", STRING | FAST);
+    let text_field_few_terms_status =
+        schema_builder.add_text_field("text_few_terms_status", STRING | FAST);
     let score_fieldtype = tantivy::schema::NumericOptions::default().set_fast();
     let score_field = schema_builder.add_u64_field("score", score_fieldtype.clone());
     let score_field_f64 = schema_builder.add_f64_field("score_f64", score_fieldtype.clone());
     let score_field_i64 = schema_builder.add_i64_field("score_i64", score_fieldtype);
     let index = Index::create_from_tempdir(schema_builder.build())?;
     let few_terms_data = ["INFO", "ERROR", "WARN", "DEBUG"];
+    // Approximate production log proportions: INFO dominant, WARN and DEBUG occasional, ERROR rare.
+    let log_level_distribution = WeightedIndex::new([80u32, 3, 12, 5]).unwrap();
 
     let lg_norm = rand_distr::LogNormal::new(2.996f64, 0.979f64).unwrap();
 
@@ -442,15 +513,21 @@ fn get_test_index_bench(cardinality: Cardinality) -> tantivy::Result<Index> {
             index_writer.add_document(doc!())?;
         }
         if cardinality == Cardinality::Multivalued {
+            let log_level_sample_a = few_terms_data[log_level_distribution.sample(&mut rng)];
+            let log_level_sample_b = few_terms_data[log_level_distribution.sample(&mut rng)];
             index_writer.add_document(doc!(
                 json_field => json!({"mixed_type": 10.0}),
                 json_field => json!({"mixed_type": 10.0}),
                 text_field => "cool",
                 text_field => "cool",
+                text_field_all_unique_terms => "cool",
+                text_field_all_unique_terms => "coolo",
                 text_field_many_terms => "cool",
                 text_field_many_terms => "cool",
                 text_field_few_terms => "cool",
                 text_field_few_terms => "cool",
+                text_field_few_terms_status => log_level_sample_a,
+                text_field_few_terms_status => log_level_sample_b,
                 score_field => 1u64,
                 score_field => 1u64,
                 score_field_f64 => lg_norm.sample(&mut rng),
@@ -475,8 +552,10 @@ fn get_test_index_bench(cardinality: Cardinality) -> tantivy::Result<Index> {
             index_writer.add_document(doc!(
                 text_field => "cool",
                 json_field => json,
+                text_field_all_unique_terms => format!("unique_term_{}", rng.gen::<u64>()),
                 text_field_many_terms => many_terms_data.choose(&mut rng).unwrap().to_string(),
                 text_field_few_terms => few_terms_data.choose(&mut rng).unwrap().to_string(),
+                text_field_few_terms_status => few_terms_data[log_level_distribution.sample(&mut rng)],
                 score_field => val as u64,
                 score_field_f64 => lg_norm.sample(&mut rng),
                 score_field_i64 => val as i64,

benches/bool_queries_with_range.rs

@@ -0,0 +1,288 @@
+use binggan::{black_box, BenchGroup, BenchRunner};
+use rand::prelude::*;
+use rand::rngs::StdRng;
+use rand::SeedableRng;
+use tantivy::collector::{Collector, Count, DocSetCollector, TopDocs};
+use tantivy::query::{Query, QueryParser};
+use tantivy::schema::{Schema, FAST, INDEXED, TEXT};
+use tantivy::{doc, Index, Order, ReloadPolicy, Searcher};
+
+#[derive(Clone)]
+struct BenchIndex {
+    #[allow(dead_code)]
+    index: Index,
+    searcher: Searcher,
+    query_parser: QueryParser,
+}
+
+fn build_shared_indices(num_docs: usize, p_title_a: f32, distribution: &str) -> BenchIndex {
+    // Unified schema
+    let mut schema_builder = Schema::builder();
+    let f_title = schema_builder.add_text_field("title", TEXT);
+    let f_num_rand = schema_builder.add_u64_field("num_rand", INDEXED);
+    let f_num_asc = schema_builder.add_u64_field("num_asc", INDEXED);
+    let f_num_rand_fast = schema_builder.add_u64_field("num_rand_fast", INDEXED | FAST);
+    let f_num_asc_fast = schema_builder.add_u64_field("num_asc_fast", INDEXED | FAST);
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema.clone());
+
+    // Populate index with stable RNG for reproducibility.
+    let mut rng = StdRng::from_seed([7u8; 32]);
+
+    {
+        let mut writer = index.writer_with_num_threads(1, 4_000_000_000).unwrap();
+
+        match distribution {
+            "dense" => {
+                for doc_id in 0..num_docs {
+                    // Always add title to avoid empty documents
+                    let title_token = if rng.gen_bool(p_title_a as f64) {
+                        "a"
+                    } else {
+                        "b"
+                    };
+
+                    let num_rand = rng.gen_range(0u64..1000u64);
+
+                    let num_asc = (doc_id / 10000) as u64;
+
+                    writer
+                        .add_document(doc!(
+                            f_title=>title_token,
+                            f_num_rand=>num_rand,
+                            f_num_asc=>num_asc,
+                            f_num_rand_fast=>num_rand,
+                            f_num_asc_fast=>num_asc,
+                        ))
+                        .unwrap();
+                }
+            }
+            "sparse" => {
+                for doc_id in 0..num_docs {
+                    // Always add title to avoid empty documents
+                    let title_token = if rng.gen_bool(p_title_a as f64) {
+                        "a"
+                    } else {
+                        "b"
+                    };
+
+                    let num_rand = rng.gen_range(0u64..10000000u64);
+
+                    let num_asc = doc_id as u64;
+
+                    writer
+                        .add_document(doc!(
+                            f_title=>title_token,
+                            f_num_rand=>num_rand,
+                            f_num_asc=>num_asc,
+                            f_num_rand_fast=>num_rand,
+                            f_num_asc_fast=>num_asc,
+                        ))
+                        .unwrap();
+                }
+            }
+            _ => {
+                panic!("Unsupported distribution type");
+            }
+        }
+        writer.commit().unwrap();
+    }
+
+    // Prepare reader/searcher once.
+    let reader = index
+        .reader_builder()
+        .reload_policy(ReloadPolicy::Manual)
+        .try_into()
+        .unwrap();
+    let searcher = reader.searcher();
+
+    // Build query parser for title field
+    let qp_title = QueryParser::for_index(&index, vec![f_title]);
+
+    BenchIndex {
+        index,
+        searcher,
+        query_parser: qp_title,
+    }
+}
+
+fn main() {
+    // Prepare corpora with varying scenarios
+    let scenarios = vec![
+        (
+            "dense and 99% a".to_string(),
+            10_000_000,
+            0.99,
+            "dense",
+            0,
+            9,
+        ),
+        (
+            "dense and 99% a".to_string(),
+            10_000_000,
+            0.99,
+            "dense",
+            990,
+            999,
+        ),
+        (
+            "sparse and 99% a".to_string(),
+            10_000_000,
+            0.99,
+            "sparse",
+            0,
+            9,
+        ),
+        (
+            "sparse and 99% a".to_string(),
+            10_000_000,
+            0.99,
+            "sparse",
+            9_999_990,
+            9_999_999,
+        ),
+    ];
+
+    let mut runner = BenchRunner::new();
+    for (scenario_id, n, p_title_a, num_rand_distribution, range_low, range_high) in scenarios {
+        // Build index for this scenario
+        let bench_index = build_shared_indices(n, p_title_a, num_rand_distribution);
+
+        // Create benchmark group
+        let mut group = runner.new_group();
+
+        // Now set the name (this moves scenario_id)
+        group.set_name(scenario_id);
+
+        // Define all four field types
+        let field_names = ["num_rand", "num_asc", "num_rand_fast", "num_asc_fast"];
+
+        // Define the three terms we want to test with
+        let terms = ["a", "b", "z"];
+
+        // Generate all combinations of terms and field names
+        let mut queries = Vec::new();
+        for &term in &terms {
+            for &field_name in &field_names {
+                let query_str = format!(
+                    "{} AND {}:[{} TO {}]",
+                    term, field_name, range_low, range_high
+                );
+                queries.push((query_str, field_name.to_string()));
+            }
+        }
+
+        let query_str = format!(
+            "{}:[{} TO {}] AND {}:[{} TO {}]",
+            "num_rand_fast", range_low, range_high, "num_asc_fast", range_low, range_high
+        );
+        queries.push((query_str, "num_asc_fast".to_string()));
+
+        // Run all benchmark tasks for each query and its corresponding field name
+        for (query_str, field_name) in queries {
+            run_benchmark_tasks(&mut group, &bench_index, &query_str, &field_name);
+        }
+
+        group.run();
+    }
+}
+
+/// Run all benchmark tasks for a given query string and field name
+fn run_benchmark_tasks(
+    bench_group: &mut BenchGroup,
+    bench_index: &BenchIndex,
+    query_str: &str,
+    field_name: &str,
+) {
+    // Test count
+    add_bench_task(bench_group, bench_index, query_str, Count, "count");
+
+    // Test all results
+    add_bench_task(
+        bench_group,
+        bench_index,
+        query_str,
+        DocSetCollector,
+        "all results",
+    );
+
+    // Test top 100 by the field (if it's a FAST field)
+    if field_name.ends_with("_fast") {
+        // Ascending order
+        {
+            let collector_name = format!("top100_by_{}_asc", field_name);
+            let field_name_owned = field_name.to_string();
+            add_bench_task(
+                bench_group,
+                bench_index,
+                query_str,
+                TopDocs::with_limit(100).order_by_fast_field::<u64>(field_name_owned, Order::Asc),
+                &collector_name,
+            );
+        }
+
+        // Descending order
+        {
+            let collector_name = format!("top100_by_{}_desc", field_name);
+            let field_name_owned = field_name.to_string();
+            add_bench_task(
+                bench_group,
+                bench_index,
+                query_str,
+                TopDocs::with_limit(100).order_by_fast_field::<u64>(field_name_owned, Order::Desc),
+                &collector_name,
+            );
+        }
+    }
+}
+
+fn add_bench_task<C: Collector + 'static>(
+    bench_group: &mut BenchGroup,
+    bench_index: &BenchIndex,
+    query_str: &str,
+    collector: C,
+    collector_name: &str,
+) {
+    let task_name = format!("{}_{}", query_str.replace(" ", "_"), collector_name);
+    let query = bench_index.query_parser.parse_query(query_str).unwrap();
+    let search_task = SearchTask {
+        searcher: bench_index.searcher.clone(),
+        collector,
+        query,
+    };
+    bench_group.register(task_name, move |_| black_box(search_task.run()));
+}
+
+struct SearchTask<C: Collector> {
+    searcher: Searcher,
+    collector: C,
+    query: Box<dyn Query>,
+}
+
+impl<C: Collector> SearchTask<C> {
+    #[inline(never)]
+    pub fn run(&self) -> usize {
+        let result = self.searcher.search(&self.query, &self.collector).unwrap();
+        if let Some(count) = (&result as &dyn std::any::Any).downcast_ref::<usize>() {
+            *count
+        } else if let Some(top_docs) = (&result as &dyn std::any::Any)
+            .downcast_ref::<Vec<(Option<u64>, tantivy::DocAddress)>>()
+        {
+            top_docs.len()
+        } else if let Some(top_docs) =
+            (&result as &dyn std::any::Any).downcast_ref::<Vec<(u64, tantivy::DocAddress)>>()
+        {
+            top_docs.len()
+        } else if let Some(doc_set) = (&result as &dyn std::any::Any)
+            .downcast_ref::<std::collections::HashSet<tantivy::DocAddress>>()
+        {
+            doc_set.len()
+        } else {
+            eprintln!(
+                "Unknown collector result type: {:?}",
+                std::any::type_name::<C::Fruit>()
+            );
+            0
+        }
+    }
+}

benches/range_queries.rs

@@ -0,0 +1,365 @@
+use std::ops::Bound;
+
+use binggan::{black_box, BenchGroup, BenchRunner};
+use rand::prelude::*;
+use rand::rngs::StdRng;
+use rand::SeedableRng;
+use tantivy::collector::{Count, DocSetCollector, TopDocs};
+use tantivy::query::RangeQuery;
+use tantivy::schema::{Schema, FAST, INDEXED};
+use tantivy::{doc, Index, Order, ReloadPolicy, Searcher, Term};
+
+#[derive(Clone)]
+struct BenchIndex {
+    #[allow(dead_code)]
+    index: Index,
+    searcher: Searcher,
+}
+
+fn build_shared_indices(num_docs: usize, distribution: &str) -> BenchIndex {
+    // Schema with fast fields only
+    let mut schema_builder = Schema::builder();
+    let f_num_rand_fast = schema_builder.add_u64_field("num_rand_fast", INDEXED | FAST);
+    let f_num_asc_fast = schema_builder.add_u64_field("num_asc_fast", INDEXED | FAST);
+    let schema = schema_builder.build();
+    let index = Index::create_in_ram(schema.clone());
+
+    // Populate index with stable RNG for reproducibility.
+    let mut rng = StdRng::from_seed([7u8; 32]);
+
+    {
+        let mut writer = index.writer_with_num_threads(1, 4_000_000_000).unwrap();
+
+        match distribution {
+            "dense" => {
+                for doc_id in 0..num_docs {
+                    let num_rand = rng.gen_range(0u64..1000u64);
+                    let num_asc = (doc_id / 10000) as u64;
+
+                    writer
+                        .add_document(doc!(
+                            f_num_rand_fast=>num_rand,
+                            f_num_asc_fast=>num_asc,
+                        ))
+                        .unwrap();
+                }
+            }
+            "sparse" => {
+                for doc_id in 0..num_docs {
+                    let num_rand = rng.gen_range(0u64..10000000u64);
+                    let num_asc = doc_id as u64;
+
+                    writer
+                        .add_document(doc!(
+                            f_num_rand_fast=>num_rand,
+                            f_num_asc_fast=>num_asc,
+                        ))
+                        .unwrap();
+                }
+            }
+            _ => {
+                panic!("Unsupported distribution type");
+            }
+        }
+        writer.commit().unwrap();
+    }
+
+    // Prepare reader/searcher once.
+    let reader = index
+        .reader_builder()
+        .reload_policy(ReloadPolicy::Manual)
+        .try_into()
+        .unwrap();
+    let searcher = reader.searcher();
+
+    BenchIndex { index, searcher }
+}
+
+fn main() {
+    // Prepare corpora with varying scenarios
+    let scenarios = vec![
+        // Dense distribution - random values in small range (0-999)
+        (
+            "dense_values_search_low_value_range".to_string(),
+            10_000_000,
+            "dense",
+            0,
+            9,
+        ),
+        (
+            "dense_values_search_high_value_range".to_string(),
+            10_000_000,
+            "dense",
+            990,
+            999,
+        ),
+        (
+            "dense_values_search_out_of_range".to_string(),
+            10_000_000,
+            "dense",
+            1000,
+            1002,
+        ),
+        (
+            "sparse_values_search_low_value_range".to_string(),
+            10_000_000,
+            "sparse",
+            0,
+            9,
+        ),
+        (
+            "sparse_values_search_high_value_range".to_string(),
+            10_000_000,
+            "sparse",
+            9_999_990,
+            9_999_999,
+        ),
+        (
+            "sparse_values_search_out_of_range".to_string(),
+            10_000_000,
+            "sparse",
+            10_000_000,
+            10_000_002,
+        ),
+    ];
+
+    let mut runner = BenchRunner::new();
+    for (scenario_id, n, num_rand_distribution, range_low, range_high) in scenarios {
+        // Build index for this scenario
+        let bench_index = build_shared_indices(n, num_rand_distribution);
+
+        // Create benchmark group
+        let mut group = runner.new_group();
+
+        // Now set the name (this moves scenario_id)
+        group.set_name(scenario_id);
+
+        // Define fast field types
+        let field_names = ["num_rand_fast", "num_asc_fast"];
+
+        // Generate range queries for fast fields
+        for &field_name in &field_names {
+            // Create the range query
+            let field = bench_index.searcher.schema().get_field(field_name).unwrap();
+            let lower_term = Term::from_field_u64(field, range_low);
+            let upper_term = Term::from_field_u64(field, range_high);
+
+            let query = RangeQuery::new(Bound::Included(lower_term), Bound::Included(upper_term));
+
+            run_benchmark_tasks(
+                &mut group,
+                &bench_index,
+                query,
+                field_name,
+                range_low,
+                range_high,
+            );
+        }
+
+        group.run();
+    }
+}
+
+/// Run all benchmark tasks for a given range query and field name
+fn run_benchmark_tasks(
+    bench_group: &mut BenchGroup,
+    bench_index: &BenchIndex,
+    query: RangeQuery,
+    field_name: &str,
+    range_low: u64,
+    range_high: u64,
+) {
+    // Test count
+    add_bench_task_count(
+        bench_group,
+        bench_index,
+        query.clone(),
+        "count",
+        field_name,
+        range_low,
+        range_high,
+    );
+
+    // Test top 100 by the field (ascending order)
+    {
+        let collector_name = format!("top100_by_{}_asc", field_name);
+        let field_name_owned = field_name.to_string();
+        add_bench_task_top100_asc(
+            bench_group,
+            bench_index,
+            query.clone(),
+            &collector_name,
+            field_name,
+            range_low,
+            range_high,
+            field_name_owned,
+        );
+    }
+
+    // Test top 100 by the field (descending order)
+    {
+        let collector_name = format!("top100_by_{}_desc", field_name);
+        let field_name_owned = field_name.to_string();
+        add_bench_task_top100_desc(
+            bench_group,
+            bench_index,
+            query,
+            &collector_name,
+            field_name,
+            range_low,
+            range_high,
+            field_name_owned,
+        );
+    }
+}
+
+fn add_bench_task_count(
+    bench_group: &mut BenchGroup,
+    bench_index: &BenchIndex,
+    query: RangeQuery,
+    collector_name: &str,
+    field_name: &str,
+    range_low: u64,
+    range_high: u64,
+) {
+    let task_name = format!(
+        "range_{}_[{} TO {}]_{}",
+        field_name, range_low, range_high, collector_name
+    );
+
+    let search_task = CountSearchTask {
+        searcher: bench_index.searcher.clone(),
+        query,
+    };
+    bench_group.register(task_name, move |_| black_box(search_task.run()));
+}
+
+fn add_bench_task_docset(
+    bench_group: &mut BenchGroup,
+    bench_index: &BenchIndex,
+    query: RangeQuery,
+    collector_name: &str,
+    field_name: &str,
+    range_low: u64,
+    range_high: u64,
+) {
+    let task_name = format!(
+        "range_{}_[{} TO {}]_{}",
+        field_name, range_low, range_high, collector_name
+    );
+
+    let search_task = DocSetSearchTask {
+        searcher: bench_index.searcher.clone(),
+        query,
+    };
+    bench_group.register(task_name, move |_| black_box(search_task.run()));
+}
+
+fn add_bench_task_top100_asc(
+    bench_group: &mut BenchGroup,
+    bench_index: &BenchIndex,
+    query: RangeQuery,
+    collector_name: &str,
+    field_name: &str,
+    range_low: u64,
+    range_high: u64,
+    field_name_owned: String,
+) {
+    let task_name = format!(
+        "range_{}_[{} TO {}]_{}",
+        field_name, range_low, range_high, collector_name
+    );
+
+    let search_task = Top100AscSearchTask {
+        searcher: bench_index.searcher.clone(),
+        query,
+        field_name: field_name_owned,
+    };
+    bench_group.register(task_name, move |_| black_box(search_task.run()));
+}
+
+fn add_bench_task_top100_desc(
+    bench_group: &mut BenchGroup,
+    bench_index: &BenchIndex,
+    query: RangeQuery,
+    collector_name: &str,
+    field_name: &str,
+    range_low: u64,
+    range_high: u64,
+    field_name_owned: String,
+) {
+    let task_name = format!(
+        "range_{}_[{} TO {}]_{}",
+        field_name, range_low, range_high, collector_name
+    );
+
+    let search_task = Top100DescSearchTask {
+        searcher: bench_index.searcher.clone(),
+        query,
+        field_name: field_name_owned,
+    };
+    bench_group.register(task_name, move |_| black_box(search_task.run()));
+}
+
+struct CountSearchTask {
+    searcher: Searcher,
+    query: RangeQuery,
+}
+
+impl CountSearchTask {
+    #[inline(never)]
+    pub fn run(&self) -> usize {
+        self.searcher.search(&self.query, &Count).unwrap()
+    }
+}
+
+struct DocSetSearchTask {
+    searcher: Searcher,
+    query: RangeQuery,
+}
+
+impl DocSetSearchTask {
+    #[inline(never)]
+    pub fn run(&self) -> usize {
+        let result = self.searcher.search(&self.query, &DocSetCollector).unwrap();
+        result.len()
+    }
+}
+
+struct Top100AscSearchTask {
+    searcher: Searcher,
+    query: RangeQuery,
+    field_name: String,
+}
+
+impl Top100AscSearchTask {
+    #[inline(never)]
+    pub fn run(&self) -> usize {
+        let collector =
+            TopDocs::with_limit(100).order_by_fast_field::<u64>(&self.field_name, Order::Asc);
+        let result = self.searcher.search(&self.query, &collector).unwrap();
+        for (_score, doc_address) in &result {
+            let _doc: tantivy::TantivyDocument = self.searcher.doc(*doc_address).unwrap();
+        }
+        result.len()
+    }
+}
+
+struct Top100DescSearchTask {
+    searcher: Searcher,
+    query: RangeQuery,
+    field_name: String,
+}
+
+impl Top100DescSearchTask {
+    #[inline(never)]
+    pub fn run(&self) -> usize {
+        let collector =
+            TopDocs::with_limit(100).order_by_fast_field::<u64>(&self.field_name, Order::Desc);
+        let result = self.searcher.search(&self.query, &collector).unwrap();
+        for (_score, doc_address) in &result {
+            let _doc: tantivy::TantivyDocument = self.searcher.doc(*doc_address).unwrap();
+        }
+        result.len()
+    }
+}

benches/range_query.rs

@@ -0,0 +1,260 @@
+use std::fmt::Display;
+use std::net::Ipv6Addr;
+use std::ops::RangeInclusive;
+
+use binggan::plugins::PeakMemAllocPlugin;
+use binggan::{black_box, BenchRunner, OutputValue, PeakMemAlloc, INSTRUMENTED_SYSTEM};
+use columnar::MonotonicallyMappableToU128;
+use rand::rngs::StdRng;
+use rand::{Rng, SeedableRng};
+use tantivy::collector::{Count, TopDocs};
+use tantivy::query::QueryParser;
+use tantivy::schema::*;
+use tantivy::{doc, Index};
+
+#[global_allocator]
+pub static GLOBAL: &PeakMemAlloc<std::alloc::System> = &INSTRUMENTED_SYSTEM;
+
+fn main() {
+    bench_range_query();
+}
+
+fn bench_range_query() {
+    let index = get_index_0_to_100();
+    let mut runner = BenchRunner::new();
+    runner.add_plugin(PeakMemAllocPlugin::new(GLOBAL));
+
+    runner.set_name("range_query on u64");
+    let field_name_and_descr: Vec<_> = vec![
+        ("id", "Single Valued Range Field"),
+        ("ids", "Multi Valued Range Field"),
+    ];
+    let range_num_hits = vec![
+        ("90_percent", get_90_percent()),
+        ("10_percent", get_10_percent()),
+        ("1_percent", get_1_percent()),
+    ];
+
+    test_range(&mut runner, &index, &field_name_and_descr, range_num_hits);
+
+    runner.set_name("range_query on ip");
+    let field_name_and_descr: Vec<_> = vec![
+        ("ip", "Single Valued Range Field"),
+        ("ips", "Multi Valued Range Field"),
+    ];
+    let range_num_hits = vec![
+        ("90_percent", get_90_percent_ip()),
+        ("10_percent", get_10_percent_ip()),
+        ("1_percent", get_1_percent_ip()),
+    ];
+
+    test_range(&mut runner, &index, &field_name_and_descr, range_num_hits);
+}
+
+fn test_range<T: Display>(
+    runner: &mut BenchRunner,
+    index: &Index,
+    field_name_and_descr: &[(&str, &str)],
+    range_num_hits: Vec<(&str, RangeInclusive<T>)>,
+) {
+    for (field, suffix) in field_name_and_descr {
+        let term_num_hits = vec![
+            ("", ""),
+            ("1_percent", "veryfew"),
+            ("10_percent", "few"),
+            ("90_percent", "most"),
+        ];
+        let mut group = runner.new_group();
+        group.set_name(suffix);
+        // all intersect combinations
+        for (range_name, range) in &range_num_hits {
+            for (term_name, term) in &term_num_hits {
+                let index = &index;
+                let test_name = if term_name.is_empty() {
+                    format!("id_range_hit_{}", range_name)
+                } else {
+                    format!(
+                        "id_range_hit_{}_intersect_with_term_{}",
+                        range_name, term_name
+                    )
+                };
+                group.register(test_name, move |_| {
+                    let query = if term_name.is_empty() {
+                        "".to_string()
+                    } else {
+                        format!("AND id_name:{}", term)
+                    };
+                    black_box(execute_query(field, range, &query, index));
+                });
+            }
+        }
+        group.run();
+    }
+}
+
+fn get_index_0_to_100() -> Index {
+    let mut rng = StdRng::from_seed([1u8; 32]);
+    let num_vals = 100_000;
+    let docs: Vec<_> = (0..num_vals)
+        .map(|_i| {
+            let id_name = if rng.gen_bool(0.01) {
+                "veryfew".to_string() // 1%
+            } else if rng.gen_bool(0.1) {
+                "few".to_string() // 9%
+            } else {
+                "most".to_string() // 90%
+            };
+            Doc {
+                id_name,
+                id: rng.gen_range(0..100),
+                // Multiply by 1000, so that we create most buckets in the compact space
+                // The benches depend on this range to select n-percent of elements with the
+                // methods below.
+                ip: Ipv6Addr::from_u128(rng.gen_range(0..100) * 1000),
+            }
+        })
+        .collect();
+
+    create_index_from_docs(&docs)
+}
+
+#[derive(Clone, Debug)]
+pub struct Doc {
+    pub id_name: String,
+    pub id: u64,
+    pub ip: Ipv6Addr,
+}
+
+pub fn create_index_from_docs(docs: &[Doc]) -> Index {
+    let mut schema_builder = Schema::builder();
+    let id_u64_field = schema_builder.add_u64_field("id", INDEXED | STORED | FAST);
+    let ids_u64_field =
+        schema_builder.add_u64_field("ids", NumericOptions::default().set_fast().set_indexed());
+
+    let id_f64_field = schema_builder.add_f64_field("id_f64", INDEXED | STORED | FAST);
+    let ids_f64_field = schema_builder.add_f64_field(
+        "ids_f64",
+        NumericOptions::default().set_fast().set_indexed(),
+    );
+
+    let id_i64_field = schema_builder.add_i64_field("id_i64", INDEXED | STORED | FAST);
+    let ids_i64_field = schema_builder.add_i64_field(
+        "ids_i64",
+        NumericOptions::default().set_fast().set_indexed(),
+    );
+
+    let text_field = schema_builder.add_text_field("id_name", STRING | STORED);
+    let text_field2 = schema_builder.add_text_field("id_name_fast", STRING | STORED | FAST);
+
+    let ip_field = schema_builder.add_ip_addr_field("ip", FAST);
+    let ips_field = schema_builder.add_ip_addr_field("ips", FAST);
+
+    let schema = schema_builder.build();
+
+    let index = Index::create_in_ram(schema);
+
+    {
+        let mut index_writer = index.writer_with_num_threads(1, 50_000_000).unwrap();
+        for doc in docs.iter() {
+            index_writer
+                .add_document(doc!(
+                    ids_i64_field => doc.id as i64,
+                    ids_i64_field => doc.id as i64,
+                    ids_f64_field => doc.id as f64,
+                    ids_f64_field => doc.id as f64,
+                    ids_u64_field => doc.id,
+                    ids_u64_field => doc.id,
+                    id_u64_field => doc.id,
+                    id_f64_field => doc.id as f64,
+                    id_i64_field => doc.id as i64,
+                    text_field => doc.id_name.to_string(),
+                    text_field2 => doc.id_name.to_string(),
+                    ips_field => doc.ip,
+                    ips_field => doc.ip,
+                    ip_field => doc.ip,
+                ))
+                .unwrap();
+        }
+
+        index_writer.commit().unwrap();
+    }
+    index
+}
+
+fn get_90_percent() -> RangeInclusive<u64> {
+    0..=90
+}
+
+fn get_10_percent() -> RangeInclusive<u64> {
+    0..=10
+}
+
+fn get_1_percent() -> RangeInclusive<u64> {
+    10..=10
+}
+
+fn get_90_percent_ip() -> RangeInclusive<Ipv6Addr> {
+    let start = Ipv6Addr::from_u128(0);
+    let end = Ipv6Addr::from_u128(90 * 1000);
+    start..=end
+}
+
+fn get_10_percent_ip() -> RangeInclusive<Ipv6Addr> {
+    let start = Ipv6Addr::from_u128(0);
+    let end = Ipv6Addr::from_u128(10 * 1000);
+    start..=end
+}
+
+fn get_1_percent_ip() -> RangeInclusive<Ipv6Addr> {
+    let start = Ipv6Addr::from_u128(10 * 1000);
+    let end = Ipv6Addr::from_u128(10 * 1000);
+    start..=end
+}
+
+struct NumHits {
+    count: usize,
+}
+impl OutputValue for NumHits {
+    fn column_title() -> &'static str {
+        "NumHits"
+    }
+    fn format(&self) -> Option<String> {
+        Some(self.count.to_string())
+    }
+}
+
+fn execute_query<T: Display>(
+    field: &str,
+    id_range: &RangeInclusive<T>,
+    suffix: &str,
+    index: &Index,
+) -> NumHits {
+    let gen_query_inclusive = |from: &T, to: &T| {
+        format!(
+            "{}:[{} TO {}] {}",
+            field,
+            &from.to_string(),
+            &to.to_string(),
+            suffix
+        )
+    };
+
+    let query = gen_query_inclusive(id_range.start(), id_range.end());
+    execute_query_(&query, index)
+}
+
+fn execute_query_(query: &str, index: &Index) -> NumHits {
+    let query_from_text = |text: &str| {
+        QueryParser::for_index(index, vec![])
+            .parse_query(text)
+            .unwrap()
+    };
+    let query = query_from_text(query);
+    let reader = index.reader().unwrap();
+    let searcher = reader.searcher();
+    let num_hits = searcher
+        .search(&query, &(TopDocs::with_limit(10).order_by_score(), Count))
+        .unwrap()
+        .1;
+    NumHits { count: num_hits }
+}

bitpacker/src/filter_vec/avx2.rs

@@ -19,7 +19,7 @@ fn u32_to_i32(val: u32) -> i32 {
 #[inline]
 unsafe fn u32_to_i32_avx2(vals_u32x8s: DataType) -> DataType {
     const HIGHEST_BIT_MASK: DataType = from_u32x8([HIGHEST_BIT; NUM_LANES]);
-    op_xor(vals_u32x8s, HIGHEST_BIT_MASK)
+    unsafe { op_xor(vals_u32x8s, HIGHEST_BIT_MASK) }
 }
 
 pub fn filter_vec_in_place(range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) {
@@ -66,17 +66,19 @@ unsafe fn filter_vec_avx2_aux(
     ]);
     const SHIFT: __m256i = from_u32x8([NUM_LANES as u32; NUM_LANES]);
     for _ in 0..num_words {
-        let word = load_unaligned(input);
-        let word = u32_to_i32_avx2(word);
-        let keeper_bitset = compute_filter_bitset(word, range_simd.clone());
-        let added_len = keeper_bitset.count_ones();
-        let filtered_doc_ids = compact(ids, keeper_bitset);
-        store_unaligned(output_tail as *mut __m256i, filtered_doc_ids);
-        output_tail = output_tail.offset(added_len as isize);
-        ids = op_add(ids, SHIFT);
-        input = input.offset(1);
+        unsafe {
+            let word = load_unaligned(input);
+            let word = u32_to_i32_avx2(word);
+            let keeper_bitset = compute_filter_bitset(word, range_simd.clone());
+            let added_len = keeper_bitset.count_ones();
+            let filtered_doc_ids = compact(ids, keeper_bitset);
+            store_unaligned(output_tail as *mut __m256i, filtered_doc_ids);
+            output_tail = output_tail.offset(added_len as isize);
+            ids = op_add(ids, SHIFT);
+            input = input.offset(1);
+        }
     }
-    output_tail.offset_from(output) as usize
+    unsafe { output_tail.offset_from(output) as usize }
 }
 
 #[inline]
@@ -92,8 +94,7 @@ unsafe fn compute_filter_bitset(val: __m256i, range: std::ops::RangeInclusive<__
     let too_low = op_greater(*range.start(), val);
     let too_high = op_greater(val, *range.end());
     let inside = op_or(too_low, too_high);
-    255 - std::arch::x86_64::_mm256_movemask_ps(std::mem::transmute::<DataType, __m256>(inside))
-        as u8
+    255 - std::arch::x86_64::_mm256_movemask_ps(_mm256_castsi256_ps(inside)) as u8
 }
 
 union U8x32 {

columnar/src/column_values/u64_based/bitpacked.rs

@@ -41,12 +41,6 @@ fn transform_range_before_linear_transformation(
     if range.is_empty() {
         return None;
     }
-    if stats.min_value > *range.end() {
-        return None;
-    }
-    if stats.max_value < *range.start() {
-        return None;
-    }
     let shifted_range =
         range.start().saturating_sub(stats.min_value)..=range.end().saturating_sub(stats.min_value);
     let start_before_gcd_multiplication: u64 = div_ceil(*shifted_range.start(), stats.gcd);

columnar/src/dynamic_column.rs

@@ -3,7 +3,8 @@ use std::sync::Arc;
 use std::{fmt, io};
 
 use common::file_slice::FileSlice;
-use common::{ByteCount, DateTime, HasLen, OwnedBytes};
+use common::{ByteCount, DateTime, OwnedBytes};
+use serde::{Deserialize, Serialize};
 
 use crate::column::{BytesColumn, Column, StrColumn};
 use crate::column_values::{StrictlyMonotonicFn, monotonic_map_column};
@@ -317,10 +318,89 @@ impl DynamicColumnHandle {
     }
 
     pub fn num_bytes(&self) -> ByteCount {
-        self.file_slice.len().into()
+        self.file_slice.num_bytes()
+    }
+
+    /// Legacy helper returning the column space usage.
+    pub fn column_and_dictionary_num_bytes(&self) -> io::Result<ColumnSpaceUsage> {
+        self.space_usage()
+    }
+
+    /// Return the space usage of the column, optionally broken down by dictionary and column
+    /// values.
+    ///
+    /// For dictionary encoded columns (strings and bytes), this splits the total footprint into
+    /// the dictionary and the remaining column data (including index and values).
+    /// For all other column types, the dictionary size is `None` and the column size
+    /// equals the total bytes.
+    pub fn space_usage(&self) -> io::Result<ColumnSpaceUsage> {
+        let total_num_bytes = self.num_bytes();
+        let dynamic_column = self.open()?;
+        let dictionary_num_bytes = match &dynamic_column {
+            DynamicColumn::Bytes(bytes_column) => bytes_column.dictionary().num_bytes(),
+            DynamicColumn::Str(str_column) => str_column.dictionary().num_bytes(),
+            _ => {
+                return Ok(ColumnSpaceUsage::new(self.num_bytes(), None));
+            }
+        };
+        assert!(dictionary_num_bytes <= total_num_bytes);
+        let column_num_bytes =
+            ByteCount::from(total_num_bytes.get_bytes() - dictionary_num_bytes.get_bytes());
+        Ok(ColumnSpaceUsage::new(
+            column_num_bytes,
+            Some(dictionary_num_bytes),
+        ))
     }
 
     pub fn column_type(&self) -> ColumnType {
         self.column_type
     }
 }
+
+/// Represents space usage of a column.
+///
+/// `column_num_bytes` tracks the column payload (index, values and footer).
+/// For dictionary encoded columns, `dictionary_num_bytes` captures the dictionary footprint.
+/// [`ColumnSpaceUsage::total_num_bytes`] returns the sum of both parts.
+#[derive(Clone, Debug, Serialize, Deserialize)]
+pub struct ColumnSpaceUsage {
+    column_num_bytes: ByteCount,
+    dictionary_num_bytes: Option<ByteCount>,
+}
+
+impl ColumnSpaceUsage {
+    pub(crate) fn new(
+        column_num_bytes: ByteCount,
+        dictionary_num_bytes: Option<ByteCount>,
+    ) -> Self {
+        ColumnSpaceUsage {
+            column_num_bytes,
+            dictionary_num_bytes,
+        }
+    }
+
+    pub fn column_num_bytes(&self) -> ByteCount {
+        self.column_num_bytes
+    }
+
+    pub fn dictionary_num_bytes(&self) -> Option<ByteCount> {
+        self.dictionary_num_bytes
+    }
+
+    pub fn total_num_bytes(&self) -> ByteCount {
+        self.column_num_bytes + self.dictionary_num_bytes.unwrap_or_default()
+    }
+
+    /// Merge two space usage values by summing their components.
+    pub fn merge(&self, other: &ColumnSpaceUsage) -> ColumnSpaceUsage {
+        let dictionary_num_bytes = match (self.dictionary_num_bytes, other.dictionary_num_bytes) {
+            (Some(lhs), Some(rhs)) => Some(lhs + rhs),
+            (Some(val), None) | (None, Some(val)) => Some(val),
+            (None, None) => None,
+        };
+        ColumnSpaceUsage {
+            column_num_bytes: self.column_num_bytes + other.column_num_bytes,
+            dictionary_num_bytes,
+        }
+    }
+}

columnar/src/lib.rs

@@ -48,7 +48,7 @@ pub use columnar::{
 use sstable::VoidSSTable;
 pub use value::{NumericalType, NumericalValue};
 
-pub use self::dynamic_column::{DynamicColumn, DynamicColumnHandle};
+pub use self::dynamic_column::{ColumnSpaceUsage, DynamicColumn, DynamicColumnHandle};
 
 pub type RowId = u32;
 pub type DocId = u32;

examples/geo_json.rs

@@ -1,66 +0,0 @@
-use geo_types::Point;
-use tantivy::collector::TopDocs;
-use tantivy::query::SpatialQuery;
-use tantivy::schema::{Schema, Value, SPATIAL, STORED, TEXT};
-use tantivy::spatial::point::GeoPoint;
-use tantivy::{Index, IndexWriter, TantivyDocument};
-fn main() -> tantivy::Result<()> {
-    let mut schema_builder = Schema::builder();
-    schema_builder.add_json_field("properties", STORED | TEXT);
-    schema_builder.add_spatial_field("geometry", STORED | SPATIAL);
-    let schema = schema_builder.build();
-    let index = Index::create_in_ram(schema.clone());
-    let mut index_writer: IndexWriter = index.writer(50_000_000)?;
-    let doc = TantivyDocument::parse_json(
-        &schema,
-        r#"{
-            "type":"Feature",
-            "geometry":{
-                "type":"Polygon",
-                "coordinates":[[[-99.483911,45.577697],[-99.483869,45.571457],[-99.481739,45.571461],[-99.474881,45.571584],[-99.473167,45.571615],[-99.463394,45.57168],[-99.463391,45.57883],[-99.463368,45.586076],[-99.48177,45.585926],[-99.48384,45.585953],[-99.483885,45.57873],[-99.483911,45.577697]]]
-            },
-            "properties":{
-                "admin_level":"8",
-                "border_type":"city",
-                "boundary":"administrative",
-                "gnis:feature_id":"1267426",
-                "name":"Hosmer",
-                "place":"city",
-                "source":"TIGER/Line® 2008 Place Shapefiles (http://www.census.gov/geo/www/tiger/)",
-                "wikidata":"Q2442118",
-                "wikipedia":"en:Hosmer, South Dakota"
-            }
-        }"#,
-    )?;
-    index_writer.add_document(doc)?;
-    index_writer.commit()?;
-
-    let reader = index.reader()?;
-    let searcher = reader.searcher();
-    let field = schema.get_field("geometry").unwrap();
-    let query = SpatialQuery::new(
-        field,
-        [
-            GeoPoint {
-                lon: -99.49,
-                lat: 45.56,
-            },
-            GeoPoint {
-                lon: -99.45,
-                lat: 45.59,
-            },
-        ],
-        tantivy::query::SpatialQueryType::Intersects,
-    );
-    let hits = searcher.search(&query, &TopDocs::with_limit(10).order_by_score())?;
-    for (_score, doc_address) in &hits {
-        let retrieved_doc: TantivyDocument = searcher.doc(*doc_address)?;
-        if let Some(field_value) = retrieved_doc.get_first(field) {
-            if let Some(geometry_box) = field_value.as_value().into_geometry() {
-                println!("Retrieved geometry: {:?}", geometry_box);
-            }
-        }
-    }
-    assert_eq!(hits.len(), 1);
-    Ok(())
-}

src/collector/sort_key/mod.rs

@@ -1,25 +1,48 @@
 mod order;
+mod sort_by_erased_type;
 mod sort_by_score;
 mod sort_by_static_fast_value;
 mod sort_by_string;
 mod sort_key_computer;
 
 pub use order::*;
+pub use sort_by_erased_type::SortByErasedType;
 pub use sort_by_score::SortBySimilarityScore;
 pub use sort_by_static_fast_value::SortByStaticFastValue;
 pub use sort_by_string::SortByString;
 pub use sort_key_computer::{SegmentSortKeyComputer, SortKeyComputer};
 
 #[cfg(test)]
-mod tests {
+pub(crate) mod tests {
+
+    // By spec, regardless of whether ascending or descending order was requested, in presence of a
+    // tie, we sort by ascending doc id/doc address.
+    pub(crate) fn sort_hits<TSortKey: Ord, D: Ord>(
+        hits: &mut [ComparableDoc<TSortKey, D>],
+        order: Order,
+    ) {
+        if order.is_asc() {
+            hits.sort_by(|l, r| l.sort_key.cmp(&r.sort_key).then(l.doc.cmp(&r.doc)));
+        } else {
+            hits.sort_by(|l, r| {
+                l.sort_key
+                    .cmp(&r.sort_key)
+                    .reverse() // This is descending
+                    .then(l.doc.cmp(&r.doc))
+            });
+        }
+    }
+
     use std::collections::HashMap;
     use std::ops::Range;
 
-    use crate::collector::sort_key::{SortBySimilarityScore, SortByStaticFastValue, SortByString};
+    use crate::collector::sort_key::{
+        SortByErasedType, SortBySimilarityScore, SortByStaticFastValue, SortByString,
+    };
     use crate::collector::{ComparableDoc, DocSetCollector, TopDocs};
     use crate::indexer::NoMergePolicy;
     use crate::query::{AllQuery, QueryParser};
-    use crate::schema::{Schema, FAST, TEXT};
+    use crate::schema::{OwnedValue, Schema, FAST, TEXT};
     use crate::{DocAddress, Document, Index, Order, Score, Searcher};
 
     fn make_index() -> crate::Result<Index> {
@@ -294,11 +317,9 @@ mod tests {
                 (SortBySimilarityScore, score_order),
                 (SortByString::for_field("city"), city_order),
             ));
-            Ok(searcher
-                .search(&AllQuery, &top_collector)?
-                .into_iter()
-                .map(|(f, doc)| (f, ids[&doc]))
-                .collect())
+            let results: Vec<((Score, Option<String>), DocAddress)> =
+                searcher.search(&AllQuery, &top_collector)?;
+            Ok(results.into_iter().map(|(f, doc)| (f, ids[&doc])).collect())
         }
 
         assert_eq!(
@@ -323,6 +344,51 @@ mod tests {
         Ok(())
     }
 
+    #[test]
+    fn test_order_by_score_then_owned_value() -> crate::Result<()> {
+        let index = make_index()?;
+
+        type SortKey = (Score, OwnedValue);
+
+        fn query(
+            index: &Index,
+            score_order: Order,
+            city_order: Order,
+        ) -> crate::Result<Vec<(SortKey, u64)>> {
+            let searcher = index.reader()?.searcher();
+            let ids = id_mapping(&searcher);
+
+            let top_collector = TopDocs::with_limit(4).order_by::<(Score, OwnedValue)>((
+                (SortBySimilarityScore, score_order),
+                (SortByErasedType::for_field("city"), city_order),
+            ));
+            let results: Vec<((Score, OwnedValue), DocAddress)> =
+                searcher.search(&AllQuery, &top_collector)?;
+            Ok(results.into_iter().map(|(f, doc)| (f, ids[&doc])).collect())
+        }
+
+        assert_eq!(
+            &query(&index, Order::Asc, Order::Asc)?,
+            &[
+                ((1.0, OwnedValue::Str("austin".to_owned())), 0),
+                ((1.0, OwnedValue::Str("greenville".to_owned())), 1),
+                ((1.0, OwnedValue::Str("tokyo".to_owned())), 2),
+                ((1.0, OwnedValue::Null), 3),
+            ]
+        );
+
+        assert_eq!(
+            &query(&index, Order::Asc, Order::Desc)?,
+            &[
+                ((1.0, OwnedValue::Str("tokyo".to_owned())), 2),
+                ((1.0, OwnedValue::Str("greenville".to_owned())), 1),
+                ((1.0, OwnedValue::Str("austin".to_owned())), 0),
+                ((1.0, OwnedValue::Null), 3),
+            ]
+        );
+        Ok(())
+    }
+
     use proptest::prelude::*;
 
     proptest! {
@@ -372,15 +438,10 @@ mod tests {
 
             // Using the TopDocs collector should always be equivalent to sorting, skipping the
             // offset, and then taking the limit.
-            let sorted_docs: Vec<_> = if order.is_desc() {
-                let mut comparable_docs: Vec<ComparableDoc<_, _, true>> =
+            let sorted_docs: Vec<_> = {
+                let mut comparable_docs: Vec<ComparableDoc<_, _>> =
                     all_results.into_iter().map(|(sort_key, doc)| ComparableDoc { sort_key, doc}).collect();
-                comparable_docs.sort();
-                comparable_docs.into_iter().map(|cd| (cd.sort_key, cd.doc)).collect()
-            } else {
-                let mut comparable_docs: Vec<ComparableDoc<_, _, false>> =
-                    all_results.into_iter().map(|(sort_key, doc)| ComparableDoc { sort_key, doc}).collect();
-                comparable_docs.sort();
+                sort_hits(&mut comparable_docs, order);
                 comparable_docs.into_iter().map(|cd| (cd.sort_key, cd.doc)).collect()
             };
             let expected_docs = sorted_docs.into_iter().skip(offset).take(limit).collect::<Vec<_>>();

src/collector/sort_key/order.rs

@@ -1,36 +1,116 @@
 use std::cmp::Ordering;
 
+use columnar::MonotonicallyMappableToU64;
 use serde::{Deserialize, Serialize};
 
 use crate::collector::{SegmentSortKeyComputer, SortKeyComputer};
-use crate::schema::Schema;
+use crate::schema::{OwnedValue, Schema};
 use crate::{DocId, Order, Score};
 
+fn compare_owned_value<const NULLS_FIRST: bool>(lhs: &OwnedValue, rhs: &OwnedValue) -> Ordering {
+    match (lhs, rhs) {
+        (OwnedValue::Null, OwnedValue::Null) => Ordering::Equal,
+        (OwnedValue::Null, _) => {
+            if NULLS_FIRST {
+                Ordering::Less
+            } else {
+                Ordering::Greater
+            }
+        }
+        (_, OwnedValue::Null) => {
+            if NULLS_FIRST {
+                Ordering::Greater
+            } else {
+                Ordering::Less
+            }
+        }
+        (OwnedValue::Str(a), OwnedValue::Str(b)) => a.cmp(b),
+        (OwnedValue::PreTokStr(a), OwnedValue::PreTokStr(b)) => a.cmp(b),
+        (OwnedValue::U64(a), OwnedValue::U64(b)) => a.cmp(b),
+        (OwnedValue::I64(a), OwnedValue::I64(b)) => a.cmp(b),
+        (OwnedValue::F64(a), OwnedValue::F64(b)) => a.to_u64().cmp(&b.to_u64()),
+        (OwnedValue::Bool(a), OwnedValue::Bool(b)) => a.cmp(b),
+        (OwnedValue::Date(a), OwnedValue::Date(b)) => a.cmp(b),
+        (OwnedValue::Facet(a), OwnedValue::Facet(b)) => a.cmp(b),
+        (OwnedValue::Bytes(a), OwnedValue::Bytes(b)) => a.cmp(b),
+        (OwnedValue::IpAddr(a), OwnedValue::IpAddr(b)) => a.cmp(b),
+        (OwnedValue::U64(a), OwnedValue::I64(b)) => {
+            if *b < 0 {
+                Ordering::Greater
+            } else {
+                a.cmp(&(*b as u64))
+            }
+        }
+        (OwnedValue::I64(a), OwnedValue::U64(b)) => {
+            if *a < 0 {
+                Ordering::Less
+            } else {
+                (*a as u64).cmp(b)
+            }
+        }
+        (OwnedValue::U64(a), OwnedValue::F64(b)) => (*a as f64).to_u64().cmp(&b.to_u64()),
+        (OwnedValue::F64(a), OwnedValue::U64(b)) => a.to_u64().cmp(&(*b as f64).to_u64()),
+        (OwnedValue::I64(a), OwnedValue::F64(b)) => (*a as f64).to_u64().cmp(&b.to_u64()),
+        (OwnedValue::F64(a), OwnedValue::I64(b)) => a.to_u64().cmp(&(*b as f64).to_u64()),
+        (a, b) => {
+            let ord = a.discriminant_value().cmp(&b.discriminant_value());
+            // If the discriminant is equal, it's because a new type was added, but hasn't been
+            // included in this `match` statement.
+            assert!(
+                ord != Ordering::Equal,
+                "Unimplemented comparison for type of {a:?}, {b:?}"
+            );
+            ord
+        }
+    }
+}
+
 /// Comparator trait defining the order in which documents should be ordered.
 pub trait Comparator<T>: Send + Sync + std::fmt::Debug + Default {
     /// Return the order between two values.
     fn compare(&self, lhs: &T, rhs: &T) -> Ordering;
 }
 
-/// With the natural comparator, the top k collector will return
-/// the top documents in decreasing order.
+/// Compare values naturally (e.g. 1 < 2).
+///
+/// When used with `TopDocs`, which reverses the order, this results in a
+/// "Descending" sort (Greatest values first).
+///
+/// `None` (or Null for `OwnedValue`) values are considered to be smaller than any other value,
+/// and will therefore appear last in a descending sort (e.g. `[Some(20), Some(10), None]`).
 #[derive(Debug, Copy, Clone, Default, Serialize, Deserialize)]
 pub struct NaturalComparator;
 
 impl<T: PartialOrd> Comparator<T> for NaturalComparator {
     #[inline(always)]
     fn compare(&self, lhs: &T, rhs: &T) -> Ordering {
-        lhs.partial_cmp(rhs).unwrap()
+        lhs.partial_cmp(rhs).unwrap_or(Ordering::Equal)
     }
 }
 
-/// Sorts document in reverse order.
+/// A (partial) implementation of comparison for OwnedValue.
 ///
-/// If the sort key is None, it will considered as the lowest value, and will therefore appear
-/// first.
+/// Intended for use within columns of homogenous types, and so will panic for OwnedValues with
+/// mismatched types. The one exception is Null, for which we do define all comparisons.
+impl Comparator<OwnedValue> for NaturalComparator {
+    #[inline(always)]
+    fn compare(&self, lhs: &OwnedValue, rhs: &OwnedValue) -> Ordering {
+        compare_owned_value::</* NULLS_FIRST= */ true>(lhs, rhs)
+    }
+}
+
+/// Compare values in reverse (e.g. 2 < 1).
+///
+/// When used with `TopDocs`, which reverses the order, this results in an
+/// "Ascending" sort (Smallest values first).
+///
+/// `None` is considered smaller than `Some` in the underlying comparator, but because the
+/// comparison is reversed, `None` is effectively treated as the lowest value in the resulting
+/// Ascending sort (e.g. `[None, Some(10), Some(20)]`).
 ///
 /// The ReverseComparator does not necessarily imply that the sort order is reversed compared
-/// to the NaturalComparator. In presence of a tie, both version will retain the higher doc ids.
+/// to the NaturalComparator. In presence of a tie on the sort key, documents will always be
+/// sorted by ascending `DocId`/`DocAddress` in TopN results, regardless of the sort key's order.
 #[derive(Debug, Copy, Clone, Default, Serialize, Deserialize)]
 pub struct ReverseComparator;
 
@@ -43,11 +123,15 @@ where NaturalComparator: Comparator<T>
     }
 }
 
-/// Sorts document in reverse order, but considers None as having the lowest value.
+/// Compare values in reverse, but treating `None` as lower than `Some`.
+///
+/// When used with `TopDocs`, which reverses the order, this results in an
+/// "Ascending" sort (Smallest values first), but with `None` values appearing last
+/// (e.g. `[Some(10), Some(20), None]`).
 ///
 /// This is usually what is wanted when sorting by a field in an ascending order.
-/// For instance, in a e-commerce website, if I sort by price ascending, I most likely want the
-/// cheapest items first, and the items without a price at last.
+/// For instance, in an e-commerce website, if sorting by price ascending,
+/// the cheapest items would appear first, and items without a price would appear last.
 #[derive(Debug, Copy, Clone, Default)]
 pub struct ReverseNoneIsLowerComparator;
 
@@ -107,6 +191,84 @@ impl Comparator<String> for ReverseNoneIsLowerComparator {
     }
 }
 
+impl Comparator<OwnedValue> for ReverseNoneIsLowerComparator {
+    #[inline(always)]
+    fn compare(&self, lhs: &OwnedValue, rhs: &OwnedValue) -> Ordering {
+        compare_owned_value::</* NULLS_FIRST= */ false>(rhs, lhs)
+    }
+}
+
+/// Compare values naturally, but treating `None` as higher than `Some`.
+///
+/// When used with `TopDocs`, which reverses the order, this results in a
+/// "Descending" sort (Greatest values first), but with `None` values appearing first
+/// (e.g. `[None, Some(20), Some(10)]`).
+#[derive(Debug, Copy, Clone, Default, Serialize, Deserialize)]
+pub struct NaturalNoneIsHigherComparator;
+
+impl<T> Comparator<Option<T>> for NaturalNoneIsHigherComparator
+where NaturalComparator: Comparator<T>
+{
+    #[inline(always)]
+    fn compare(&self, lhs_opt: &Option<T>, rhs_opt: &Option<T>) -> Ordering {
+        match (lhs_opt, rhs_opt) {
+            (None, None) => Ordering::Equal,
+            (None, Some(_)) => Ordering::Greater,
+            (Some(_), None) => Ordering::Less,
+            (Some(lhs), Some(rhs)) => NaturalComparator.compare(lhs, rhs),
+        }
+    }
+}
+
+impl Comparator<u32> for NaturalNoneIsHigherComparator {
+    #[inline(always)]
+    fn compare(&self, lhs: &u32, rhs: &u32) -> Ordering {
+        NaturalComparator.compare(lhs, rhs)
+    }
+}
+
+impl Comparator<u64> for NaturalNoneIsHigherComparator {
+    #[inline(always)]
+    fn compare(&self, lhs: &u64, rhs: &u64) -> Ordering {
+        NaturalComparator.compare(lhs, rhs)
+    }
+}
+
+impl Comparator<f64> for NaturalNoneIsHigherComparator {
+    #[inline(always)]
+    fn compare(&self, lhs: &f64, rhs: &f64) -> Ordering {
+        NaturalComparator.compare(lhs, rhs)
+    }
+}
+
+impl Comparator<f32> for NaturalNoneIsHigherComparator {
+    #[inline(always)]
+    fn compare(&self, lhs: &f32, rhs: &f32) -> Ordering {
+        NaturalComparator.compare(lhs, rhs)
+    }
+}
+
+impl Comparator<i64> for NaturalNoneIsHigherComparator {
+    #[inline(always)]
+    fn compare(&self, lhs: &i64, rhs: &i64) -> Ordering {
+        NaturalComparator.compare(lhs, rhs)
+    }
+}
+
+impl Comparator<String> for NaturalNoneIsHigherComparator {
+    #[inline(always)]
+    fn compare(&self, lhs: &String, rhs: &String) -> Ordering {
+        NaturalComparator.compare(lhs, rhs)
+    }
+}
+
+impl Comparator<OwnedValue> for NaturalNoneIsHigherComparator {
+    #[inline(always)]
+    fn compare(&self, lhs: &OwnedValue, rhs: &OwnedValue) -> Ordering {
+        compare_owned_value::</* NULLS_FIRST= */ false>(lhs, rhs)
+    }
+}
+
 /// An enum representing the different sort orders.
 #[derive(Debug, Clone, Copy, Eq, PartialEq, Default)]
 pub enum ComparatorEnum {
@@ -115,8 +277,10 @@ pub enum ComparatorEnum {
     Natural,
     /// Reverse order (See [ReverseComparator])
     Reverse,
-    /// Reverse order by treating None as the lowest value.(See [ReverseNoneLowerComparator])
+    /// Reverse order by treating None as the lowest value. (See [ReverseNoneLowerComparator])
     ReverseNoneLower,
+    /// Natural order but treating None as the highest value. (See [NaturalNoneIsHigherComparator])
+    NaturalNoneHigher,
 }
 
 impl From<Order> for ComparatorEnum {
@@ -133,6 +297,7 @@ where
     ReverseNoneIsLowerComparator: Comparator<T>,
     NaturalComparator: Comparator<T>,
     ReverseComparator: Comparator<T>,
+    NaturalNoneIsHigherComparator: Comparator<T>,
 {
     #[inline(always)]
     fn compare(&self, lhs: &T, rhs: &T) -> Ordering {
@@ -140,6 +305,7 @@ where
             ComparatorEnum::Natural => NaturalComparator.compare(lhs, rhs),
             ComparatorEnum::Reverse => ReverseComparator.compare(lhs, rhs),
             ComparatorEnum::ReverseNoneLower => ReverseNoneIsLowerComparator.compare(lhs, rhs),
+            ComparatorEnum::NaturalNoneHigher => NaturalNoneIsHigherComparator.compare(lhs, rhs),
         }
     }
 }
@@ -322,11 +488,12 @@ impl<TSegmentSortKeyComputer, TSegmentSortKey, TComparator> SegmentSortKeyComput
     for SegmentSortKeyComputerWithComparator<TSegmentSortKeyComputer, TComparator>
 where
     TSegmentSortKeyComputer: SegmentSortKeyComputer<SegmentSortKey = TSegmentSortKey>,
-    TSegmentSortKey: PartialOrd + Clone + 'static + Sync + Send,
+    TSegmentSortKey: Clone + 'static + Sync + Send,
     TComparator: Comparator<TSegmentSortKey> + 'static + Sync + Send,
 {
     type SortKey = TSegmentSortKeyComputer::SortKey;
     type SegmentSortKey = TSegmentSortKey;
+    type SegmentComparator = TComparator;
 
     fn segment_sort_key(&mut self, doc: DocId, score: Score) -> Self::SegmentSortKey {
         self.segment_sort_key_computer.segment_sort_key(doc, score)
@@ -346,3 +513,55 @@ where
             .convert_segment_sort_key(sort_key)
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::schema::OwnedValue;
+
+    #[test]
+    fn test_natural_none_is_higher() {
+        let comp = NaturalNoneIsHigherComparator;
+        let null = None;
+        let v1 = Some(1_u64);
+        let v2 = Some(2_u64);
+
+        // NaturalNoneIsGreaterComparator logic:
+        // 1. Delegates to NaturalComparator for non-nulls.
+        // NaturalComparator compare(2, 1) -> 2.cmp(1) -> Greater.
+        assert_eq!(comp.compare(&v2, &v1), Ordering::Greater);
+
+        // 2. Treats None (Null) as Greater than any value.
+        // compare(None, Some(2)) should be Greater.
+        assert_eq!(comp.compare(&null, &v2), Ordering::Greater);
+
+        // compare(Some(1), None) should be Less.
+        assert_eq!(comp.compare(&v1, &null), Ordering::Less);
+
+        // compare(None, None) should be Equal.
+        assert_eq!(comp.compare(&null, &null), Ordering::Equal);
+    }
+
+    #[test]
+    fn test_mixed_ownedvalue_compare() {
+        let u = OwnedValue::U64(10);
+        let i = OwnedValue::I64(10);
+        let f = OwnedValue::F64(10.0);
+
+        let nc = NaturalComparator;
+        assert_eq!(nc.compare(&u, &i), Ordering::Equal);
+        assert_eq!(nc.compare(&u, &f), Ordering::Equal);
+        assert_eq!(nc.compare(&i, &f), Ordering::Equal);
+
+        let u2 = OwnedValue::U64(11);
+        assert_eq!(nc.compare(&u2, &f), Ordering::Greater);
+
+        let s = OwnedValue::Str("a".to_string());
+        // Str < U64
+        assert_eq!(nc.compare(&s, &u), Ordering::Less);
+        // Str < I64
+        assert_eq!(nc.compare(&s, &i), Ordering::Less);
+        // Str < F64
+        assert_eq!(nc.compare(&s, &f), Ordering::Less);
+    }
+}

src/collector/sort_key/sort_by_erased_type.rs

@@ -0,0 +1,361 @@
+use columnar::{ColumnType, MonotonicallyMappableToU64};
+
+use crate::collector::sort_key::{
+    NaturalComparator, SortBySimilarityScore, SortByStaticFastValue, SortByString,
+};
+use crate::collector::{SegmentSortKeyComputer, SortKeyComputer};
+use crate::fastfield::FastFieldNotAvailableError;
+use crate::schema::OwnedValue;
+use crate::{DateTime, DocId, Score};
+
+/// Sort by the boxed / OwnedValue representation of either a fast field, or of the score.
+///
+/// Using the OwnedValue representation allows for type erasure, and can be useful when sort orders
+/// are not known until runtime. But it comes with a performance cost: wherever possible, prefer to
+/// use a SortKeyComputer implementation with a known-type at compile time.
+#[derive(Debug, Clone)]
+pub enum SortByErasedType {
+    /// Sort by a fast field
+    Field(String),
+    /// Sort by score
+    Score,
+}
+
+impl SortByErasedType {
+    /// Creates a new sort key computer which will sort by the given fast field column, with type
+    /// erasure.
+    pub fn for_field(column_name: impl ToString) -> Self {
+        Self::Field(column_name.to_string())
+    }
+
+    /// Creates a new sort key computer which will sort by score, with type erasure.
+    pub fn for_score() -> Self {
+        Self::Score
+    }
+}
+
+trait ErasedSegmentSortKeyComputer: Send + Sync {
+    fn segment_sort_key(&mut self, doc: DocId, score: Score) -> Option<u64>;
+    fn convert_segment_sort_key(&self, sort_key: Option<u64>) -> OwnedValue;
+}
+
+struct ErasedSegmentSortKeyComputerWrapper<C, F> {
+    inner: C,
+    converter: F,
+}
+
+impl<C, F> ErasedSegmentSortKeyComputer for ErasedSegmentSortKeyComputerWrapper<C, F>
+where
+    C: SegmentSortKeyComputer<SegmentSortKey = Option<u64>> + Send + Sync,
+    F: Fn(C::SortKey) -> OwnedValue + Send + Sync + 'static,
+{
+    fn segment_sort_key(&mut self, doc: DocId, score: Score) -> Option<u64> {
+        self.inner.segment_sort_key(doc, score)
+    }
+
+    fn convert_segment_sort_key(&self, sort_key: Option<u64>) -> OwnedValue {
+        let val = self.inner.convert_segment_sort_key(sort_key);
+        (self.converter)(val)
+    }
+}
+
+struct ScoreSegmentSortKeyComputer {
+    segment_computer: SortBySimilarityScore,
+}
+
+impl ErasedSegmentSortKeyComputer for ScoreSegmentSortKeyComputer {
+    fn segment_sort_key(&mut self, doc: DocId, score: Score) -> Option<u64> {
+        let score_value: f64 = self.segment_computer.segment_sort_key(doc, score).into();
+        Some(score_value.to_u64())
+    }
+
+    fn convert_segment_sort_key(&self, sort_key: Option<u64>) -> OwnedValue {
+        let score_value: u64 = sort_key.expect("This implementation always produces a score.");
+        OwnedValue::F64(f64::from_u64(score_value))
+    }
+}
+
+impl SortKeyComputer for SortByErasedType {
+    type SortKey = OwnedValue;
+    type Child = ErasedColumnSegmentSortKeyComputer;
+    type Comparator = NaturalComparator;
+
+    fn requires_scoring(&self) -> bool {
+        matches!(self, Self::Score)
+    }
+
+    fn segment_sort_key_computer(
+        &self,
+        segment_reader: &crate::SegmentReader,
+    ) -> crate::Result<Self::Child> {
+        let inner: Box<dyn ErasedSegmentSortKeyComputer> = match self {
+            Self::Field(column_name) => {
+                let fast_fields = segment_reader.fast_fields();
+                // TODO: We currently double-open the column to avoid relying on the implementation
+                // details of `SortByString` or `SortByStaticFastValue`. Once
+                // https://github.com/quickwit-oss/tantivy/issues/2776 is resolved, we should
+                // consider directly constructing the appropriate `SegmentSortKeyComputer` type for
+                // the column that we open here.
+                let (_column, column_type) =
+                    fast_fields.u64_lenient(column_name)?.ok_or_else(|| {
+                        FastFieldNotAvailableError {
+                            field_name: column_name.to_owned(),
+                        }
+                    })?;
+
+                match column_type {
+                    ColumnType::Str => {
+                        let computer = SortByString::for_field(column_name);
+                        let inner = computer.segment_sort_key_computer(segment_reader)?;
+                        Box::new(ErasedSegmentSortKeyComputerWrapper {
+                            inner,
+                            converter: |val: Option<String>| {
+                                val.map(OwnedValue::Str).unwrap_or(OwnedValue::Null)
+                            },
+                        })
+                    }
+                    ColumnType::U64 => {
+                        let computer = SortByStaticFastValue::<u64>::for_field(column_name);
+                        let inner = computer.segment_sort_key_computer(segment_reader)?;
+                        Box::new(ErasedSegmentSortKeyComputerWrapper {
+                            inner,
+                            converter: |val: Option<u64>| {
+                                val.map(OwnedValue::U64).unwrap_or(OwnedValue::Null)
+                            },
+                        })
+                    }
+                    ColumnType::I64 => {
+                        let computer = SortByStaticFastValue::<i64>::for_field(column_name);
+                        let inner = computer.segment_sort_key_computer(segment_reader)?;
+                        Box::new(ErasedSegmentSortKeyComputerWrapper {
+                            inner,
+                            converter: |val: Option<i64>| {
+                                val.map(OwnedValue::I64).unwrap_or(OwnedValue::Null)
+                            },
+                        })
+                    }
+                    ColumnType::F64 => {
+                        let computer = SortByStaticFastValue::<f64>::for_field(column_name);
+                        let inner = computer.segment_sort_key_computer(segment_reader)?;
+                        Box::new(ErasedSegmentSortKeyComputerWrapper {
+                            inner,
+                            converter: |val: Option<f64>| {
+                                val.map(OwnedValue::F64).unwrap_or(OwnedValue::Null)
+                            },
+                        })
+                    }
+                    ColumnType::Bool => {
+                        let computer = SortByStaticFastValue::<bool>::for_field(column_name);
+                        let inner = computer.segment_sort_key_computer(segment_reader)?;
+                        Box::new(ErasedSegmentSortKeyComputerWrapper {
+                            inner,
+                            converter: |val: Option<bool>| {
+                                val.map(OwnedValue::Bool).unwrap_or(OwnedValue::Null)
+                            },
+                        })
+                    }
+                    ColumnType::DateTime => {
+                        let computer = SortByStaticFastValue::<DateTime>::for_field(column_name);
+                        let inner = computer.segment_sort_key_computer(segment_reader)?;
+                        Box::new(ErasedSegmentSortKeyComputerWrapper {
+                            inner,
+                            converter: |val: Option<DateTime>| {
+                                val.map(OwnedValue::Date).unwrap_or(OwnedValue::Null)
+                            },
+                        })
+                    }
+                    column_type => {
+                        return Err(crate::TantivyError::SchemaError(format!(
+                            "Field `{}` is of type {column_type:?}, which is not supported for \
+                             sorting by owned value yet.",
+                            column_name
+                        )))
+                    }
+                }
+            }
+            Self::Score => Box::new(ScoreSegmentSortKeyComputer {
+                segment_computer: SortBySimilarityScore,
+            }),
+        };
+        Ok(ErasedColumnSegmentSortKeyComputer { inner })
+    }
+}
+
+pub struct ErasedColumnSegmentSortKeyComputer {
+    inner: Box<dyn ErasedSegmentSortKeyComputer>,
+}
+
+impl SegmentSortKeyComputer for ErasedColumnSegmentSortKeyComputer {
+    type SortKey = OwnedValue;
+    type SegmentSortKey = Option<u64>;
+    type SegmentComparator = NaturalComparator;
+
+    #[inline(always)]
+    fn segment_sort_key(&mut self, doc: DocId, score: Score) -> Option<u64> {
+        self.inner.segment_sort_key(doc, score)
+    }
+
+    fn convert_segment_sort_key(&self, segment_sort_key: Self::SegmentSortKey) -> OwnedValue {
+        self.inner.convert_segment_sort_key(segment_sort_key)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::collector::sort_key::{ComparatorEnum, SortByErasedType};
+    use crate::collector::TopDocs;
+    use crate::query::AllQuery;
+    use crate::schema::{OwnedValue, Schema, FAST, TEXT};
+    use crate::Index;
+
+    #[test]
+    fn test_sort_by_owned_u64() {
+        let mut schema_builder = Schema::builder();
+        let id_field = schema_builder.add_u64_field("id", FAST);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut writer = index.writer_for_tests().unwrap();
+        writer.add_document(doc!(id_field => 10u64)).unwrap();
+        writer.add_document(doc!(id_field => 2u64)).unwrap();
+        writer.add_document(doc!()).unwrap();
+        writer.commit().unwrap();
+
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+
+        let collector = TopDocs::with_limit(10)
+            .order_by((SortByErasedType::for_field("id"), ComparatorEnum::Natural));
+        let top_docs = searcher.search(&AllQuery, &collector).unwrap();
+
+        let values: Vec<OwnedValue> = top_docs.into_iter().map(|(key, _)| key).collect();
+
+        assert_eq!(
+            values,
+            vec![OwnedValue::U64(10), OwnedValue::U64(2), OwnedValue::Null]
+        );
+
+        let collector = TopDocs::with_limit(10).order_by((
+            SortByErasedType::for_field("id"),
+            ComparatorEnum::ReverseNoneLower,
+        ));
+        let top_docs = searcher.search(&AllQuery, &collector).unwrap();
+
+        let values: Vec<OwnedValue> = top_docs.into_iter().map(|(key, _)| key).collect();
+
+        assert_eq!(
+            values,
+            vec![OwnedValue::U64(2), OwnedValue::U64(10), OwnedValue::Null]
+        );
+    }
+
+    #[test]
+    fn test_sort_by_owned_string() {
+        let mut schema_builder = Schema::builder();
+        let city_field = schema_builder.add_text_field("city", FAST | TEXT);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut writer = index.writer_for_tests().unwrap();
+        writer.add_document(doc!(city_field => "tokyo")).unwrap();
+        writer.add_document(doc!(city_field => "austin")).unwrap();
+        writer.add_document(doc!()).unwrap();
+        writer.commit().unwrap();
+
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+
+        let collector = TopDocs::with_limit(10).order_by((
+            SortByErasedType::for_field("city"),
+            ComparatorEnum::ReverseNoneLower,
+        ));
+        let top_docs = searcher.search(&AllQuery, &collector).unwrap();
+
+        let values: Vec<OwnedValue> = top_docs.into_iter().map(|(key, _)| key).collect();
+
+        assert_eq!(
+            values,
+            vec![
+                OwnedValue::Str("austin".to_string()),
+                OwnedValue::Str("tokyo".to_string()),
+                OwnedValue::Null
+            ]
+        );
+    }
+
+    #[test]
+    fn test_sort_by_owned_reverse() {
+        let mut schema_builder = Schema::builder();
+        let id_field = schema_builder.add_u64_field("id", FAST);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut writer = index.writer_for_tests().unwrap();
+        writer.add_document(doc!(id_field => 10u64)).unwrap();
+        writer.add_document(doc!(id_field => 2u64)).unwrap();
+        writer.add_document(doc!()).unwrap();
+        writer.commit().unwrap();
+
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+
+        let collector = TopDocs::with_limit(10)
+            .order_by((SortByErasedType::for_field("id"), ComparatorEnum::Reverse));
+        let top_docs = searcher.search(&AllQuery, &collector).unwrap();
+
+        let values: Vec<OwnedValue> = top_docs.into_iter().map(|(key, _)| key).collect();
+
+        assert_eq!(
+            values,
+            vec![OwnedValue::Null, OwnedValue::U64(2), OwnedValue::U64(10)]
+        );
+    }
+
+    #[test]
+    fn test_sort_by_owned_score() {
+        let mut schema_builder = Schema::builder();
+        let body_field = schema_builder.add_text_field("body", TEXT);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut writer = index.writer_for_tests().unwrap();
+        writer.add_document(doc!(body_field => "a a")).unwrap();
+        writer.add_document(doc!(body_field => "a")).unwrap();
+        writer.commit().unwrap();
+
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+        let query_parser = crate::query::QueryParser::for_index(&index, vec![body_field]);
+        let query = query_parser.parse_query("a").unwrap();
+
+        // Sort by score descending (Natural)
+        let collector = TopDocs::with_limit(10)
+            .order_by((SortByErasedType::for_score(), ComparatorEnum::Natural));
+        let top_docs = searcher.search(&query, &collector).unwrap();
+
+        let values: Vec<f64> = top_docs
+            .into_iter()
+            .map(|(key, _)| match key {
+                OwnedValue::F64(val) => val,
+                _ => panic!("Wrong type {key:?}"),
+            })
+            .collect();
+
+        assert_eq!(values.len(), 2);
+        assert!(values[0] > values[1]);
+
+        // Sort by score ascending (ReverseNoneLower)
+        let collector = TopDocs::with_limit(10).order_by((
+            SortByErasedType::for_score(),
+            ComparatorEnum::ReverseNoneLower,
+        ));
+        let top_docs = searcher.search(&query, &collector).unwrap();
+
+        let values: Vec<f64> = top_docs
+            .into_iter()
+            .map(|(key, _)| match key {
+                OwnedValue::F64(val) => val,
+                _ => panic!("Wrong type {key:?}"),
+            })
+            .collect();
+
+        assert_eq!(values.len(), 2);
+        assert!(values[0] < values[1]);
+    }
+}

src/collector/sort_key/sort_by_score.rs

@@ -63,8 +63,8 @@ impl SortKeyComputer for SortBySimilarityScore {
 
 impl SegmentSortKeyComputer for SortBySimilarityScore {
     type SortKey = Score;
-
     type SegmentSortKey = Score;
+    type SegmentComparator = NaturalComparator;
 
     #[inline(always)]
     fn segment_sort_key(&mut self, _doc: DocId, score: Score) -> Score {

src/collector/sort_key/sort_by_static_fast_value.rs

@@ -34,9 +34,7 @@ impl<T: FastValue> SortByStaticFastValue<T> {
 
 impl<T: FastValue> SortKeyComputer for SortByStaticFastValue<T> {
     type Child = SortByFastValueSegmentSortKeyComputer<T>;
-
     type SortKey = Option<T>;
-
     type Comparator = NaturalComparator;
 
     fn check_schema(&self, schema: &crate::schema::Schema) -> crate::Result<()> {
@@ -84,8 +82,8 @@ pub struct SortByFastValueSegmentSortKeyComputer<T> {
 
 impl<T: FastValue> SegmentSortKeyComputer for SortByFastValueSegmentSortKeyComputer<T> {
     type SortKey = Option<T>;
-
     type SegmentSortKey = Option<u64>;
+    type SegmentComparator = NaturalComparator;
 
     #[inline(always)]
     fn segment_sort_key(&mut self, doc: DocId, _score: Score) -> Self::SegmentSortKey {

src/collector/sort_key/sort_by_string.rs

@@ -30,9 +30,7 @@ impl SortByString {
 
 impl SortKeyComputer for SortByString {
     type SortKey = Option<String>;
-
     type Child = ByStringColumnSegmentSortKeyComputer;
-
     type Comparator = NaturalComparator;
 
     fn segment_sort_key_computer(
@@ -50,8 +48,8 @@ pub struct ByStringColumnSegmentSortKeyComputer {
 
 impl SegmentSortKeyComputer for ByStringColumnSegmentSortKeyComputer {
     type SortKey = Option<String>;
-
     type SegmentSortKey = Option<TermOrdinal>;
+    type SegmentComparator = NaturalComparator;
 
     #[inline(always)]
     fn segment_sort_key(&mut self, doc: DocId, _score: Score) -> Option<TermOrdinal> {
@@ -60,6 +58,8 @@ impl SegmentSortKeyComputer for ByStringColumnSegmentSortKeyComputer {
     }
 
     fn convert_segment_sort_key(&self, term_ord_opt: Option<TermOrdinal>) -> Option<String> {
+        // TODO: Individual lookups to the dictionary like this are very likely to repeatedly
+        // decompress the same blocks. See https://github.com/quickwit-oss/tantivy/issues/2776
         let term_ord = term_ord_opt?;
         let str_column = self.str_column_opt.as_ref()?;
         let mut bytes = Vec::new();

src/collector/sort_key/sort_key_computer.rs

@@ -12,13 +12,21 @@ use crate::{DocAddress, DocId, Result, Score, SegmentReader};
 /// It is the segment local version of the [`SortKeyComputer`].
 pub trait SegmentSortKeyComputer: 'static {
     /// The final score being emitted.
-    type SortKey: 'static + PartialOrd + Send + Sync + Clone;
+    type SortKey: 'static + Send + Sync + Clone;
 
     /// Sort key used by at the segment level by the `SegmentSortKeyComputer`.
     ///
     /// It is typically small like a `u64`, and is meant to be converted
     /// to the final score at the end of the collection of the segment.
-    type SegmentSortKey: 'static + PartialOrd + Clone + Send + Sync + Clone;
+    type SegmentSortKey: 'static + Clone + Send + Sync + Clone;
+
+    /// Comparator type.
+    type SegmentComparator: Comparator<Self::SegmentSortKey> + 'static;
+
+    /// Returns the segment sort key comparator.
+    fn segment_comparator(&self) -> Self::SegmentComparator {
+        Self::SegmentComparator::default()
+    }
 
     /// Computes the sort key for the given document and score.
     fn segment_sort_key(&mut self, doc: DocId, score: Score) -> Self::SegmentSortKey;
@@ -47,7 +55,7 @@ pub trait SegmentSortKeyComputer: 'static {
         left: &Self::SegmentSortKey,
         right: &Self::SegmentSortKey,
     ) -> Ordering {
-        NaturalComparator.compare(left, right)
+        self.segment_comparator().compare(left, right)
     }
 
     /// Implementing this method makes it possible to avoid computing
@@ -81,7 +89,7 @@ pub trait SegmentSortKeyComputer: 'static {
 /// the sort key at a segment scale.
 pub trait SortKeyComputer: Sync {
     /// The sort key type.
-    type SortKey: 'static + Send + Sync + PartialOrd + Clone + std::fmt::Debug;
+    type SortKey: 'static + Send + Sync + Clone + std::fmt::Debug;
     /// Type of the associated [`SegmentSortKeyComputer`].
     type Child: SegmentSortKeyComputer<SortKey = Self::SortKey>;
     /// Comparator type.
@@ -136,10 +144,7 @@ where
     HeadSortKeyComputer: SortKeyComputer,
     TailSortKeyComputer: SortKeyComputer,
 {
-    type SortKey = (
-        <HeadSortKeyComputer::Child as SegmentSortKeyComputer>::SortKey,
-        <TailSortKeyComputer::Child as SegmentSortKeyComputer>::SortKey,
-    );
+    type SortKey = (HeadSortKeyComputer::SortKey, TailSortKeyComputer::SortKey);
     type Child = (HeadSortKeyComputer::Child, TailSortKeyComputer::Child);
 
     type Comparator = (
@@ -188,6 +193,11 @@ where
         TailSegmentSortKeyComputer::SegmentSortKey,
     );
 
+    type SegmentComparator = (
+        HeadSegmentSortKeyComputer::SegmentComparator,
+        TailSegmentSortKeyComputer::SegmentComparator,
+    );
+
     /// A SegmentSortKeyComputer maps to a SegmentSortKey, but it can also decide on
     /// its ordering.
     ///
@@ -269,11 +279,12 @@ impl<T, PreviousScore, NewScore> SegmentSortKeyComputer
     for MappedSegmentSortKeyComputer<T, PreviousScore, NewScore>
 where
     T: SegmentSortKeyComputer<SortKey = PreviousScore>,
-    PreviousScore: 'static + Clone + Send + Sync + PartialOrd,
-    NewScore: 'static + Clone + Send + Sync + PartialOrd,
+    PreviousScore: 'static + Clone + Send + Sync,
+    NewScore: 'static + Clone + Send + Sync,
 {
     type SortKey = NewScore;
     type SegmentSortKey = T::SegmentSortKey;
+    type SegmentComparator = T::SegmentComparator;
 
     fn segment_sort_key(&mut self, doc: DocId, score: Score) -> Self::SegmentSortKey {
         self.sort_key_computer.segment_sort_key(doc, score)
@@ -463,6 +474,7 @@ where
 {
     type SortKey = TSortKey;
     type SegmentSortKey = TSortKey;
+    type SegmentComparator = NaturalComparator;
 
     fn segment_sort_key(&mut self, doc: DocId, _score: Score) -> TSortKey {
         (self)(doc)

src/collector/top_collector.rs

@@ -1,64 +1,22 @@
-use std::cmp::Ordering;
-
 use serde::{Deserialize, Serialize};
 
 /// Contains a feature (field, score, etc.) of a document along with the document address.
 ///
-/// It guarantees stable sorting: in case of a tie on the feature, the document
-/// address is used.
-///
-/// The REVERSE_ORDER generic parameter controls whether the by-feature order
-/// should be reversed, which is useful for achieving for example largest-first
-/// semantics without having to wrap the feature in a `Reverse`.
-#[derive(Clone, Default, Serialize, Deserialize)]
-pub struct ComparableDoc<T, D, const REVERSE_ORDER: bool = false> {
+/// Used only by TopNComputer, which implements the actual comparison via a `Comparator`.
+#[derive(Clone, Default, Eq, PartialEq, Serialize, Deserialize)]
+pub struct ComparableDoc<T, D> {
     /// The feature of the document. In practice, this is
-    /// is any type that implements `PartialOrd`.
+    /// is a type which can be compared with a `Comparator<T>`.
     pub sort_key: T,
-    /// The document address. In practice, this is any
-    /// type that implements `PartialOrd`, and is guaranteed
-    /// to be unique for each document.
+    /// The document address. In practice, this is either a `DocId` or `DocAddress`.
     pub doc: D,
 }
-impl<T: std::fmt::Debug, D: std::fmt::Debug, const R: bool> std::fmt::Debug
-    for ComparableDoc<T, D, R>
-{
+
+impl<T: std::fmt::Debug, D: std::fmt::Debug> std::fmt::Debug for ComparableDoc<T, D> {
     fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
-        f.debug_struct(format!("ComparableDoc<_, _ {R}").as_str())
+        f.debug_struct("ComparableDoc")
             .field("feature", &self.sort_key)
             .field("doc", &self.doc)
             .finish()
     }
 }
-
-impl<T: PartialOrd, D: PartialOrd, const R: bool> PartialOrd for ComparableDoc<T, D, R> {
-    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
-        Some(self.cmp(other))
-    }
-}
-
-impl<T: PartialOrd, D: PartialOrd, const R: bool> Ord for ComparableDoc<T, D, R> {
-    #[inline]
-    fn cmp(&self, other: &Self) -> Ordering {
-        let by_feature = self
-            .sort_key
-            .partial_cmp(&other.sort_key)
-            .map(|ord| if R { ord.reverse() } else { ord })
-            .unwrap_or(Ordering::Equal);
-
-        let lazy_by_doc_address = || self.doc.partial_cmp(&other.doc).unwrap_or(Ordering::Equal);
-
-        // In case of a tie on the feature, we sort by ascending
-        // `DocAddress` in order to ensure a stable sorting of the
-        // documents.
-        by_feature.then_with(lazy_by_doc_address)
-    }
-}
-
-impl<T: PartialOrd, D: PartialOrd, const R: bool> PartialEq for ComparableDoc<T, D, R> {
-    fn eq(&self, other: &Self) -> bool {
-        self.cmp(other) == Ordering::Equal
-    }
-}
-
-impl<T: PartialOrd, D: PartialOrd, const R: bool> Eq for ComparableDoc<T, D, R> {}

src/collector/top_score_collector.rs

@@ -23,10 +23,9 @@ use crate::{DocAddress, DocId, Order, Score, SegmentReader};
 /// The theoretical complexity for collecting the top `K` out of `N` documents
 /// is `O(N + K)`.
 ///
-/// This collector does not guarantee a stable sorting in case of a tie on the
-/// document score, for stable sorting `PartialOrd` needs to resolve on other fields
-/// like docid in case of score equality.
-/// Only then, it is suitable for pagination.
+/// This collector guarantees a stable sorting in case of a tie on the
+/// document score/sort key: The document address (`DocAddress`) is used as a tie breaker.
+/// In case of a tie on the sort key, documents are always sorted by ascending `DocAddress`.
 ///
 /// ```rust
 /// use tantivy::collector::TopDocs;
@@ -325,7 +324,7 @@ impl TopDocs {
         sort_key_computer: impl SortKeyComputer<SortKey = TSortKey> + Send + 'static,
     ) -> impl Collector<Fruit = Vec<(TSortKey, DocAddress)>>
     where
-        TSortKey: 'static + Clone + Send + Sync + PartialOrd + std::fmt::Debug,
+        TSortKey: 'static + Clone + Send + Sync + std::fmt::Debug,
     {
         TopBySortKeyCollector::new(sort_key_computer, self.doc_range())
     }
@@ -446,7 +445,7 @@ where
     F: 'static + Send + Sync + Fn(&SegmentReader) -> TTweakScoreSortKeyFn,
     TTweakScoreSortKeyFn: 'static + Fn(DocId, Score) -> TSortKey,
     TweakScoreSegmentSortKeyComputer<TTweakScoreSortKeyFn>:
-        SegmentSortKeyComputer<SortKey = TSortKey>,
+        SegmentSortKeyComputer<SortKey = TSortKey, SegmentSortKey = TSortKey>,
     TSortKey: 'static + PartialOrd + Clone + Send + Sync + std::fmt::Debug,
 {
     type SortKey = TSortKey;
@@ -481,6 +480,7 @@ where
 {
     type SortKey = TSortKey;
     type SegmentSortKey = TSortKey;
+    type SegmentComparator = NaturalComparator;
 
     fn segment_sort_key(&mut self, doc: DocId, score: Score) -> TSortKey {
         (self.sort_key_fn)(doc, score)
@@ -500,8 +500,13 @@ where
 ///
 /// For TopN == 0, it will be relative expensive.
 ///
-/// When using the natural comparator, the top N computer returns the top N elements in
-/// descending order, as expected for a top N.
+/// The TopNComputer will tiebreak by using ascending `D` (DocId or DocAddress):
+/// i.e., in case of a tie on the sort key, the `DocId|DocAddress` are always sorted in
+/// ascending order, regardless of the `Comparator` used for the `Score` type.
+///
+/// NOTE: Items must be `push`ed to the TopNComputer in ascending `DocId|DocAddress` order, as the
+/// threshold used to eliminate docs does not include the `DocId` or `DocAddress`: this provides
+/// the ascending `DocId|DocAddress` tie-breaking behavior without additional comparisons.
 #[derive(Serialize, Deserialize)]
 #[serde(from = "TopNComputerDeser<Score, D, C>")]
 pub struct TopNComputer<Score, D, C> {
@@ -580,6 +585,18 @@ where
     }
 }
 
+#[inline(always)]
+fn compare_for_top_k<TSortKey, D: Ord, C: Comparator<TSortKey>>(
+    c: &C,
+    lhs: &ComparableDoc<TSortKey, D>,
+    rhs: &ComparableDoc<TSortKey, D>,
+) -> std::cmp::Ordering {
+    c.compare(&lhs.sort_key, &rhs.sort_key)
+        .reverse() // Reverse here because we want top K.
+        .then_with(|| lhs.doc.cmp(&rhs.doc)) // Regardless of asc/desc, in presence of a tie, we
+                                             // sort by doc id
+}
+
 impl<TSortKey, D, C> TopNComputer<TSortKey, D, C>
 where
     D: Ord,
@@ -600,10 +617,13 @@ where
 
     /// Push a new document to the top n.
     /// If the document is below the current threshold, it will be ignored.
+    ///
+    /// NOTE: `push` must be called in ascending `DocId`/`DocAddress` order.
     #[inline]
     pub fn push(&mut self, sort_key: TSortKey, doc: D) {
         if let Some(last_median) = &self.threshold {
-            if self.comparator.compare(&sort_key, last_median) == Ordering::Less {
+            // See the struct docs for an explanation of why this comparison is strict.
+            if self.comparator.compare(&sort_key, last_median) != Ordering::Greater {
                 return;
             }
         }
@@ -629,9 +649,7 @@ where
     fn truncate_top_n(&mut self) -> TSortKey {
         // Use select_nth_unstable to find the top nth score
         let (_, median_el, _) = self.buffer.select_nth_unstable_by(self.top_n, |lhs, rhs| {
-            self.comparator
-                .compare(&rhs.sort_key, &lhs.sort_key)
-                .then_with(|| lhs.doc.cmp(&rhs.doc))
+            compare_for_top_k(&self.comparator, lhs, rhs)
         });
 
         let median_score = median_el.sort_key.clone();
@@ -646,11 +664,8 @@ where
         if self.buffer.len() > self.top_n {
             self.truncate_top_n();
         }
-        self.buffer.sort_unstable_by(|left, right| {
-            self.comparator
-                .compare(&right.sort_key, &left.sort_key)
-                .then_with(|| left.doc.cmp(&right.doc))
-        });
+        self.buffer
+            .sort_unstable_by(|lhs, rhs| compare_for_top_k(&self.comparator, lhs, rhs));
         self.buffer
     }
 
@@ -755,6 +770,33 @@ mod tests {
         );
     }
 
+    #[test]
+    fn test_topn_computer_duplicates() {
+        let mut computer: TopNComputer<u32, u32, NaturalComparator> =
+            TopNComputer::new_with_comparator(2, NaturalComparator);
+
+        computer.push(1u32, 1u32);
+        computer.push(1u32, 2u32);
+        computer.push(1u32, 3u32);
+        computer.push(1u32, 4u32);
+        computer.push(1u32, 5u32);
+
+        // In the presence of duplicates, DocIds are always ascending order.
+        assert_eq!(
+            computer.into_sorted_vec(),
+            &[
+                ComparableDoc {
+                    sort_key: 1u32,
+                    doc: 1u32,
+                },
+                ComparableDoc {
+                    sort_key: 1u32,
+                    doc: 2u32,
+                }
+            ]
+        );
+    }
+
     #[test]
     fn test_topn_computer_no_panic() {
         for top_n in 0..10 {
@@ -772,14 +814,17 @@ mod tests {
         #[test]
         fn test_topn_computer_asc_prop(
           limit in 0..10_usize,
-          docs in proptest::collection::vec((0..100_u64, 0..100_u64), 0..100_usize),
+          mut docs in proptest::collection::vec((0..100_u64, 0..100_u64), 0..100_usize),
         ) {
+            // NB: TopNComputer must receive inputs in ascending DocId order.
+            docs.sort_by_key(|(_, doc_id)| *doc_id);
             let mut computer: TopNComputer<_, _, ReverseComparator> = TopNComputer::new_with_comparator(limit, ReverseComparator);
             for (feature, doc) in &docs {
                 computer.push(*feature, *doc);
             }
-            let mut comparable_docs: Vec<ComparableDoc<u64, u64>> = docs.into_iter().map(|(sort_key, doc)| ComparableDoc { sort_key, doc }).collect::<Vec<_>>();
-            comparable_docs.sort();
+            let mut comparable_docs: Vec<ComparableDoc<u64, u64>> =
+                docs.into_iter().map(|(sort_key, doc)| ComparableDoc { sort_key, doc }).collect();
+            crate::collector::sort_key::tests::sort_hits(&mut comparable_docs, Order::Asc);
             comparable_docs.truncate(limit);
             prop_assert_eq!(
                 computer.into_sorted_vec(),
@@ -1406,15 +1451,10 @@ mod tests {
 
             // Using the TopDocs collector should always be equivalent to sorting, skipping the
             // offset, and then taking the limit.
-            let sorted_docs: Vec<_> = if order.is_desc() {
-                let mut comparable_docs: Vec<ComparableDoc<_, _, true>> =
+            let sorted_docs: Vec<_> = {
+                let mut comparable_docs: Vec<ComparableDoc<_, _>> =
                     all_results.into_iter().map(|(sort_key, doc)| ComparableDoc { sort_key, doc}).collect();
-                comparable_docs.sort();
-                comparable_docs.into_iter().map(|cd| (cd.sort_key, cd.doc)).collect()
-            } else {
-                let mut comparable_docs: Vec<ComparableDoc<_, _, false>> =
-                    all_results.into_iter().map(|(sort_key, doc)| ComparableDoc { sort_key, doc}).collect();
-                comparable_docs.sort();
+                crate::collector::sort_key::tests::sort_hits(&mut comparable_docs, order);
                 comparable_docs.into_iter().map(|cd| (cd.sort_key, cd.doc)).collect()
             };
             let expected_docs = sorted_docs.into_iter().skip(offset).take(limit).collect::<Vec<_>>();

src/core/json_utils.rs

@@ -227,9 +227,6 @@ pub(crate) fn index_json_value<'a, V: Value<'a>>(
             ReferenceValueLeaf::IpAddr(_) => {
                 unimplemented!("IP address support in dynamic fields is not yet implemented")
             }
-            ReferenceValueLeaf::Geometry(_) => {
-                unimplemented!("Geometry support in dynamic fields is not implemented")
-            }
         },
         ReferenceValue::Array(elements) => {
             for val in elements {
@@ -409,7 +406,7 @@ mod tests {
         let mut term = Term::from_field_json_path(field, "color", false);
         term.append_type_and_str("red");
 
-        assert_eq!(term.serialized_term(), b"\x00\x00\x00\x01jcolor\x00sred")
+        assert_eq!(term.serialized_value_bytes(), b"color\x00sred".to_vec())
     }
 
     #[test]
@@ -419,8 +416,8 @@ mod tests {
         term.append_type_and_fast_value(-4i64);
 
         assert_eq!(
-            term.serialized_term(),
-            b"\x00\x00\x00\x01jcolor\x00i\x7f\xff\xff\xff\xff\xff\xff\xfc"
+            term.serialized_value_bytes(),
+            b"color\x00i\x7f\xff\xff\xff\xff\xff\xff\xfc".to_vec()
         )
     }
 
@@ -431,8 +428,8 @@ mod tests {
         term.append_type_and_fast_value(4u64);
 
         assert_eq!(
-            term.serialized_term(),
-            b"\x00\x00\x00\x01jcolor\x00u\x00\x00\x00\x00\x00\x00\x00\x04"
+            term.serialized_value_bytes(),
+            b"color\x00u\x00\x00\x00\x00\x00\x00\x00\x04".to_vec()
         )
     }
 
@@ -442,8 +439,8 @@ mod tests {
         let mut term = Term::from_field_json_path(field, "color", false);
         term.append_type_and_fast_value(4.0f64);
         assert_eq!(
-            term.serialized_term(),
-            b"\x00\x00\x00\x01jcolor\x00f\xc0\x10\x00\x00\x00\x00\x00\x00"
+            term.serialized_value_bytes(),
+            b"color\x00f\xc0\x10\x00\x00\x00\x00\x00\x00".to_vec()
         )
     }
 
@@ -453,8 +450,8 @@ mod tests {
         let mut term = Term::from_field_json_path(field, "color", false);
         term.append_type_and_fast_value(true);
         assert_eq!(
-            term.serialized_term(),
-            b"\x00\x00\x00\x01jcolor\x00o\x00\x00\x00\x00\x00\x00\x00\x01"
+            term.serialized_value_bytes(),
+            b"color\x00o\x00\x00\x00\x00\x00\x00\x00\x01".to_vec()
         )
     }
 

src/directory/composite_file.rs

@@ -5,7 +5,7 @@ use std::ops::Range;
 use common::{BinarySerializable, CountingWriter, HasLen, VInt};
 
 use crate::directory::{FileSlice, TerminatingWrite, WritePtr};
-use crate::schema::Field;
+use crate::schema::{Field, Schema};
 use crate::space_usage::{FieldUsage, PerFieldSpaceUsage};
 
 #[derive(Eq, PartialEq, Hash, Copy, Ord, PartialOrd, Clone, Debug)]
@@ -167,10 +167,11 @@ impl CompositeFile {
             .map(|byte_range| self.data.slice(byte_range.clone()))
     }
 
-    pub fn space_usage(&self) -> PerFieldSpaceUsage {
+    pub fn space_usage(&self, schema: &Schema) -> PerFieldSpaceUsage {
         let mut fields = Vec::new();
         for (&field_addr, byte_range) in &self.offsets_index {
-            let mut field_usage = FieldUsage::empty(field_addr.field);
+            let field_name = schema.get_field_name(field_addr.field).to_string();
+            let mut field_usage = FieldUsage::empty(field_name);
             field_usage.add_field_idx(field_addr.idx, byte_range.len().into());
             fields.push(field_usage);
         }

src/directory/mmap_directory/mod.rs

@@ -1,3 +1,5 @@
+mod file_watcher;
+
 use std::collections::HashMap;
 use std::fmt;
 use std::fs::{self, File, OpenOptions};
@@ -7,6 +9,7 @@ use std::path::{Path, PathBuf};
 use std::sync::{Arc, RwLock, Weak};
 
 use common::StableDeref;
+use file_watcher::FileWatcher;
 use fs4::fs_std::FileExt;
 #[cfg(all(feature = "mmap", unix))]
 pub use memmap2::Advice;
@@ -18,7 +21,6 @@ use crate::core::META_FILEPATH;
 use crate::directory::error::{
     DeleteError, LockError, OpenDirectoryError, OpenReadError, OpenWriteError,
 };
-use crate::directory::file_watcher::FileWatcher;
 use crate::directory::{
     AntiCallToken, Directory, DirectoryLock, FileHandle, Lock, OwnedBytes, TerminatingWrite,
     WatchCallback, WatchHandle, WritePtr,

src/directory/mod.rs

@@ -5,7 +5,6 @@ mod mmap_directory;
 
 mod directory;
 mod directory_lock;
-mod file_watcher;
 pub mod footer;
 mod managed_directory;
 mod ram_directory;

src/docset.rs

@@ -40,6 +40,8 @@ pub trait DocSet: Send {
     /// of `DocSet` should support it.
     ///
     /// Calling `seek(TERMINATED)` is also legal and is the normal way to consume a `DocSet`.
+    ///
+    /// `target` has to be larger or equal to `.doc()` when calling `seek`.
     fn seek(&mut self, target: DocId) -> DocId {
         let mut doc = self.doc();
         debug_assert!(doc <= target);
@@ -49,6 +51,33 @@ pub trait DocSet: Send {
         doc
     }
 
+    /// Seeks to the target if possible and returns true if the target is in the DocSet.
+    ///
+    /// DocSets that already have an efficient `seek` method don't need to implement
+    /// `seek_into_the_danger_zone`. All wrapper DocSets should forward
+    /// `seek_into_the_danger_zone` to the underlying DocSet.
+    ///
+    /// ## API Behaviour
+    /// If `seek_into_the_danger_zone` is returning true, a call to `doc()` has to return target.
+    /// If `seek_into_the_danger_zone` is returning false, a call to `doc()` may return any doc
+    /// between the last doc that matched and target or a doc that is a valid next hit after
+    /// target. The DocSet is considered to be in an invalid state until
+    /// `seek_into_the_danger_zone` returns true again.
+    ///
+    /// `target` needs to be equal or larger than `doc` when in a valid state.
+    ///
+    /// Consecutive calls are not allowed to have decreasing `target` values.
+    ///
+    /// # Warning
+    /// This is an advanced API used by intersection. The API contract is tricky, avoid using it.
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        let current_doc = self.doc();
+        if current_doc < target {
+            self.seek(target);
+        }
+        self.doc() == target
+    }
+
     /// Fills a given mutable buffer with the next doc ids from the
     /// `DocSet`
     ///
@@ -94,6 +123,15 @@ pub trait DocSet: Send {
     /// which would be the number of documents in the DocSet.
     ///
     /// By default this returns `size_hint()`.
+    ///
+    /// DocSets may have vastly different cost depending on their type,
+    /// e.g. an intersection with 10 hits is much cheaper than
+    /// a phrase search with 10 hits, since it needs to load positions.
+    ///
+    /// ### Future Work
+    /// We may want to differentiate `DocSet` costs more more granular, e.g.
+    /// creation_cost, advance_cost, seek_cost on to get a good estimation
+    /// what query types to choose.
     fn cost(&self) -> u64 {
         self.size_hint() as u64
     }
@@ -137,6 +175,10 @@ impl DocSet for &mut dyn DocSet {
         (**self).seek(target)
     }
 
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        (**self).seek_into_the_danger_zone(target)
+    }
+
     fn doc(&self) -> u32 {
         (**self).doc()
     }
@@ -169,6 +211,11 @@ impl<TDocSet: DocSet + ?Sized> DocSet for Box<TDocSet> {
         unboxed.seek(target)
     }
 
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        let unboxed: &mut TDocSet = self.borrow_mut();
+        unboxed.seek_into_the_danger_zone(target)
+    }
+
     fn fill_buffer(&mut self, buffer: &mut [DocId; COLLECT_BLOCK_BUFFER_LEN]) -> usize {
         let unboxed: &mut TDocSet = self.borrow_mut();
         unboxed.fill_buffer(buffer)

src/fastfield/mod.rs

@@ -683,7 +683,7 @@ mod tests {
     }
 
     #[test]
-    fn test_datefastfield() {
+    fn test_datefastfield() -> crate::Result<()> {
         let mut schema_builder = Schema::builder();
         let date_field = schema_builder.add_date_field(
             "date",
@@ -697,28 +697,22 @@ mod tests {
         );
         let schema = schema_builder.build();
         let index = Index::create_in_ram(schema);
-        let mut index_writer = index.writer_for_tests().unwrap();
+        let mut index_writer = index.writer_for_tests()?;
         index_writer.set_merge_policy(Box::new(NoMergePolicy));
-        index_writer
-            .add_document(doc!(
-                date_field => DateTime::from_u64(1i64.to_u64()),
-                multi_date_field => DateTime::from_u64(2i64.to_u64()),
-                multi_date_field => DateTime::from_u64(3i64.to_u64())
-            ))
-            .unwrap();
-        index_writer
-            .add_document(doc!(
-                date_field => DateTime::from_u64(4i64.to_u64())
-            ))
-            .unwrap();
-        index_writer
-            .add_document(doc!(
-                multi_date_field => DateTime::from_u64(5i64.to_u64()),
-                multi_date_field => DateTime::from_u64(6i64.to_u64())
-            ))
-            .unwrap();
-        index_writer.commit().unwrap();
-        let reader = index.reader().unwrap();
+        index_writer.add_document(doc!(
+            date_field => DateTime::from_u64(1i64.to_u64()),
+            multi_date_field => DateTime::from_u64(2i64.to_u64()),
+            multi_date_field => DateTime::from_u64(3i64.to_u64())
+        ))?;
+        index_writer.add_document(doc!(
+            date_field => DateTime::from_u64(4i64.to_u64())
+        ))?;
+        index_writer.add_document(doc!(
+            multi_date_field => DateTime::from_u64(5i64.to_u64()),
+            multi_date_field => DateTime::from_u64(6i64.to_u64())
+        ))?;
+        index_writer.commit()?;
+        let reader = index.reader()?;
         let searcher = reader.searcher();
         assert_eq!(searcher.segment_readers().len(), 1);
         let segment_reader = searcher.segment_reader(0);
@@ -752,6 +746,7 @@ mod tests {
             assert_eq!(dates[0].into_timestamp_nanos(), 5i64);
             assert_eq!(dates[1].into_timestamp_nanos(), 6i64);
         }
+        Ok(())
     }
 
     #[test]

src/fastfield/readers.rs

@@ -8,7 +8,7 @@ use columnar::{
 };
 use common::ByteCount;
 
-use crate::core::json_utils::encode_column_name;
+use crate::core::json_utils::{encode_column_name, json_path_sep_to_dot};
 use crate::directory::FileSlice;
 use crate::schema::{Field, FieldEntry, FieldType, Schema};
 use crate::space_usage::{FieldUsage, PerFieldSpaceUsage};
@@ -39,19 +39,15 @@ impl FastFieldReaders {
         self.resolve_column_name_given_default_field(column_name, default_field_opt)
     }
 
-    pub(crate) fn space_usage(&self, schema: &Schema) -> io::Result<PerFieldSpaceUsage> {
+    pub(crate) fn space_usage(&self) -> io::Result<PerFieldSpaceUsage> {
         let mut per_field_usages: Vec<FieldUsage> = Default::default();
-        for (field, field_entry) in schema.fields() {
-            let column_handles = self.columnar.read_columns(field_entry.name())?;
-            let num_bytes: ByteCount = column_handles
-                .iter()
-                .map(|column_handle| column_handle.num_bytes())
-                .sum();
-            let mut field_usage = FieldUsage::empty(field);
-            field_usage.add_field_idx(0, num_bytes);
+        for (mut field_name, column_handle) in self.columnar.iter_columns()? {
+            json_path_sep_to_dot(&mut field_name);
+            let space_usage = column_handle.space_usage()?;
+            let mut field_usage = FieldUsage::empty(field_name);
+            field_usage.set_column_usage(space_usage);
             per_field_usages.push(field_usage);
         }
-        // TODO fix space usage for JSON fields.
         Ok(PerFieldSpaceUsage::new(per_field_usages))
     }
 

src/fastfield/writer.rs

@@ -189,9 +189,6 @@ impl FastFieldsWriter {
                             .record_str(doc_id, field_name, &token.text);
                     }
                 }
-                ReferenceValueLeaf::Geometry(_) => {
-                    panic!("Geometry fields should not be routed to fast field writer")
-                }
             },
             ReferenceValue::Array(val) => {
                 // TODO: Check this is the correct behaviour we want.
@@ -323,9 +320,6 @@ fn record_json_value_to_columnar_writer<'a, V: Value<'a>>(
                     "Pre-tokenized string support in dynamic fields is not yet implemented"
                 )
             }
-            ReferenceValueLeaf::Geometry(_) => {
-                unimplemented!("Geometry support in dynamic fields is not yet implemented")
-            }
         },
         ReferenceValue::Array(elements) => {
             for el in elements {

src/fieldnorm/reader.rs

@@ -2,7 +2,7 @@ use std::sync::Arc;
 
 use super::{fieldnorm_to_id, id_to_fieldnorm};
 use crate::directory::{CompositeFile, FileSlice, OwnedBytes};
-use crate::schema::Field;
+use crate::schema::{Field, Schema};
 use crate::space_usage::PerFieldSpaceUsage;
 use crate::DocId;
 
@@ -37,8 +37,8 @@ impl FieldNormReaders {
     }
 
     /// Return a break down of the space usage per field.
-    pub fn space_usage(&self) -> PerFieldSpaceUsage {
-        self.data.space_usage()
+    pub fn space_usage(&self, schema: &Schema) -> PerFieldSpaceUsage {
+        self.data.space_usage(schema)
     }
 
     /// Returns a handle to inner file

src/index/index_meta.rs

@@ -13,9 +13,9 @@ use crate::store::Compressor;
 use crate::{Inventory, Opstamp, TrackedObject};
 
 #[derive(Clone, Debug, Serialize, Deserialize)]
-struct DeleteMeta {
+pub struct DeleteMeta {
     num_deleted_docs: u32,
-    opstamp: Opstamp,
+    pub opstamp: Opstamp,
 }
 
 #[derive(Clone, Default)]
@@ -142,7 +142,6 @@ impl SegmentMeta {
             SegmentComponent::FastFields => ".fast".to_string(),
             SegmentComponent::FieldNorms => ".fieldnorm".to_string(),
             SegmentComponent::Delete => format!(".{}.del", self.delete_opstamp().unwrap_or(0)),
-            SegmentComponent::Spatial => ".spatial".to_string(),
         });
         PathBuf::from(path)
     }
@@ -214,7 +213,7 @@ impl SegmentMeta {
 struct InnerSegmentMeta {
     segment_id: SegmentId,
     max_doc: u32,
-    deletes: Option<DeleteMeta>,
+    pub deletes: Option<DeleteMeta>,
     /// If you want to avoid the SegmentComponent::TempStore file to be covered by
     /// garbage collection and deleted, set this to true. This is used during merge.
     #[serde(skip)]
@@ -405,7 +404,10 @@ mod tests {
             schema_builder.build()
         };
         let index_metas = IndexMeta {
-            index_settings: IndexSettings::default(),
+            index_settings: IndexSettings {
+                docstore_compression: Compressor::None,
+                ..Default::default()
+            },
             segments: Vec::new(),
             schema,
             opstamp: 0u64,
@@ -414,7 +416,7 @@ mod tests {
         let json = serde_json::ser::to_string(&index_metas).expect("serialization failed");
         assert_eq!(
             json,
-            r#"{"index_settings":{"docstore_compression":"lz4","docstore_blocksize":16384},"segments":[],"schema":[{"name":"text","type":"text","options":{"indexing":{"record":"position","fieldnorms":true,"tokenizer":"default"},"stored":false,"fast":false}}],"opstamp":0}"#
+            r#"{"index_settings":{"docstore_compression":"none","docstore_blocksize":16384},"segments":[],"schema":[{"name":"text","type":"text","options":{"indexing":{"record":"position","fieldnorms":true,"tokenizer":"default"},"stored":false,"fast":false}}],"opstamp":0}"#
         );
 
         let deser_meta: UntrackedIndexMeta = serde_json::from_str(&json).unwrap();
@@ -495,6 +497,8 @@ mod tests {
     #[test]
     #[cfg(feature = "lz4-compression")]
     fn test_index_settings_default() {
+        use crate::store::Compressor;
+
         let mut index_settings = IndexSettings::default();
         assert_eq!(
             index_settings,

src/index/segment.rs

@@ -46,7 +46,7 @@ impl Segment {
     ///
     /// This method is only used when updating `max_doc` from 0
     /// as we finalize a fresh new segment.
-    pub(crate) fn with_max_doc(self, max_doc: u32) -> Segment {
+    pub fn with_max_doc(self, max_doc: u32) -> Segment {
         Segment {
             index: self.index,
             meta: self.meta.with_max_doc(max_doc),

src/index/segment_component.rs

@@ -28,14 +28,12 @@ pub enum SegmentComponent {
     /// Bitset describing which document of the segment is alive.
     /// (It was representing deleted docs but changed to represent alive docs from v0.17)
     Delete,
-    /// HUSH
-    Spatial,
 }
 
 impl SegmentComponent {
     /// Iterates through the components.
     pub fn iterator() -> slice::Iter<'static, SegmentComponent> {
-        static SEGMENT_COMPONENTS: [SegmentComponent; 9] = [
+        static SEGMENT_COMPONENTS: [SegmentComponent; 8] = [
             SegmentComponent::Postings,
             SegmentComponent::Positions,
             SegmentComponent::FastFields,
@@ -44,7 +42,6 @@ impl SegmentComponent {
             SegmentComponent::Store,
             SegmentComponent::TempStore,
             SegmentComponent::Delete,
-            SegmentComponent::Spatial,
         ];
         SEGMENT_COMPONENTS.iter()
     }

src/index/segment_reader.rs

@@ -14,7 +14,6 @@ use crate::index::{InvertedIndexReader, Segment, SegmentComponent, SegmentId};
 use crate::json_utils::json_path_sep_to_dot;
 use crate::schema::{Field, IndexRecordOption, Schema, Type};
 use crate::space_usage::SegmentSpaceUsage;
-use crate::spatial::reader::SpatialReaders;
 use crate::store::StoreReader;
 use crate::termdict::TermDictionary;
 use crate::{DocId, Opstamp};
@@ -44,7 +43,6 @@ pub struct SegmentReader {
     positions_composite: CompositeFile,
     fast_fields_readers: FastFieldReaders,
     fieldnorm_readers: FieldNormReaders,
-    spatial_readers: SpatialReaders,
 
     store_file: FileSlice,
     alive_bitset_opt: Option<AliveBitSet>,
@@ -94,11 +92,6 @@ impl SegmentReader {
         &self.fast_fields_readers
     }
 
-    /// HUSH
-    pub fn spatial_fields(&self) -> &SpatialReaders {
-        &self.spatial_readers
-    }
-
     /// Accessor to the `FacetReader` associated with a given `Field`.
     pub fn facet_reader(&self, field_name: &str) -> crate::Result<FacetReader> {
         let schema = self.schema();
@@ -180,12 +173,6 @@ impl SegmentReader {
         let fast_fields_readers = FastFieldReaders::open(fast_fields_data, schema.clone())?;
         let fieldnorm_data = segment.open_read(SegmentComponent::FieldNorms)?;
         let fieldnorm_readers = FieldNormReaders::open(fieldnorm_data)?;
-        let spatial_readers = if schema.contains_spatial_field() {
-            let spatial_data = segment.open_read(SegmentComponent::Spatial)?;
-            SpatialReaders::open(spatial_data)?
-        } else {
-            SpatialReaders::empty()
-        };
 
         let original_bitset = if segment.meta().has_deletes() {
             let alive_doc_file_slice = segment.open_read(SegmentComponent::Delete)?;
@@ -211,7 +198,6 @@ impl SegmentReader {
             postings_composite,
             fast_fields_readers,
             fieldnorm_readers,
-            spatial_readers,
             segment_id: segment.id(),
             delete_opstamp: segment.meta().delete_opstamp(),
             store_file,
@@ -469,12 +455,11 @@ impl SegmentReader {
     pub fn space_usage(&self) -> io::Result<SegmentSpaceUsage> {
         Ok(SegmentSpaceUsage::new(
             self.num_docs(),
-            self.termdict_composite.space_usage(),
-            self.postings_composite.space_usage(),
-            self.positions_composite.space_usage(),
-            self.fast_fields_readers.space_usage(self.schema())?,
-            self.fieldnorm_readers.space_usage(),
-            self.spatial_readers.space_usage(),
+            self.termdict_composite.space_usage(self.schema()),
+            self.postings_composite.space_usage(self.schema()),
+            self.positions_composite.space_usage(self.schema()),
+            self.fast_fields_readers.space_usage()?,
+            self.fieldnorm_readers.space_usage(self.schema()),
             self.get_store_reader(0)?.space_usage(),
             self.alive_bitset_opt
                 .as_ref()

src/indexer/delete_queue.rs

@@ -4,38 +4,37 @@ use std::sync::{Arc, RwLock, Weak};
 use super::operation::DeleteOperation;
 use crate::Opstamp;
 
-// The DeleteQueue is similar in conceptually to a multiple
-// consumer single producer broadcast channel.
-//
-// All consumer will receive all messages.
-//
-// Consumer of the delete queue are holding a `DeleteCursor`,
-// which points to a specific place of the `DeleteQueue`.
-//
-// New consumer can be created in two ways
-// - calling `delete_queue.cursor()` returns a cursor, that will include all future delete operation
-//   (and some or none of the past operations... The client is in charge of checking the opstamps.).
-// - cloning an existing cursor returns a new cursor, that is at the exact same position, and can
-//   now advance independently from the original cursor.
+/// The DeleteQueue is similar in conceptually to a multiple
+/// consumer single producer broadcast channel.
+///
+/// All consumer will receive all messages.
+///
+/// Consumer of the delete queue are holding a `DeleteCursor`,
+/// which points to a specific place of the `DeleteQueue`.
+///
+/// New consumer can be created in two ways
+/// - calling `delete_queue.cursor()` returns a cursor, that will include all future delete
+///   operation (and some or none of the past operations... The client is in charge of checking the
+///   opstamps.).
+/// - cloning an existing cursor returns a new cursor, that is at the exact same position, and can
+///   now advance independently from the original cursor.
 #[derive(Default)]
 struct InnerDeleteQueue {
     writer: Vec<DeleteOperation>,
     last_block: Weak<Block>,
 }
 
-#[derive(Clone)]
+/// The delete queue is a linked list storing delete operations.
+///
+/// Several consumers can hold a reference to it. Delete operations
+/// get dropped/gc'ed when no more consumers are holding a reference
+/// to them.
+#[derive(Clone, Default)]
 pub struct DeleteQueue {
     inner: Arc<RwLock<InnerDeleteQueue>>,
 }
 
 impl DeleteQueue {
-    // Creates a new delete queue.
-    pub fn new() -> DeleteQueue {
-        DeleteQueue {
-            inner: Arc::default(),
-        }
-    }
-
     fn get_last_block(&self) -> Arc<Block> {
         {
             // try get the last block with simply acquiring the read lock.
@@ -58,10 +57,10 @@ impl DeleteQueue {
         block
     }
 
-    // Creates a new cursor that makes it possible to
-    // consume future delete operations.
-    //
-    // Past delete operations are not accessible.
+    /// Creates a new cursor that makes it possible to
+    /// consume future delete operations.
+    ///
+    /// Past delete operations are not accessible.
     pub fn cursor(&self) -> DeleteCursor {
         let last_block = self.get_last_block();
         let operations_len = last_block.operations.len();
@@ -71,7 +70,7 @@ impl DeleteQueue {
         }
     }
 
-    // Appends a new delete operations.
+    /// Appends a new delete operations.
     pub fn push(&self, delete_operation: DeleteOperation) {
         self.inner
             .write()
@@ -169,6 +168,7 @@ struct Block {
     next: NextBlock,
 }
 
+/// As we process delete operations, keeps track of our position.
 #[derive(Clone)]
 pub struct DeleteCursor {
     block: Arc<Block>,
@@ -261,7 +261,7 @@ mod tests {
 
     #[test]
     fn test_deletequeue() {
-        let delete_queue = DeleteQueue::new();
+        let delete_queue = DeleteQueue::default();
 
         let make_op = |i: usize| DeleteOperation {
             opstamp: i as u64,

src/indexer/index_writer.rs

@@ -128,7 +128,7 @@ fn compute_deleted_bitset(
 /// is `==` target_opstamp.
 /// For instance, there was no delete operation between the state of the `segment_entry` and
 /// the `target_opstamp`, `segment_entry` is not updated.
-pub(crate) fn advance_deletes(
+pub fn advance_deletes(
     mut segment: Segment,
     segment_entry: &mut SegmentEntry,
     target_opstamp: Opstamp,
@@ -303,7 +303,7 @@ impl<D: Document> IndexWriter<D> {
         let (document_sender, document_receiver) =
             crossbeam_channel::bounded(PIPELINE_MAX_SIZE_IN_DOCS);
 
-        let delete_queue = DeleteQueue::new();
+        let delete_queue = DeleteQueue::default();
 
         let current_opstamp = index.load_metas()?.opstamp;
 

src/indexer/indexing_term.rs

@@ -3,21 +3,21 @@ use std::net::Ipv6Addr;
 use columnar::MonotonicallyMappableToU128;
 
 use crate::fastfield::FastValue;
-use crate::schema::{Field, Type};
+use crate::schema::Field;
 
-/// Term represents the value that the token can take.
-/// It's a serialized representation over different types.
+/// IndexingTerm is used to represent a term during indexing.
+/// It's a serialized representation over field and value.
 ///
-/// It actually wraps a `Vec<u8>`. The first 5 bytes are metadata.
-/// 4 bytes are the field id, and the last byte is the type.
+/// It actually wraps a `Vec<u8>`. The first 4 bytes are the field.
 ///
-/// The serialized value `ValueBytes` is considered everything after the 4 first bytes (term id).
+/// We serialize the field, because we index everything in a single
+/// global term dictionary during indexing.
 #[derive(Clone)]
 pub(crate) struct IndexingTerm<B = Vec<u8>>(B)
 where B: AsRef<[u8]>;
 
 /// The number of bytes used as metadata by `Term`.
-const TERM_METADATA_LENGTH: usize = 5;
+const TERM_METADATA_LENGTH: usize = 4;
 
 impl IndexingTerm {
     /// Create a new Term with a buffer with a given capacity.
@@ -31,10 +31,9 @@ impl IndexingTerm {
     /// Use `clear_with_field_and_type` in that case.
     ///
     /// Sets field and the type.
-    pub(crate) fn set_field_and_type(&mut self, field: Field, typ: Type) {
+    pub(crate) fn set_field(&mut self, field: Field) {
         assert!(self.is_empty());
         self.0[0..4].clone_from_slice(field.field_id().to_be_bytes().as_ref());
-        self.0[4] = typ.to_code();
     }
 
     /// Is empty if there are no value bytes.
@@ -42,10 +41,10 @@ impl IndexingTerm {
         self.0.len() == TERM_METADATA_LENGTH
     }
 
-    /// Removes the value_bytes and set the field and type code.
-    pub(crate) fn clear_with_field_and_type(&mut self, typ: Type, field: Field) {
+    /// Removes the value_bytes and set the field
+    pub(crate) fn clear_with_field(&mut self, field: Field) {
         self.truncate_value_bytes(0);
-        self.set_field_and_type(field, typ);
+        self.set_field(field);
     }
 
     /// Sets a u64 value in the term.
@@ -122,6 +121,23 @@ impl IndexingTerm {
 impl<B> IndexingTerm<B>
 where B: AsRef<[u8]>
 {
+    /// Wraps serialized term bytes.
+    ///
+    /// The input buffer is expected to be the concatenation of the big endian encoded field id
+    /// followed by the serialized value bytes (type tag + payload).
+    #[inline]
+    pub fn wrap(serialized_term: B) -> IndexingTerm<B> {
+        debug_assert!(serialized_term.as_ref().len() >= TERM_METADATA_LENGTH);
+        IndexingTerm(serialized_term)
+    }
+
+    /// Returns the field this term belongs to.
+    #[inline]
+    pub fn field(&self) -> Field {
+        let field_id_bytes: [u8; 4] = self.0.as_ref()[..4].try_into().unwrap();
+        Field::from_field_id(u32::from_be_bytes(field_id_bytes))
+    }
+
     /// Returns the serialized representation of Term.
     /// This includes field_id, value type and value.
     ///
@@ -136,6 +152,7 @@ where B: AsRef<[u8]>
 #[cfg(test)]
 mod tests {
 
+    use super::IndexingTerm;
     use crate::schema::*;
 
     #[test]
@@ -143,42 +160,55 @@ mod tests {
         let mut schema_builder = Schema::builder();
         schema_builder.add_text_field("text", STRING);
         let title_field = schema_builder.add_text_field("title", STRING);
-        let term = Term::from_field_text(title_field, "test");
+        let mut term = IndexingTerm::with_capacity(0);
+        term.set_field(title_field);
+        term.set_bytes(b"test");
         assert_eq!(term.field(), title_field);
-        assert_eq!(term.typ(), Type::Str);
-        assert_eq!(term.value().as_str(), Some("test"))
+        assert_eq!(term.serialized_term(), b"\x00\x00\x00\x01test".to_vec())
     }
 
     /// Size (in bytes) of the buffer of a fast value (u64, i64, f64, or date) term.
     /// <field> + <type byte> + <value len>
     ///
     /// - <field> is a big endian encoded u32 field id
-    /// - <type_byte>'s most significant bit expresses whether the term is a json term or not The
-    ///   remaining 7 bits are used to encode the type of the value. If this is a JSON term, the
-    ///   type is the type of the leaf of the json.
     /// - <value> is,  if this is not the json term, a binary representation specific to the type.
     ///   If it is a JSON Term, then it is prepended with the path that leads to this leaf value.
-    const FAST_VALUE_TERM_LEN: usize = 4 + 1 + 8;
+    const FAST_VALUE_TERM_LEN: usize = 4 + 8;
 
     #[test]
     pub fn test_term_u64() {
         let mut schema_builder = Schema::builder();
         let count_field = schema_builder.add_u64_field("count", INDEXED);
-        let term = Term::from_field_u64(count_field, 983u64);
+        let mut term = IndexingTerm::with_capacity(0);
+        term.set_field(count_field);
+        term.set_u64(983u64);
         assert_eq!(term.field(), count_field);
-        assert_eq!(term.typ(), Type::U64);
         assert_eq!(term.serialized_term().len(), FAST_VALUE_TERM_LEN);
-        assert_eq!(term.value().as_u64(), Some(983u64))
     }
 
     #[test]
     pub fn test_term_bool() {
         let mut schema_builder = Schema::builder();
         let bool_field = schema_builder.add_bool_field("bool", INDEXED);
-        let term = Term::from_field_bool(bool_field, true);
+        let term = {
+            let mut term = IndexingTerm::with_capacity(0);
+            term.set_field(bool_field);
+            term.set_bool(true);
+            term
+        };
         assert_eq!(term.field(), bool_field);
-        assert_eq!(term.typ(), Type::Bool);
         assert_eq!(term.serialized_term().len(), FAST_VALUE_TERM_LEN);
-        assert_eq!(term.value().as_bool(), Some(true))
+    }
+
+    #[test]
+    pub fn indexing_term_wrap_extracts_field() {
+        let field = Field::from_field_id(7u32);
+        let mut term = IndexingTerm::with_capacity(0);
+        term.set_field(field);
+        term.append_bytes(b"abc");
+
+        let wrapped = IndexingTerm::wrap(term.serialized_term());
+        assert_eq!(wrapped.field(), field);
+        assert_eq!(wrapped.serialized_term(), term.serialized_term());
     }
 }

src/indexer/merger.rs

@@ -1,5 +1,3 @@
-use std::collections::HashMap;
-use std::io::{BufWriter, Write};
 use std::sync::Arc;
 
 use columnar::{
@@ -8,7 +6,6 @@ use columnar::{
 use common::ReadOnlyBitSet;
 use itertools::Itertools;
 use measure_time::debug_time;
-use tempfile::NamedTempFile;
 
 use crate::directory::WritePtr;
 use crate::docset::{DocSet, TERMINATED};
@@ -20,8 +17,6 @@ use crate::indexer::doc_id_mapping::{MappingType, SegmentDocIdMapping};
 use crate::indexer::SegmentSerializer;
 use crate::postings::{InvertedIndexSerializer, Postings, SegmentPostings};
 use crate::schema::{value_type_to_column_type, Field, FieldType, Schema};
-use crate::spatial::bkd::LeafPageIterator;
-use crate::spatial::triangle::Triangle;
 use crate::store::StoreWriter;
 use crate::termdict::{TermMerger, TermOrdinal};
 use crate::{DocAddress, DocId, InvertedIndexReader};
@@ -175,7 +170,6 @@ impl IndexMerger {
         let mut readers = vec![];
         for (segment, new_alive_bitset_opt) in segments.iter().zip(alive_bitset_opt) {
             if segment.meta().num_docs() > 0 {
-                dbg!("segment");
                 let reader =
                     SegmentReader::open_with_custom_alive_set(segment, new_alive_bitset_opt)?;
                 readers.push(reader);
@@ -526,89 +520,6 @@ impl IndexMerger {
         Ok(())
     }
 
-    fn write_spatial_fields(
-        &self,
-        serializer: &mut SegmentSerializer,
-        doc_id_mapping: &SegmentDocIdMapping,
-    ) -> crate::Result<()> {
-        /// We need to rebuild a BKD-tree based off the list of triangles.
-        ///
-        /// Because the data can be large, we do this by writing the sequence of triangles to
-        /// disk, and mmapping it as mutable slice, and calling the same code as what
-        /// is done for the segment serialization.
-        ///
-        /// The OS is in charge of deciding how to handle its page cache.
-        /// This is the same as what would have happened with swapping,
-        /// except by explicitly mapping the file, the OS is more likely to
-        /// swap, the memory will not be accounted as anonymous memory,
-        /// swap space is reserved etc.
-        use crate::spatial::bkd::Segment;
-
-        let Some(mut spatial_serializer) = serializer.extract_spatial_serializer() else {
-            // The schema does not contain any spatial field.
-            return Ok(());
-        };
-
-        let mut segment_mappings: Vec<Vec<Option<DocId>>> = Vec::new();
-        for reader in &self.readers {
-            let max_doc = reader.max_doc();
-            segment_mappings.push(vec![None; max_doc as usize]);
-        }
-        for (new_doc_id, old_doc_addr) in doc_id_mapping.iter_old_doc_addrs().enumerate() {
-            segment_mappings[old_doc_addr.segment_ord as usize][old_doc_addr.doc_id as usize] =
-                Some(new_doc_id as DocId);
-        }
-        let mut temp_files: HashMap<Field, NamedTempFile> = HashMap::new();
-
-        for (field, field_entry) in self.schema.fields() {
-            if matches!(field_entry.field_type(), FieldType::Spatial(_)) {
-                temp_files.insert(field, NamedTempFile::new()?);
-            }
-        }
-        for (segment_ord, reader) in self.readers.iter().enumerate() {
-            for (field, temp_file) in &mut temp_files {
-                let mut buf_temp_file = BufWriter::new(temp_file);
-                let spatial_readers = reader.spatial_fields();
-                let Some(spatial_reader) = spatial_readers.get_field(*field)? else {
-                    continue;
-                };
-                let segment = Segment::new(spatial_reader.get_bytes());
-                for triangle_result in LeafPageIterator::new(&segment) {
-                    let triangles = triangle_result?;
-                    for triangle in triangles {
-                        if let Some(new_doc_id) =
-                            segment_mappings[segment_ord][triangle.doc_id as usize]
-                        {
-                            // This is really just a temporary file, not meant to be portable, so we
-                            // use native endianness here.
-                            for &word in &triangle.words {
-                                buf_temp_file.write_all(&word.to_ne_bytes())?;
-                            }
-                            buf_temp_file.write_all(&new_doc_id.to_ne_bytes())?;
-                        }
-                    }
-                }
-                buf_temp_file.flush()?;
-                // No need to fsync here. This file is not here for persistency.
-            }
-        }
-        for (field, temp_file) in temp_files {
-            // Memory map the triangle file.
-            use memmap2::MmapOptions;
-            let mmap = unsafe { MmapOptions::new().map_mut(temp_file.as_file())? };
-            // Cast to &[Triangle] slice
-            let triangle_count = mmap.len() / std::mem::size_of::<Triangle>();
-            let triangles = unsafe {
-                std::slice::from_raw_parts_mut(mmap.as_ptr() as *mut Triangle, triangle_count)
-            };
-            // Get spatial writer and rebuild block kd-tree.
-            spatial_serializer.serialize_field(field, triangles)?;
-        }
-        spatial_serializer.close()?;
-
-        Ok(())
-    }
-
     /// Writes the merged segment by pushing information
     /// to the `SegmentSerializer`.
     ///
@@ -633,10 +544,9 @@ impl IndexMerger {
 
         debug!("write-storagefields");
         self.write_storable_fields(serializer.get_store_writer())?;
-        debug!("write-spatialfields");
-        self.write_spatial_fields(&mut serializer, &doc_id_mapping)?;
         debug!("write-fastfields");
         self.write_fast_fields(serializer.get_fast_field_write(), doc_id_mapping)?;
+
         debug!("close-serializer");
         serializer.close()?;
         Ok(self.max_doc)

src/indexer/mod.rs

@@ -4,6 +4,7 @@
 //! `IndexWriter` is the main entry point for that, which created from
 //! [`Index::writer`](crate::Index::writer).
 
+/// Delete queue implementation for broadcasting delete operations to consumers.
 pub(crate) mod delete_queue;
 pub(crate) mod path_to_unordered_id;
 
@@ -32,12 +33,11 @@ mod stamper;
 use crossbeam_channel as channel;
 use smallvec::SmallVec;
 
-pub use self::index_writer::{IndexWriter, IndexWriterOptions};
+pub use self::index_writer::{advance_deletes, IndexWriter, IndexWriterOptions};
 pub use self::log_merge_policy::LogMergePolicy;
 pub use self::merge_operation::MergeOperation;
 pub use self::merge_policy::{MergeCandidate, MergePolicy, NoMergePolicy};
-use self::operation::AddOperation;
-pub use self::operation::UserOperation;
+pub use self::operation::{AddOperation, DeleteOperation, UserOperation};
 pub use self::prepared_commit::PreparedCommit;
 pub use self::segment_entry::SegmentEntry;
 pub(crate) use self::segment_serializer::SegmentSerializer;

src/indexer/operation.rs

@@ -5,14 +5,20 @@ use crate::Opstamp;
 
 /// Timestamped Delete operation.
 pub struct DeleteOperation {
+    /// Operation stamp.
+    /// It is used to check whether the delete operation
+    /// applies to an added document operation.
     pub opstamp: Opstamp,
+    /// Weight is used to define the set of documents to be deleted.
     pub target: Box<dyn Weight>,
 }
 
 /// Timestamped Add operation.
 #[derive(Eq, PartialEq, Debug)]
 pub struct AddOperation<D: Document = TantivyDocument> {
+    /// Operation stamp.
     pub opstamp: Opstamp,
+    /// Document to be added.
     pub document: D,
 }
 

src/indexer/segment_register.rs

@@ -117,7 +117,7 @@ mod tests {
     #[test]
     fn test_segment_register() {
         let inventory = SegmentMetaInventory::default();
-        let delete_queue = DeleteQueue::new();
+        let delete_queue = DeleteQueue::default();
 
         let mut segment_register = SegmentRegister::default();
         let segment_id_a = SegmentId::generate_random();

src/indexer/segment_serializer.rs

@@ -4,7 +4,6 @@ use crate::directory::WritePtr;
 use crate::fieldnorm::FieldNormsSerializer;
 use crate::index::{Segment, SegmentComponent};
 use crate::postings::InvertedIndexSerializer;
-use crate::spatial::serializer::SpatialSerializer;
 use crate::store::StoreWriter;
 
 /// Segment serializer is in charge of laying out on disk
@@ -13,7 +12,6 @@ pub struct SegmentSerializer {
     segment: Segment,
     pub(crate) store_writer: StoreWriter,
     fast_field_write: WritePtr,
-    spatial_serializer: Option<SpatialSerializer>,
     fieldnorms_serializer: Option<FieldNormsSerializer>,
     postings_serializer: InvertedIndexSerializer,
 }
@@ -37,20 +35,11 @@ impl SegmentSerializer {
         let fieldnorms_write = segment.open_write(SegmentComponent::FieldNorms)?;
         let fieldnorms_serializer = FieldNormsSerializer::from_write(fieldnorms_write)?;
 
-        let spatial_serializer: Option<SpatialSerializer> =
-            if segment.schema().contains_spatial_field() {
-                let spatial_write = segment.open_write(SegmentComponent::Spatial)?;
-                Some(SpatialSerializer::from_write(spatial_write)?)
-            } else {
-                None
-            };
-
         let postings_serializer = InvertedIndexSerializer::open(&mut segment)?;
         Ok(SegmentSerializer {
             segment,
             store_writer,
             fast_field_write,
-            spatial_serializer,
             fieldnorms_serializer: Some(fieldnorms_serializer),
             postings_serializer,
         })
@@ -75,11 +64,6 @@ impl SegmentSerializer {
         &mut self.fast_field_write
     }
 
-    /// Accessor to the `SpatialSerializer`
-    pub fn extract_spatial_serializer(&mut self) -> Option<SpatialSerializer> {
-        self.spatial_serializer.take()
-    }
-
     /// Extract the field norm serializer.
     ///
     /// Note the fieldnorms serializer can only be extracted once.
@@ -97,9 +81,6 @@ impl SegmentSerializer {
         if let Some(fieldnorms_serializer) = self.extract_fieldnorms_serializer() {
             fieldnorms_serializer.close()?;
         }
-        if let Some(spatial_serializer) = self.extract_spatial_serializer() {
-            spatial_serializer.close()?;
-        }
         self.fast_field_write.terminate()?;
         self.postings_serializer.close()?;
         self.store_writer.close()?;

src/indexer/segment_writer.rs

@@ -16,7 +16,6 @@ use crate::postings::{
 };
 use crate::schema::document::{Document, Value};
 use crate::schema::{FieldEntry, FieldType, Schema, DATE_TIME_PRECISION_INDEXED};
-use crate::spatial::writer::SpatialWriter;
 use crate::tokenizer::{FacetTokenizer, PreTokenizedStream, TextAnalyzer, Tokenizer};
 use crate::{DocId, Opstamp, TantivyError};
 
@@ -53,7 +52,6 @@ pub struct SegmentWriter {
     pub(crate) segment_serializer: SegmentSerializer,
     pub(crate) fast_field_writers: FastFieldsWriter,
     pub(crate) fieldnorms_writer: FieldNormsWriter,
-    pub(crate) spatial_writer: SpatialWriter,
     pub(crate) json_path_writer: JsonPathWriter,
     pub(crate) json_positions_per_path: IndexingPositionsPerPath,
     pub(crate) doc_opstamps: Vec<Opstamp>,
@@ -106,7 +104,6 @@ impl SegmentWriter {
             ctx: IndexingContext::new(table_size),
             per_field_postings_writers,
             fieldnorms_writer: FieldNormsWriter::for_schema(&schema),
-            spatial_writer: SpatialWriter::default(),
             json_path_writer: JsonPathWriter::default(),
             json_positions_per_path: IndexingPositionsPerPath::default(),
             segment_serializer,
@@ -133,7 +130,6 @@ impl SegmentWriter {
             self.ctx,
             self.fast_field_writers,
             &self.fieldnorms_writer,
-            &mut self.spatial_writer,
             self.segment_serializer,
         )?;
         Ok(self.doc_opstamps)
@@ -146,7 +142,6 @@ impl SegmentWriter {
             + self.fieldnorms_writer.mem_usage()
             + self.fast_field_writers.mem_usage()
             + self.segment_serializer.mem_usage()
-            + self.spatial_writer.mem_usage()
     }
 
     fn index_document<D: Document>(&mut self, doc: &D) -> crate::Result<()> {
@@ -176,7 +171,7 @@ impl SegmentWriter {
             let (term_buffer, ctx) = (&mut self.term_buffer, &mut self.ctx);
             let postings_writer: &mut dyn PostingsWriter =
                 self.per_field_postings_writers.get_for_field_mut(field);
-            term_buffer.clear_with_field_and_type(field_entry.field_type().value_type(), field);
+            term_buffer.clear_with_field(field);
 
             match field_entry.field_type() {
                 FieldType::Facet(_) => {
@@ -343,13 +338,6 @@ impl SegmentWriter {
                         self.fieldnorms_writer.record(doc_id, field, num_vals);
                     }
                 }
-                FieldType::Spatial(_) => {
-                    for value in values {
-                        if let Some(geometry) = value.as_geometry() {
-                            self.spatial_writer.add_geometry(doc_id, field, *geometry);
-                        }
-                    }
-                }
             }
         }
         Ok(())
@@ -404,16 +392,12 @@ fn remap_and_write(
     ctx: IndexingContext,
     fast_field_writers: FastFieldsWriter,
     fieldnorms_writer: &FieldNormsWriter,
-    spatial_writer: &mut SpatialWriter,
     mut serializer: SegmentSerializer,
 ) -> crate::Result<()> {
     debug!("remap-and-write");
     if let Some(fieldnorms_serializer) = serializer.extract_fieldnorms_serializer() {
         fieldnorms_writer.serialize(fieldnorms_serializer)?;
     }
-    if let Some(spatial_serializer) = serializer.extract_spatial_serializer() {
-        spatial_writer.serialize(spatial_serializer)?;
-    }
     let fieldnorm_data = serializer
         .segment()
         .open_read(SegmentComponent::FieldNorms)?;
@@ -437,10 +421,9 @@ fn remap_and_write(
 #[cfg(test)]
 mod tests {
     use std::collections::BTreeMap;
-    use std::path::{Path, PathBuf};
+    use std::path::Path;
 
     use columnar::ColumnType;
-    use tempfile::TempDir;
 
     use crate::collector::{Count, TopDocs};
     use crate::directory::RamDirectory;
@@ -1083,10 +1066,7 @@ mod tests {
         let mut schema_builder = Schema::builder();
         schema_builder.add_text_field("title", text_options);
         let schema = schema_builder.build();
-        let tempdir = TempDir::new().unwrap();
-        let tempdir_path = PathBuf::from(tempdir.path());
-        Index::create_in_dir(&tempdir_path, schema).unwrap();
-        let index = Index::open_in_dir(tempdir_path).unwrap();
+        let index = Index::create_in_ram(schema);
         let schema = index.schema();
         let mut index_writer = index.writer(50_000_000).unwrap();
         let title = schema.get_field("title").unwrap();

src/lib.rs

@@ -17,6 +17,7 @@
 //!
 //! ```rust
 //! # use std::path::Path;
+//! # use std::fs;
 //! # use tempfile::TempDir;
 //! # use tantivy::collector::TopDocs;
 //! # use tantivy::query::QueryParser;
@@ -27,8 +28,11 @@
 //! #     // Let's create a temporary directory for the
 //! #     // sake of this example
 //! #     if let Ok(dir) = TempDir::new() {
-//! #         run_example(dir.path()).unwrap();
-//! #         dir.close().unwrap();
+//! #         let index_path = dir.path().join("index");
+//! #         // In case the directory already exists, we remove it
+//! #         let _ = fs::remove_dir_all(&index_path);
+//! #         fs::create_dir_all(&index_path).unwrap();
+//! #         run_example(&index_path).unwrap();
 //! #     }
 //! # }
 //! #
@@ -191,7 +195,6 @@ pub mod fieldnorm;
 pub mod index;
 pub mod positions;
 pub mod postings;
-pub mod spatial;
 
 /// Module containing the different query implementations.
 pub mod query;
@@ -204,6 +207,7 @@ mod docset;
 mod reader;
 
 #[cfg(test)]
+#[cfg(feature = "mmap")]
 mod compat_tests;
 
 pub use self::reader::{IndexReader, IndexReaderBuilder, ReloadPolicy, Warmer};
@@ -217,9 +221,7 @@ use once_cell::sync::Lazy;
 use serde::{Deserialize, Serialize};
 
 pub use self::docset::{DocSet, COLLECT_BLOCK_BUFFER_LEN, TERMINATED};
-#[doc(hidden)]
-pub use crate::core::json_utils;
-pub use crate::core::{Executor, Searcher, SearcherGeneration};
+pub use crate::core::{json_utils, Executor, Searcher, SearcherGeneration};
 pub use crate::directory::Directory;
 pub use crate::index::{
     Index, IndexBuilder, IndexMeta, IndexSettings, InvertedIndexReader, Order, Segment,
@@ -1173,12 +1175,11 @@ pub mod tests {
 
     #[test]
     fn test_validate_checksum() -> crate::Result<()> {
-        let index_path = tempfile::tempdir().expect("dir");
         let mut builder = Schema::builder();
         let body = builder.add_text_field("body", TEXT | STORED);
         let schema = builder.build();
-        let index = Index::create_in_dir(&index_path, schema)?;
-        let mut writer: IndexWriter = index.writer(50_000_000)?;
+        let index = Index::create_in_ram(schema);
+        let mut writer: IndexWriter = index.writer_for_tests()?;
         writer.set_merge_policy(Box::new(NoMergePolicy));
         for _ in 0..5000 {
             writer.add_document(doc!(body => "foo"))?;

src/postings/compression/mod.rs

@@ -1,12 +1,15 @@
 use bitpacking::{BitPacker, BitPacker4x};
-use common::FixedSize;
 
 pub const COMPRESSION_BLOCK_SIZE: usize = BitPacker4x::BLOCK_LEN;
-const COMPRESSED_BLOCK_MAX_SIZE: usize = COMPRESSION_BLOCK_SIZE * u32::SIZE_IN_BYTES;
+// in vint encoding, each byte stores 7 bits of data, so we need at most 32 / 7 = 4.57 bytes to
+// store a u32 in the worst case, rounding up to 5 bytes total
+const MAX_VINT_SIZE: usize = 5;
+const COMPRESSED_BLOCK_MAX_SIZE: usize = COMPRESSION_BLOCK_SIZE * MAX_VINT_SIZE;
 
 mod vint;
 
 /// Returns the size in bytes of a compressed block, given `num_bits`.
+#[inline]
 pub fn compressed_block_size(num_bits: u8) -> usize {
     (num_bits as usize) * COMPRESSION_BLOCK_SIZE / 8
 }
@@ -267,7 +270,6 @@ impl VIntDecoder for BlockDecoder {
 
 #[cfg(test)]
 pub(crate) mod tests {
-
     use super::*;
     use crate::TERMINATED;
 
@@ -372,6 +374,13 @@ pub(crate) mod tests {
             }
         }
     }
+
+    #[test]
+    fn test_compress_vint_unsorted_does_not_overflow() {
+        let mut encoder = BlockEncoder::new();
+        let input: Vec<u32> = vec![u32::MAX; COMPRESSION_BLOCK_SIZE];
+        encoder.compress_vint_unsorted(&input);
+    }
 }
 
 #[cfg(all(test, feature = "unstable"))]

src/postings/json_postings_writer.rs

@@ -8,7 +8,7 @@ use crate::indexer::path_to_unordered_id::OrderedPathId;
 use crate::postings::postings_writer::SpecializedPostingsWriter;
 use crate::postings::recorder::{BufferLender, DocIdRecorder, Recorder};
 use crate::postings::{FieldSerializer, IndexingContext, IndexingPosition, PostingsWriter};
-use crate::schema::{Field, Type, ValueBytes};
+use crate::schema::{Field, Type};
 use crate::tokenizer::TokenStream;
 use crate::DocId;
 
@@ -79,8 +79,7 @@ impl<Rec: Recorder> PostingsWriter for JsonPostingsWriter<Rec> {
             term_buffer.truncate(term_path_len);
             term_buffer.append_bytes(term);
 
-            let json_value = ValueBytes::wrap(term);
-            let typ = json_value.typ();
+            let typ = Type::from_code(term[0]).expect("Invalid type code in JSON term");
             if typ == Type::Str {
                 SpecializedPostingsWriter::<Rec>::serialize_one_term(
                     term_buffer.as_bytes(),
@@ -107,6 +106,8 @@ impl<Rec: Recorder> PostingsWriter for JsonPostingsWriter<Rec> {
     }
 }
 
+/// Helper to build the JSON term bytes that land in the term dictionary.
+/// Format: `[json path utf8][JSON_END_OF_PATH][type tag][payload]`
 struct JsonTermSerializer(Vec<u8>);
 impl JsonTermSerializer {
     /// Appends a JSON path to the Term.

src/postings/mod.rs

@@ -527,6 +527,7 @@ pub(crate) mod tests {
     }
 
     impl<TScorer: Scorer> Scorer for UnoptimizedDocSet<TScorer> {
+        #[inline]
         fn score(&mut self) -> Score {
             self.0.score()
         }

src/postings/per_field_postings_writer.rs

@@ -51,7 +51,6 @@ fn posting_writer_from_field_entry(field_entry: &FieldEntry) -> Box<dyn Postings
         | FieldType::Date(_)
         | FieldType::Bytes(_)
         | FieldType::IpAddr(_)
-        | FieldType::Spatial(_)
         | FieldType::Facet(_) => Box::<SpecializedPostingsWriter<DocIdRecorder>>::default(),
         FieldType::JsonObject(ref json_object_options) => {
             if let Some(text_indexing_option) = json_object_options.get_text_indexing_options() {

src/postings/postings_writer.rs

@@ -11,7 +11,7 @@ use crate::postings::recorder::{BufferLender, Recorder};
 use crate::postings::{
     FieldSerializer, IndexingContext, InvertedIndexSerializer, PerFieldPostingsWriter,
 };
-use crate::schema::{Field, Schema, Term, Type};
+use crate::schema::{Field, Schema, Type};
 use crate::tokenizer::{Token, TokenStream, MAX_TOKEN_LEN};
 use crate::DocId;
 
@@ -59,14 +59,14 @@ pub(crate) fn serialize_postings(
     let mut term_offsets: Vec<(Field, OrderedPathId, &[u8], Addr)> =
         Vec::with_capacity(ctx.term_index.len());
     term_offsets.extend(ctx.term_index.iter().map(|(key, addr)| {
-        let field = Term::wrap(key).field();
+        let field = IndexingTerm::wrap(key).field();
         if schema.get_field_entry(field).field_type().value_type() == Type::Json {
-            let byte_range_path = 5..5 + 4;
+            let byte_range_path = 4..4 + 4;
             let unordered_id = u32::from_be_bytes(key[byte_range_path.clone()].try_into().unwrap());
             let path_id = unordered_id_to_ordered_id[unordered_id as usize];
             (field, path_id, &key[byte_range_path.end..], addr)
         } else {
-            (field, 0.into(), &key[5..], addr)
+            (field, 0.into(), &key[4..], addr)
         }
     }));
     // Sort by field, path, and term

src/postings/skip.rs

@@ -6,17 +6,21 @@ use crate::{DocId, Score, TERMINATED};
 
 // doc num bits uses the following encoding:
 // given 0b a b cdefgh
-//         |1|2|   3  |
+//         |1|2|3|  4  |
 // - 1: unused
 // - 2: is delta-1 encoded. 0 if not, 1, if yes
-// - 3: a 6 bit number in 0..=32, the actual bitwidth
+// - 3: unused
+// - 4: a 5 bit number in 0..32, the actual bitwidth. Bitpacking could in theory say this is 32
+//   (requiring a 6th bit), but the biggest doc_id we can want to encode is TERMINATED-1, which can
+//   be represented on 31b without delta encoding.
 fn encode_bitwidth(bitwidth: u8, delta_1: bool) -> u8 {
+    assert!(bitwidth < 32);
     bitwidth | ((delta_1 as u8) << 6)
 }
 
 fn decode_bitwidth(raw_bitwidth: u8) -> (u8, bool) {
     let delta_1 = ((raw_bitwidth >> 6) & 1) != 0;
-    let bitwidth = raw_bitwidth & 0x3f;
+    let bitwidth = raw_bitwidth & 0x1f;
     (bitwidth, delta_1)
 }
 
@@ -430,7 +434,7 @@ mod tests {
 
     #[test]
     fn test_encode_decode_bitwidth() {
-        for bitwidth in 0..=32 {
+        for bitwidth in 0..32 {
             for delta_1 in [false, true] {
                 assert_eq!(
                     (bitwidth, delta_1),

src/query/all_query.rs

@@ -23,7 +23,11 @@ pub struct AllWeight;
 impl Weight for AllWeight {
     fn scorer(&self, reader: &SegmentReader, boost: Score) -> crate::Result<Box<dyn Scorer>> {
         let all_scorer = AllScorer::new(reader.max_doc());
-        Ok(Box::new(BoostScorer::new(all_scorer, boost)))
+        if boost != 1.0 {
+            Ok(Box::new(BoostScorer::new(all_scorer, boost)))
+        } else {
+            Ok(Box::new(all_scorer))
+        }
     }
 
     fn explain(&self, reader: &SegmentReader, doc: DocId) -> crate::Result<Explanation> {
@@ -58,6 +62,15 @@ impl DocSet for AllScorer {
         self.doc
     }
 
+    fn seek(&mut self, target: DocId) -> DocId {
+        debug_assert!(target >= self.doc);
+        self.doc = target;
+        if self.doc >= self.max_doc {
+            self.doc = TERMINATED;
+        }
+        self.doc
+    }
+
     fn fill_buffer(&mut self, buffer: &mut [DocId; COLLECT_BLOCK_BUFFER_LEN]) -> usize {
         if self.doc() == TERMINATED {
             return 0;
@@ -92,6 +105,7 @@ impl DocSet for AllScorer {
 }
 
 impl Scorer for AllScorer {
+    #[inline]
     fn score(&mut self) -> Score {
         1.0
     }

src/query/boolean_query/block_wand.rs

@@ -483,7 +483,7 @@ mod tests {
             let checkpoints_for_each_pruning =
                 compute_checkpoints_for_each_pruning(term_scorers.clone(), top_k);
             let checkpoints_manual =
-                compute_checkpoints_manual(term_scorers.clone(), top_k, 100_000);
+                compute_checkpoints_manual(term_scorers.clone(), top_k, max_doc as u32);
             assert_eq!(checkpoints_for_each_pruning.len(), checkpoints_manual.len());
             for (&(left_doc, left_score), &(right_doc, right_score)) in checkpoints_for_each_pruning
                 .iter()

src/query/boolean_query/boolean_weight.rs

@@ -97,6 +97,65 @@ fn into_box_scorer<TScoreCombiner: ScoreCombiner>(
     }
 }
 
+/// Returns the effective MUST scorer, accounting for removed AllScorers.
+///
+/// When AllScorer instances are removed from must_scorers as an optimization,
+/// we must restore the "match all" semantics if the list becomes empty.
+fn effective_must_scorer(
+    must_scorers: Vec<Box<dyn Scorer>>,
+    removed_all_scorer_count: usize,
+    max_doc: DocId,
+    num_docs: u32,
+) -> Option<Box<dyn Scorer>> {
+    if must_scorers.is_empty() {
+        if removed_all_scorer_count > 0 {
+            // Had AllScorer(s) only - all docs match
+            Some(Box::new(AllScorer::new(max_doc)))
+        } else {
+            // No MUST constraint at all
+            None
+        }
+    } else {
+        Some(intersect_scorers(must_scorers, num_docs))
+    }
+}
+
+/// Returns a SHOULD scorer with AllScorer union if any were removed.
+///
+/// For union semantics (OR): if any SHOULD clause was an AllScorer, the result
+/// should include all documents. We restore this by unioning with AllScorer.
+///
+/// When `scoring_enabled` is false, we can just return AllScorer alone since
+/// we don't need score contributions from the should_scorer.
+fn effective_should_scorer_for_union<TScoreCombiner: ScoreCombiner>(
+    should_scorer: SpecializedScorer,
+    removed_all_scorer_count: usize,
+    max_doc: DocId,
+    num_docs: u32,
+    score_combiner_fn: impl Fn() -> TScoreCombiner,
+    scoring_enabled: bool,
+) -> SpecializedScorer {
+    if removed_all_scorer_count > 0 {
+        if scoring_enabled {
+            // Need to union to get score contributions from both
+            let all_scorers: Vec<Box<dyn Scorer>> = vec![
+                into_box_scorer(should_scorer, &score_combiner_fn, num_docs),
+                Box::new(AllScorer::new(max_doc)),
+            ];
+            SpecializedScorer::Other(Box::new(BufferedUnionScorer::build(
+                all_scorers,
+                score_combiner_fn,
+                num_docs,
+            )))
+        } else {
+            // Scoring disabled - AllScorer alone is sufficient
+            SpecializedScorer::Other(Box::new(AllScorer::new(max_doc)))
+        }
+    } else {
+        should_scorer
+    }
+}
+
 enum ShouldScorersCombinationMethod {
     // Should scorers are irrelevant.
     Ignored,
@@ -193,18 +252,18 @@ impl<TScoreCombiner: ScoreCombiner> BooleanWeight<TScoreCombiner> {
             return Ok(SpecializedScorer::Other(Box::new(EmptyScorer)));
         }
 
-        let minimum_number_should_match = self
+        let effective_minimum_number_should_match = self
             .minimum_number_should_match
             .saturating_sub(should_special_scorer_counts.num_all_scorers);
 
         let should_scorers: ShouldScorersCombinationMethod = {
             let num_of_should_scorers = should_scorers.len();
-            if minimum_number_should_match > num_of_should_scorers {
+            if effective_minimum_number_should_match > num_of_should_scorers {
                 // We don't have enough scorers to satisfy the minimum number of should matches.
                 // The request will match no documents.
                 return Ok(SpecializedScorer::Other(Box::new(EmptyScorer)));
             }
-            match minimum_number_should_match {
+            match effective_minimum_number_should_match {
                 0 if num_of_should_scorers == 0 => ShouldScorersCombinationMethod::Ignored,
                 0 => ShouldScorersCombinationMethod::Optional(scorer_union(
                     should_scorers,
@@ -226,7 +285,7 @@ impl<TScoreCombiner: ScoreCombiner> BooleanWeight<TScoreCombiner> {
                     scorer_disjunction(
                         should_scorers,
                         score_combiner_fn(),
-                        self.minimum_number_should_match,
+                        effective_minimum_number_should_match,
                     ),
                 )),
             }
@@ -246,53 +305,78 @@ impl<TScoreCombiner: ScoreCombiner> BooleanWeight<TScoreCombiner> {
 
         let include_scorer = match (should_scorers, must_scorers) {
             (ShouldScorersCombinationMethod::Ignored, must_scorers) => {
-                let boxed_scorer: Box<dyn Scorer> = if must_scorers.is_empty() {
-                    // We do not have any should scorers, nor all scorers.
-                    // There are still two cases here.
-                    //
-                    // If this follows the removal of some AllScorers in the should/must clauses,
-                    // then we match all documents.
-                    //
-                    // Otherwise, it is really just an EmptyScorer.
-                    if must_special_scorer_counts.num_all_scorers
-                        + should_special_scorer_counts.num_all_scorers
-                        > 0
-                    {
-                        Box::new(AllScorer::new(reader.max_doc()))
-                    } else {
-                        Box::new(EmptyScorer)
-                    }
-                } else {
-                    intersect_scorers(must_scorers, num_docs)
-                };
+                // No SHOULD clauses (or they were absorbed into MUST).
+                // Result depends entirely on MUST + any removed AllScorers.
+                let combined_all_scorer_count = must_special_scorer_counts.num_all_scorers
+                    + should_special_scorer_counts.num_all_scorers;
+                let boxed_scorer: Box<dyn Scorer> = effective_must_scorer(
+                    must_scorers,
+                    combined_all_scorer_count,
+                    reader.max_doc(),
+                    num_docs,
+                )
+                .unwrap_or_else(|| Box::new(EmptyScorer));
                 SpecializedScorer::Other(boxed_scorer)
             }
             (ShouldScorersCombinationMethod::Optional(should_scorer), must_scorers) => {
-                if must_scorers.is_empty() && must_special_scorer_counts.num_all_scorers == 0 {
-                    // Optional options are promoted to required if no must scorers exists.
-                    should_scorer
-                } else {
-                    let must_scorer = intersect_scorers(must_scorers, num_docs);
-                    if self.scoring_enabled {
-                        SpecializedScorer::Other(Box::new(RequiredOptionalScorer::<
-                            _,
-                            _,
-                            TScoreCombiner,
-                        >::new(
-                            must_scorer,
-                            into_box_scorer(should_scorer, &score_combiner_fn, num_docs),
-                        )))
-                    } else {
-                        SpecializedScorer::Other(must_scorer)
+                // Optional SHOULD: contributes to scoring but not required for matching.
+                match effective_must_scorer(
+                    must_scorers,
+                    must_special_scorer_counts.num_all_scorers,
+                    reader.max_doc(),
+                    num_docs,
+                ) {
+                    None => {
+                        // No MUST constraint: promote SHOULD to required.
+                        // Must preserve any removed AllScorers from SHOULD via union.
+                        effective_should_scorer_for_union(
+                            should_scorer,
+                            should_special_scorer_counts.num_all_scorers,
+                            reader.max_doc(),
+                            num_docs,
+                            &score_combiner_fn,
+                            self.scoring_enabled,
+                        )
+                    }
+                    Some(must_scorer) => {
+                        // Has MUST constraint: SHOULD only affects scoring.
+                        if self.scoring_enabled {
+                            SpecializedScorer::Other(Box::new(RequiredOptionalScorer::<
+                                _,
+                                _,
+                                TScoreCombiner,
+                            >::new(
+                                must_scorer,
+                                into_box_scorer(should_scorer, &score_combiner_fn, num_docs),
+                            )))
+                        } else {
+                            SpecializedScorer::Other(must_scorer)
+                        }
                     }
                 }
             }
-            (ShouldScorersCombinationMethod::Required(should_scorer), mut must_scorers) => {
-                if must_scorers.is_empty() {
-                    should_scorer
-                } else {
-                    must_scorers.push(into_box_scorer(should_scorer, &score_combiner_fn, num_docs));
-                    SpecializedScorer::Other(intersect_scorers(must_scorers, num_docs))
+            (ShouldScorersCombinationMethod::Required(should_scorer), must_scorers) => {
+                // Required SHOULD: at least `minimum_number_should_match` must match.
+                // Semantics: (MUST constraint) AND (SHOULD constraint)
+                match effective_must_scorer(
+                    must_scorers,
+                    must_special_scorer_counts.num_all_scorers,
+                    reader.max_doc(),
+                    num_docs,
+                ) {
+                    None => {
+                        // No MUST constraint: SHOULD alone determines matching.
+                        should_scorer
+                    }
+                    Some(must_scorer) => {
+                        // Has MUST constraint: intersect MUST with SHOULD.
+                        let should_boxed =
+                            into_box_scorer(should_scorer, &score_combiner_fn, num_docs);
+                        SpecializedScorer::Other(intersect_scorers(
+                            vec![must_scorer, should_boxed],
+                            num_docs,
+                        ))
+                    }
                 }
             }
         };

src/query/boolean_query/mod.rs

@@ -9,12 +9,14 @@ pub use self::boolean_weight::BooleanWeight;
 #[cfg(test)]
 mod tests {
 
+    use std::ops::Bound;
+
     use super::*;
     use crate::collector::tests::TEST_COLLECTOR_WITH_SCORE;
-    use crate::collector::TopDocs;
+    use crate::collector::{Count, TopDocs};
     use crate::query::term_query::TermScorer;
     use crate::query::{
-        AllScorer, EmptyScorer, EnableScoring, Intersection, Occur, Query, QueryParser,
+        AllScorer, EmptyScorer, EnableScoring, Intersection, Occur, Query, QueryParser, RangeQuery,
         RequiredOptionalScorer, Scorer, SumCombiner, TermQuery,
     };
     use crate::schema::*;
@@ -374,4 +376,466 @@ mod tests {
         }
         Ok(())
     }
+
+    #[test]
+    pub fn test_min_should_match_with_all_query() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let text_field = schema_builder.add_text_field("text", TEXT);
+        let num_field =
+            schema_builder.add_i64_field("num", NumericOptions::default().set_fast().set_indexed());
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
+
+        index_writer.add_document(doc!(text_field => "apple", num_field => 10i64))?;
+        index_writer.add_document(doc!(text_field => "banana", num_field => 20i64))?;
+        index_writer.commit()?;
+
+        let searcher = index.reader()?.searcher();
+
+        let effective_all_match_query: Box<dyn Query> = Box::new(RangeQuery::new(
+            Bound::Excluded(Term::from_field_i64(num_field, 0)),
+            Bound::Unbounded,
+        ));
+        let term_query: Box<dyn Query> = Box::new(TermQuery::new(
+            Term::from_field_text(text_field, "apple"),
+            IndexRecordOption::Basic,
+        ));
+
+        // in some previous version, we would remove the 2 all_match, but then say we need *4*
+        // matches out of the 3 term queries, which matches nothing.
+        let mut bool_query = BooleanQuery::new(vec![
+            (Occur::Should, effective_all_match_query.box_clone()),
+            (Occur::Should, effective_all_match_query.box_clone()),
+            (Occur::Should, term_query.box_clone()),
+            (Occur::Should, term_query.box_clone()),
+            (Occur::Should, term_query.box_clone()),
+        ]);
+        bool_query.set_minimum_number_should_match(4);
+        let count = searcher.search(&bool_query, &Count)?;
+        assert_eq!(count, 1);
+
+        Ok(())
+    }
+
+    // =========================================================================
+    // AllScorer Preservation Regression Tests
+    // =========================================================================
+    //
+    // These tests verify the fix for a bug where AllScorer instances (produced by
+    // queries matching all documents, such as range queries covering all values)
+    // were incorrectly removed from Boolean query processing, causing documents
+    // to be unexpectedly excluded from results.
+    //
+    // The bug manifested in several scenarios:
+    // 1. SHOULD + SHOULD where one clause is AllScorer
+    // 2. MUST (AllScorer) + SHOULD
+    // 3. Range queries in Boolean clauses when all documents match the range
+
+    /// Regression test: SHOULD clause with AllScorer combined with other SHOULD clauses.
+    ///
+    /// When a SHOULD clause produces an AllScorer (e.g., from a range query matching
+    /// all documents), the Boolean query should still match all documents.
+    ///
+    /// Bug before fix: AllScorer was removed during optimization, leaving only the
+    /// other SHOULD clauses, which incorrectly excluded documents.
+    #[test]
+    pub fn test_should_with_all_scorer_regression() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let text_field = schema_builder.add_text_field("text", TEXT);
+        let num_field =
+            schema_builder.add_i64_field("num", NumericOptions::default().set_fast().set_indexed());
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
+
+        // All docs have num > 0, so range query will return AllScorer
+        index_writer.add_document(doc!(text_field => "hello", num_field => 10i64))?;
+        index_writer.add_document(doc!(text_field => "world", num_field => 20i64))?;
+        index_writer.add_document(doc!(text_field => "hello world", num_field => 30i64))?;
+        index_writer.add_document(doc!(text_field => "foo", num_field => 40i64))?;
+        index_writer.add_document(doc!(text_field => "bar", num_field => 50i64))?;
+        index_writer.add_document(doc!(text_field => "baz", num_field => 60i64))?;
+        index_writer.commit()?;
+
+        let searcher = index.reader()?.searcher();
+
+        // Range query matching all docs (returns AllScorer)
+        let all_match_query: Box<dyn Query> = Box::new(RangeQuery::new(
+            Bound::Excluded(Term::from_field_i64(num_field, 0)),
+            Bound::Unbounded,
+        ));
+        let term_query: Box<dyn Query> = Box::new(TermQuery::new(
+            Term::from_field_text(text_field, "hello"),
+            IndexRecordOption::Basic,
+        ));
+
+        // Verify range matches all 6 docs
+        assert_eq!(searcher.search(all_match_query.as_ref(), &Count)?, 6);
+
+        // RangeQuery(all) OR TermQuery should match all 6 docs
+        let bool_query = BooleanQuery::new(vec![
+            (Occur::Should, all_match_query.box_clone()),
+            (Occur::Should, term_query.box_clone()),
+        ]);
+        let count = searcher.search(&bool_query, &Count)?;
+        assert_eq!(count, 6, "SHOULD with AllScorer should match all docs");
+
+        // Order should not matter
+        let bool_query_reversed = BooleanQuery::new(vec![
+            (Occur::Should, term_query.box_clone()),
+            (Occur::Should, all_match_query.box_clone()),
+        ]);
+        let count_reversed = searcher.search(&bool_query_reversed, &Count)?;
+        assert_eq!(
+            count_reversed, 6,
+            "Order of SHOULD clauses should not matter"
+        );
+
+        Ok(())
+    }
+
+    /// Regression test: MUST clause with AllScorer combined with SHOULD clause.
+    ///
+    /// When MUST contains an AllScorer, all documents satisfy the MUST constraint.
+    /// The SHOULD clause should only affect scoring, not filtering.
+    ///
+    /// Bug before fix: AllScorer was removed, leaving an empty must_scorers vector.
+    /// intersect_scorers([]) incorrectly returned EmptyScorer, matching 0 documents.
+    #[test]
+    pub fn test_must_all_with_should_regression() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let text_field = schema_builder.add_text_field("text", TEXT);
+        let num_field =
+            schema_builder.add_i64_field("num", NumericOptions::default().set_fast().set_indexed());
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
+
+        // All docs have num > 0, so range query will return AllScorer
+        index_writer.add_document(doc!(text_field => "apple", num_field => 10i64))?;
+        index_writer.add_document(doc!(text_field => "banana", num_field => 20i64))?;
+        index_writer.add_document(doc!(text_field => "cherry", num_field => 30i64))?;
+        index_writer.add_document(doc!(text_field => "date", num_field => 40i64))?;
+        index_writer.commit()?;
+
+        let searcher = index.reader()?.searcher();
+
+        // Range query matching all docs (returns AllScorer)
+        let all_match_query: Box<dyn Query> = Box::new(RangeQuery::new(
+            Bound::Excluded(Term::from_field_i64(num_field, 0)),
+            Bound::Unbounded,
+        ));
+        let term_query: Box<dyn Query> = Box::new(TermQuery::new(
+            Term::from_field_text(text_field, "apple"),
+            IndexRecordOption::Basic,
+        ));
+
+        // Verify range matches all 4 docs
+        assert_eq!(searcher.search(all_match_query.as_ref(), &Count)?, 4);
+
+        // MUST(range matching all) AND SHOULD(term) should match all 4 docs
+        let bool_query = BooleanQuery::new(vec![
+            (Occur::Must, all_match_query.box_clone()),
+            (Occur::Should, term_query.box_clone()),
+        ]);
+        let count = searcher.search(&bool_query, &Count)?;
+        assert_eq!(count, 4, "MUST AllScorer + SHOULD should match all docs");
+
+        Ok(())
+    }
+
+    /// Regression test: Range queries in Boolean clauses when all documents match.
+    ///
+    /// Range queries can return AllScorer as an optimization when all indexed values
+    /// fall within the range. This test ensures such queries work correctly in
+    /// Boolean combinations.
+    ///
+    /// This is the most common real-world manifestation of the bug, occurring in
+    /// queries like: (age > 50 OR name = 'Alice') AND status = 'active'
+    /// when all documents have age > 50.
+    #[test]
+    pub fn test_range_query_all_match_in_boolean() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let name_field = schema_builder.add_text_field("name", TEXT);
+        let age_field =
+            schema_builder.add_i64_field("age", NumericOptions::default().set_fast().set_indexed());
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
+
+        // All documents have age > 50, so range query will return AllScorer
+        index_writer.add_document(doc!(name_field => "alice", age_field => 55_i64))?;
+        index_writer.add_document(doc!(name_field => "bob", age_field => 60_i64))?;
+        index_writer.add_document(doc!(name_field => "charlie", age_field => 70_i64))?;
+        index_writer.add_document(doc!(name_field => "diana", age_field => 80_i64))?;
+        index_writer.commit()?;
+
+        let searcher = index.reader()?.searcher();
+
+        let range_query: Box<dyn Query> = Box::new(RangeQuery::new(
+            Bound::Excluded(Term::from_field_i64(age_field, 50)),
+            Bound::Unbounded,
+        ));
+        let term_query: Box<dyn Query> = Box::new(TermQuery::new(
+            Term::from_field_text(name_field, "alice"),
+            IndexRecordOption::Basic,
+        ));
+
+        // Verify preconditions
+        assert_eq!(searcher.search(range_query.as_ref(), &Count)?, 4);
+        assert_eq!(searcher.search(term_query.as_ref(), &Count)?, 1);
+
+        // SHOULD(range) OR SHOULD(term): range matches all, so result is 4
+        let should_query = BooleanQuery::new(vec![
+            (Occur::Should, range_query.box_clone()),
+            (Occur::Should, term_query.box_clone()),
+        ]);
+        assert_eq!(
+            searcher.search(&should_query, &Count)?,
+            4,
+            "SHOULD range OR term should match all"
+        );
+
+        // MUST(range) AND SHOULD(term): range matches all, term is optional
+        let must_should_query = BooleanQuery::new(vec![
+            (Occur::Must, range_query.box_clone()),
+            (Occur::Should, term_query.box_clone()),
+        ]);
+        assert_eq!(
+            searcher.search(&must_should_query, &Count)?,
+            4,
+            "MUST range + SHOULD term should match all"
+        );
+
+        Ok(())
+    }
+
+    /// Test multiple AllScorer instances in different clause types.
+    ///
+    /// Verifies correct behavior when AllScorers appear in multiple positions.
+    #[test]
+    pub fn test_multiple_all_scorers() -> crate::Result<()> {
+        let mut schema_builder = Schema::builder();
+        let text_field = schema_builder.add_text_field("text", TEXT);
+        let num_field =
+            schema_builder.add_i64_field("num", NumericOptions::default().set_fast().set_indexed());
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer: IndexWriter = index.writer_for_tests()?;
+
+        // All docs have num > 0, so range queries will return AllScorer
+        index_writer.add_document(doc!(text_field => "doc1", num_field => 10i64))?;
+        index_writer.add_document(doc!(text_field => "doc2", num_field => 20i64))?;
+        index_writer.add_document(doc!(text_field => "doc3", num_field => 30i64))?;
+        index_writer.commit()?;
+
+        let searcher = index.reader()?.searcher();
+
+        // Two different range queries that both match all docs (return AllScorer)
+        let all_query1: Box<dyn Query> = Box::new(RangeQuery::new(
+            Bound::Excluded(Term::from_field_i64(num_field, 0)),
+            Bound::Unbounded,
+        ));
+        let all_query2: Box<dyn Query> = Box::new(RangeQuery::new(
+            Bound::Excluded(Term::from_field_i64(num_field, 5)),
+            Bound::Unbounded,
+        ));
+        let term_query: Box<dyn Query> = Box::new(TermQuery::new(
+            Term::from_field_text(text_field, "doc1"),
+            IndexRecordOption::Basic,
+        ));
+
+        // Multiple AllScorers in SHOULD
+        let multi_all_should = BooleanQuery::new(vec![
+            (Occur::Should, all_query1.box_clone()),
+            (Occur::Should, all_query2.box_clone()),
+            (Occur::Should, term_query.box_clone()),
+        ]);
+        assert_eq!(
+            searcher.search(&multi_all_should, &Count)?,
+            3,
+            "Multiple AllScorers in SHOULD"
+        );
+
+        // AllScorer in both MUST and SHOULD
+        let all_must_and_should = BooleanQuery::new(vec![
+            (Occur::Must, all_query1.box_clone()),
+            (Occur::Should, all_query2.box_clone()),
+        ]);
+        assert_eq!(
+            searcher.search(&all_must_and_should, &Count)?,
+            3,
+            "AllScorer in both MUST and SHOULD"
+        );
+
+        Ok(())
+    }
+}
+
+/// A proptest which generates arbitrary permutations of a simple boolean AST, and then matches
+/// the result against an index which contains all permutations of documents with N fields.
+#[cfg(test)]
+mod proptest_boolean_query {
+    use std::collections::{BTreeMap, HashSet};
+    use std::ops::{Bound, Range};
+
+    use proptest::collection::vec;
+    use proptest::prelude::*;
+
+    use crate::collector::DocSetCollector;
+    use crate::query::{AllQuery, BooleanQuery, Occur, Query, RangeQuery, TermQuery};
+    use crate::schema::{Field, NumericOptions, OwnedValue, Schema, TEXT};
+    use crate::{DocId, Index, Term};
+
+    #[derive(Debug, Clone)]
+    enum BooleanQueryAST {
+        /// Matches all documents via AllQuery (wraps AllScorer in BoostScorer)
+        All,
+        /// Matches all documents via RangeQuery (returns bare AllScorer)
+        /// This is the actual trigger for the AllScorer preservation bug
+        RangeAll,
+        /// Matches documents where the field has value "true"
+        Leaf {
+            field_idx: usize,
+        },
+        Union(Vec<BooleanQueryAST>),
+        Intersection(Vec<BooleanQueryAST>),
+    }
+
+    impl BooleanQueryAST {
+        fn matches(&self, doc_id: DocId) -> bool {
+            match self {
+                BooleanQueryAST::All => true,
+                BooleanQueryAST::RangeAll => true,
+                BooleanQueryAST::Leaf { field_idx } => Self::matches_field(doc_id, *field_idx),
+                BooleanQueryAST::Union(children) => {
+                    children.iter().any(|child| child.matches(doc_id))
+                }
+                BooleanQueryAST::Intersection(children) => {
+                    children.iter().all(|child| child.matches(doc_id))
+                }
+            }
+        }
+
+        fn matches_field(doc_id: DocId, field_idx: usize) -> bool {
+            ((doc_id as usize) >> field_idx) & 1 == 1
+        }
+
+        fn to_query(&self, fields: &[Field], range_field: Field) -> Box<dyn Query> {
+            match self {
+                BooleanQueryAST::All => Box::new(AllQuery),
+                BooleanQueryAST::RangeAll => {
+                    // Range query that matches all docs (all have value >= 0)
+                    // This returns bare AllScorer, triggering the bug we fixed
+                    Box::new(RangeQuery::new(
+                        Bound::Included(Term::from_field_i64(range_field, 0)),
+                        Bound::Unbounded,
+                    ))
+                }
+                BooleanQueryAST::Leaf { field_idx } => Box::new(TermQuery::new(
+                    Term::from_field_text(fields[*field_idx], "true"),
+                    crate::schema::IndexRecordOption::Basic,
+                )),
+                BooleanQueryAST::Union(children) => {
+                    let sub_queries = children
+                        .iter()
+                        .map(|child| (Occur::Should, child.to_query(fields, range_field)))
+                        .collect();
+                    Box::new(BooleanQuery::new(sub_queries))
+                }
+                BooleanQueryAST::Intersection(children) => {
+                    let sub_queries = children
+                        .iter()
+                        .map(|child| (Occur::Must, child.to_query(fields, range_field)))
+                        .collect();
+                    Box::new(BooleanQuery::new(sub_queries))
+                }
+            }
+        }
+    }
+
+    fn doc_ids(num_docs: usize, num_fields: usize) -> Range<DocId> {
+        let permutations = 1 << num_fields;
+        let copies = (num_docs as f32 / permutations as f32).ceil() as u32;
+        0..(permutations * copies)
+    }
+
+    fn create_index_with_boolean_permutations(
+        num_docs: usize,
+        num_fields: usize,
+    ) -> (Index, Vec<Field>, Field) {
+        let mut schema_builder = Schema::builder();
+        let fields: Vec<Field> = (0..num_fields)
+            .map(|i| schema_builder.add_text_field(&format!("field_{}", i), TEXT))
+            .collect();
+        // Add a numeric field for RangeQuery tests - all docs have value = doc_id
+        let range_field = schema_builder.add_i64_field(
+            "range_field",
+            NumericOptions::default().set_fast().set_indexed(),
+        );
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut writer = index.writer_for_tests().unwrap();
+
+        for doc_id in doc_ids(num_docs, num_fields) {
+            let mut doc: BTreeMap<_, OwnedValue> = BTreeMap::default();
+            for (field_idx, &field) in fields.iter().enumerate() {
+                if (doc_id >> field_idx) & 1 == 1 {
+                    doc.insert(field, "true".into());
+                }
+            }
+            // All docs have non-negative values, so RangeQuery(>=0) matches all
+            doc.insert(range_field, (doc_id as i64).into());
+            writer.add_document(doc).unwrap();
+        }
+        writer.commit().unwrap();
+        (index, fields, range_field)
+    }
+
+    fn arb_boolean_query_ast(num_fields: usize) -> impl Strategy<Value = BooleanQueryAST> {
+        // Leaf strategies: term queries, AllQuery, and RangeQuery matching all docs
+        let leaf = prop_oneof![
+            (0..num_fields).prop_map(|field_idx| BooleanQueryAST::Leaf { field_idx }),
+            Just(BooleanQueryAST::All),
+            Just(BooleanQueryAST::RangeAll),
+        ];
+        leaf.prop_recursive(
+            8,   // 8 levels of recursion
+            256, // 256 nodes max
+            10,  // 10 items per collection
+            |inner| {
+                prop_oneof![
+                    vec(inner.clone(), 1..10).prop_map(BooleanQueryAST::Union),
+                    vec(inner, 1..10).prop_map(BooleanQueryAST::Intersection),
+                ]
+            },
+        )
+    }
+
+    #[test]
+    fn proptest_boolean_query() {
+        // In the presence of optimizations around buffering, it can take large numbers of
+        // documents to uncover some issues.
+        let num_fields = 8;
+        let num_docs = 1 << num_fields;
+        let (index, fields, range_field) =
+            create_index_with_boolean_permutations(num_docs, num_fields);
+        let searcher = index.reader().unwrap().searcher();
+        proptest!(|(ast in arb_boolean_query_ast(num_fields))| {
+            let query = ast.to_query(&fields, range_field);
+
+            let mut matching_docs = HashSet::new();
+            for doc_id in doc_ids(num_docs, num_fields) {
+                if ast.matches(doc_id as DocId) {
+                    matching_docs.insert(doc_id as DocId);
+                }
+            }
+
+            let doc_addresses = searcher.search(&*query, &DocSetCollector).unwrap();
+            let result_docs: HashSet<DocId> =
+                doc_addresses.into_iter().map(|doc_address| doc_address.doc_id).collect();
+            prop_assert_eq!(result_docs, matching_docs);
+        });
+    }
 }

src/query/boost_query.rs

@@ -104,6 +104,9 @@ impl<S: Scorer> DocSet for BoostScorer<S> {
     fn seek(&mut self, target: DocId) -> DocId {
         self.underlying.seek(target)
     }
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        self.underlying.seek_into_the_danger_zone(target)
+    }
 
     fn fill_buffer(&mut self, buffer: &mut [DocId; COLLECT_BLOCK_BUFFER_LEN]) -> usize {
         self.underlying.fill_buffer(buffer)
@@ -131,6 +134,7 @@ impl<S: Scorer> DocSet for BoostScorer<S> {
 }
 
 impl<S: Scorer> Scorer for BoostScorer<S> {
+    #[inline]
     fn score(&mut self) -> Score {
         self.underlying.score() * self.boost
     }

src/query/const_score_query.rs

@@ -137,6 +137,7 @@ impl<TDocSet: DocSet> DocSet for ConstScorer<TDocSet> {
 }
 
 impl<TDocSet: DocSet + 'static> Scorer for ConstScorer<TDocSet> {
+    #[inline]
     fn score(&mut self) -> Score {
         self.score
     }

src/query/disjunction.rs

@@ -62,6 +62,16 @@ impl<T: Scorer> DocSet for ScorerWrapper<T> {
         self.current_doc = doc_id;
         doc_id
     }
+    fn seek(&mut self, target: DocId) -> DocId {
+        let doc_id = self.scorer.seek(target);
+        self.current_doc = doc_id;
+        doc_id
+    }
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        let found = self.scorer.seek_into_the_danger_zone(target);
+        self.current_doc = self.scorer.doc();
+        found
+    }
 
     fn doc(&self) -> DocId {
         self.current_doc
@@ -163,6 +173,7 @@ impl<TScorer: Scorer, TScoreCombiner: ScoreCombiner> DocSet
 impl<TScorer: Scorer, TScoreCombiner: ScoreCombiner> Scorer
     for Disjunction<TScorer, TScoreCombiner>
 {
+    #[inline]
     fn score(&mut self) -> Score {
         self.current_score
     }
@@ -297,6 +308,7 @@ mod tests {
     }
 
     impl Scorer for DummyScorer {
+        #[inline]
         fn score(&mut self) -> Score {
             self.foo.get(self.cursor).map(|x| x.1).unwrap_or(0.0)
         }

src/query/empty_query.rs

@@ -55,6 +55,7 @@ impl DocSet for EmptyScorer {
 }
 
 impl Scorer for EmptyScorer {
+    #[inline]
     fn score(&mut self) -> Score {
         0.0
     }

src/query/exclude.rs

@@ -84,6 +84,7 @@ where
     TScorer: Scorer,
     TDocSetExclude: DocSet + 'static,
 {
+    #[inline]
     fn score(&mut self) -> Score {
         self.underlying_docset.score()
     }

src/query/intersection.rs

@@ -1,5 +1,5 @@
+use super::size_hint::estimate_intersection;
 use crate::docset::{DocSet, TERMINATED};
-use crate::query::size_hint::estimate_intersection;
 use crate::query::term_query::TermScorer;
 use crate::query::{EmptyScorer, Scorer};
 use crate::{DocId, Score};
@@ -12,6 +12,9 @@ use crate::{DocId, Score};
 /// For better performance, the function uses a
 /// specialized implementation if the two
 /// shortest scorers are `TermScorer`s.
+///
+/// num_docs_segment is the number of documents in the segment. It is used for estimating the
+/// `size_hint` of the intersection.
 pub fn intersect_scorers(
     mut scorers: Vec<Box<dyn Scorer>>,
     num_docs_segment: u32,
@@ -102,35 +105,48 @@ impl<TDocSet: DocSet> Intersection<TDocSet, TDocSet> {
 }
 
 impl<TDocSet: DocSet, TOtherDocSet: DocSet> DocSet for Intersection<TDocSet, TOtherDocSet> {
+    #[inline]
     fn advance(&mut self) -> DocId {
         let (left, right) = (&mut self.left, &mut self.right);
         let mut candidate = left.advance();
+        if candidate == TERMINATED {
+            return TERMINATED;
+        }
 
-        'outer: loop {
+        loop {
             // In the first part we look for a document in the intersection
             // of the two rarest `DocSet` in the intersection.
 
             loop {
-                let right_doc = right.seek(candidate);
-                candidate = left.seek(right_doc);
-                if candidate == right_doc {
+                if right.seek_into_the_danger_zone(candidate) {
                     break;
                 }
+                let right_doc = right.doc();
+                // TODO: Think about which value would make sense here
+                // It depends on the DocSet implementation, when a seek would outweigh an advance.
+                if right_doc > candidate.wrapping_add(100) {
+                    candidate = left.seek(right_doc);
+                } else {
+                    candidate = left.advance();
+                }
+                if candidate == TERMINATED {
+                    return TERMINATED;
+                }
             }
 
             debug_assert_eq!(left.doc(), right.doc());
-            // test the remaining scorers;
-            for docset in self.others.iter_mut() {
-                let seek_doc = docset.seek(candidate);
-                if seek_doc > candidate {
-                    candidate = left.seek(seek_doc);
-                    continue 'outer;
-                }
+            // test the remaining scorers
+            if self
+                .others
+                .iter_mut()
+                .all(|docset| docset.seek_into_the_danger_zone(candidate))
+            {
+                debug_assert_eq!(candidate, self.left.doc());
+                debug_assert_eq!(candidate, self.right.doc());
+                debug_assert!(self.others.iter().all(|docset| docset.doc() == candidate));
+                return candidate;
             }
-            debug_assert_eq!(candidate, self.left.doc());
-            debug_assert_eq!(candidate, self.right.doc());
-            debug_assert!(self.others.iter().all(|docset| docset.doc() == candidate));
-            return candidate;
+            candidate = left.advance();
         }
     }
 
@@ -146,6 +162,20 @@ impl<TDocSet: DocSet, TOtherDocSet: DocSet> DocSet for Intersection<TDocSet, TOt
         doc
     }
 
+    /// Seeks to the target if necessary and checks if the target is an exact match.
+    ///
+    /// Some implementations may choose to advance past the target if beneficial for performance.
+    /// The return value is `true` if the target is in the docset, and `false` otherwise.
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        self.left.seek_into_the_danger_zone(target)
+            && self.right.seek_into_the_danger_zone(target)
+            && self
+                .others
+                .iter_mut()
+                .all(|docset| docset.seek_into_the_danger_zone(target))
+    }
+
+    #[inline]
     fn doc(&self) -> DocId {
         self.left.doc()
     }
@@ -172,6 +202,7 @@ where
     TScorer: Scorer,
     TOtherScorer: Scorer,
 {
+    #[inline]
     fn score(&mut self) -> Score {
         self.left.score()
             + self.right.score()
@@ -181,6 +212,8 @@ where
 
 #[cfg(test)]
 mod tests {
+    use proptest::prelude::*;
+
     use super::Intersection;
     use crate::docset::{DocSet, TERMINATED};
     use crate::postings::tests::test_skip_against_unoptimized;
@@ -270,4 +303,38 @@ mod tests {
         let intersection = Intersection::new(vec![a, b, c], 10);
         assert_eq!(intersection.doc(), TERMINATED);
     }
+
+    // Strategy to generate sorted and deduplicated vectors of u32 document IDs
+    fn sorted_deduped_vec(max_val: u32, max_size: usize) -> impl Strategy<Value = Vec<u32>> {
+        prop::collection::vec(0..max_val, 0..max_size).prop_map(|mut vec| {
+            vec.sort();
+            vec.dedup();
+            vec
+        })
+    }
+
+    proptest! {
+        #[test]
+        fn prop_test_intersection_consistency(
+            a in sorted_deduped_vec(100, 10),
+            b in sorted_deduped_vec(100, 10),
+            num_docs in 100u32..500u32
+        ) {
+            let left = VecDocSet::from(a.clone());
+            let right = VecDocSet::from(b.clone());
+            let mut intersection = Intersection::new(vec![left, right], num_docs);
+
+            let expected: Vec<u32> = a.iter()
+                .cloned()
+                .filter(|doc| b.contains(doc))
+                .collect();
+
+            for expected_doc in expected {
+                assert_eq!(intersection.doc(), expected_doc);
+                intersection.advance();
+            }
+            assert_eq!(intersection.doc(), TERMINATED);
+        }
+
+    }
 }

src/query/mod.rs

@@ -24,7 +24,6 @@ mod reqopt_scorer;
 mod scorer;
 mod set_query;
 mod size_hint;
-mod spatial_query;
 mod term_query;
 mod union;
 mod weight;
@@ -63,7 +62,6 @@ pub use self::reqopt_scorer::RequiredOptionalScorer;
 pub use self::score_combiner::{DisjunctionMaxCombiner, ScoreCombiner, SumCombiner};
 pub use self::scorer::Scorer;
 pub use self::set_query::TermSetQuery;
-pub use self::spatial_query::{SpatialQuery, SpatialQueryType};
 pub use self::term_query::TermQuery;
 pub use self::union::BufferedUnionScorer;
 #[cfg(test)]
@@ -72,9 +70,83 @@ pub use self::weight::Weight;
 
 #[cfg(test)]
 mod tests {
+    use crate::collector::TopDocs;
+    use crate::query::phrase_query::tests::create_index;
     use crate::query::QueryParser;
     use crate::schema::{Schema, TEXT};
-    use crate::{Index, Term};
+    use crate::{DocAddress, Index, Term};
+
+    #[test]
+    pub fn test_mixed_intersection_and_union() -> crate::Result<()> {
+        let index = create_index(&["a b", "a c", "a b c", "b"])?;
+        let schema = index.schema();
+        let text_field = schema.get_field("text").unwrap();
+        let searcher = index.reader()?.searcher();
+
+        let do_search = |term: &str| {
+            let query = QueryParser::for_index(&index, vec![text_field])
+                .parse_query(term)
+                .unwrap();
+            let top_docs: Vec<(f32, DocAddress)> = searcher
+                .search(&query, &TopDocs::with_limit(10).order_by_score())
+                .unwrap();
+
+            top_docs.iter().map(|el| el.1.doc_id).collect::<Vec<_>>()
+        };
+
+        assert_eq!(do_search("a AND b"), vec![0, 2]);
+        assert_eq!(do_search("(a OR b) AND C"), vec![2, 1]);
+        // The intersection code has special code for more than 2 intersections
+        // left, right + others
+        // The will place the union in the "others" insersection to that seek_into_the_danger_zone
+        // is called
+        assert_eq!(
+            do_search("(a OR b) AND (c OR a) AND (b OR c)"),
+            vec![2, 1, 0]
+        );
+
+        Ok(())
+    }
+
+    #[test]
+    pub fn test_mixed_intersection_and_union_with_skip() -> crate::Result<()> {
+        // Test 4096 skip in BufferedUnionScorer
+        let mut data: Vec<&str> = Vec::new();
+        data.push("a b");
+        let zz_data = vec!["z z"; 5000];
+        data.extend_from_slice(&zz_data);
+        data.extend_from_slice(&["a c"]);
+        data.extend_from_slice(&zz_data);
+        data.extend_from_slice(&["a b c", "b"]);
+        let index = create_index(&data)?;
+        let schema = index.schema();
+        let text_field = schema.get_field("text").unwrap();
+        let searcher = index.reader()?.searcher();
+
+        let do_search = |term: &str| {
+            let query = QueryParser::for_index(&index, vec![text_field])
+                .parse_query(term)
+                .unwrap();
+            let top_docs: Vec<(f32, DocAddress)> = searcher
+                .search(&query, &TopDocs::with_limit(10).order_by_score())
+                .unwrap();
+
+            top_docs.iter().map(|el| el.1.doc_id).collect::<Vec<_>>()
+        };
+
+        assert_eq!(do_search("a AND b"), vec![0, 10002]);
+        assert_eq!(do_search("(a OR b) AND C"), vec![10002, 5001]);
+        // The intersection code has special code for more than 2 intersections
+        // left, right + others
+        // The will place the union in the "others" insersection to that seek_into_the_danger_zone
+        // is called
+        assert_eq!(
+            do_search("(a OR b) AND (c OR a) AND (b OR c)"),
+            vec![10002, 5001, 0]
+        );
+
+        Ok(())
+    }
 
     #[test]
     fn test_query_terms() {

src/query/phrase_prefix_query/phrase_prefix_scorer.rs

@@ -81,6 +81,7 @@ impl<TPostings: Postings> DocSet for PhraseKind<TPostings> {
 }
 
 impl<TPostings: Postings> Scorer for PhraseKind<TPostings> {
+    #[inline]
     fn score(&mut self) -> Score {
         match self {
             PhraseKind::SinglePrefix { positions, .. } => {
@@ -193,6 +194,14 @@ impl<TPostings: Postings> DocSet for PhrasePrefixScorer<TPostings> {
         self.advance()
     }
 
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        if self.phrase_scorer.seek_into_the_danger_zone(target) {
+            self.matches_prefix()
+        } else {
+            false
+        }
+    }
+
     fn doc(&self) -> DocId {
         self.phrase_scorer.doc()
     }
@@ -207,6 +216,7 @@ impl<TPostings: Postings> DocSet for PhrasePrefixScorer<TPostings> {
 }
 
 impl<TPostings: Postings> Scorer for PhrasePrefixScorer<TPostings> {
+    #[inline]
     fn score(&mut self) -> Score {
         // TODO modify score??
         self.phrase_scorer.score()

src/query/phrase_query/phrase_scorer.rs

@@ -382,8 +382,9 @@ impl<TPostings: Postings> PhraseScorer<TPostings> {
                 PostingsWithOffset::new(postings, (max_offset - offset) as u32)
             })
             .collect::<Vec<_>>();
+        let intersection_docset = Intersection::new(postings_with_offsets, num_docs);
         let mut scorer = PhraseScorer {
-            intersection_docset: Intersection::new(postings_with_offsets, num_docs),
+            intersection_docset,
             num_terms: num_docsets,
             left_positions: Vec::with_capacity(100),
             right_positions: Vec::with_capacity(100),
@@ -529,25 +530,40 @@ impl<TPostings: Postings> DocSet for PhraseScorer<TPostings> {
         self.advance()
     }
 
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        debug_assert!(target >= self.doc());
+        if self.intersection_docset.seek_into_the_danger_zone(target) && self.phrase_match() {
+            return true;
+        }
+        false
+    }
+
     fn doc(&self) -> DocId {
         self.intersection_docset.doc()
     }
 
     fn size_hint(&self) -> u32 {
-        self.intersection_docset.size_hint()
+        // We adjust the intersection estimate, since actual phrase hits are much lower than where
+        // the all appear.
+        // The estimate should depend on average field length, e.g. if the field is really short
+        // a phrase hit is more likely
+        self.intersection_docset.size_hint() / (10 * self.num_terms as u32)
     }
 
     /// Returns a best-effort hint of the
     /// cost to drive the docset.
     fn cost(&self) -> u64 {
-        // Evaluating phrase matches is generally more expensive than simple term matches,
-        // as it requires loading and comparing positions. Use a conservative multiplier
-        // based on the number of terms.
+        // While determing a potential hit is cheap for phrases, evaluating an actual hit is
+        // expensive since it requires to load positions for a doc and check if they are next to
+        // each other.
+        // So the cost estimation would be the number of times we need to check if a doc is a hit *
+        // 10 * self.num_terms.
         self.intersection_docset.size_hint() as u64 * 10 * self.num_terms as u64
     }
 }
 
 impl<TPostings: Postings> Scorer for PhraseScorer<TPostings> {
+    #[inline]
     fn score(&mut self) -> Score {
         let doc = self.doc();
         let fieldnorm_id = self.fieldnorm_reader.fieldnorm_id(doc);

src/query/query_parser/query_parser.rs

@@ -524,9 +524,6 @@ impl QueryParser {
                 let ip_v6 = IpAddr::from_str(phrase)?.into_ipv6_addr();
                 Ok(Term::from_field_ip_addr(field, ip_v6))
             }
-            FieldType::Spatial(_) => Err(QueryParserError::UnsupportedQuery(
-                "Spatial queries are not yet supported in text query parser".to_string(),
-            )),
         }
     }
 
@@ -627,10 +624,6 @@ impl QueryParser {
                 let term = Term::from_field_ip_addr(field, ip_v6);
                 Ok(vec![LogicalLiteral::Term(term)])
             }
-            FieldType::Spatial(_) => Err(QueryParserError::UnsupportedQuery(format!(
-                "Spatial queries are not yet supported for field '{}'",
-                field_name
-            ))),
         }
     }
 

src/query/range_query/fast_field_range_doc_set.rs

@@ -62,6 +62,17 @@ pub(crate) struct RangeDocSet<T> {
 const DEFAULT_FETCH_HORIZON: u32 = 128;
 impl<T: Send + Sync + PartialOrd + Copy + Debug + 'static> RangeDocSet<T> {
     pub(crate) fn new(value_range: RangeInclusive<T>, column: Column<T>) -> Self {
+        if *value_range.start() > column.max_value() || *value_range.end() < column.min_value() {
+            return Self {
+                value_range,
+                column,
+                loaded_docs: VecCursor::new(),
+                next_fetch_start: TERMINATED,
+                fetch_horizon: DEFAULT_FETCH_HORIZON,
+                last_seek_pos_opt: None,
+            };
+        }
+
         let mut range_docset = Self {
             value_range,
             column,
@@ -81,6 +92,9 @@ impl<T: Send + Sync + PartialOrd + Copy + Debug + 'static> RangeDocSet<T> {
 
     /// Returns true if more data could be fetched
     fn fetch_block(&mut self) {
+        if self.next_fetch_start >= self.column.num_docs() {
+            return;
+        }
         const MAX_HORIZON: u32 = 100_000;
         while self.loaded_docs.is_empty() {
             let finished_to_end = self.fetch_horizon(self.fetch_horizon);
@@ -105,10 +119,10 @@ impl<T: Send + Sync + PartialOrd + Copy + Debug + 'static> RangeDocSet<T> {
     fn fetch_horizon(&mut self, horizon: u32) -> bool {
         let mut finished_to_end = false;
 
-        let limit = self.column.num_docs();
-        let mut end = self.next_fetch_start + horizon;
-        if end >= limit {
-            end = limit;
+        let num_docs = self.column.num_docs();
+        let mut fetch_end = self.next_fetch_start + horizon;
+        if fetch_end >= num_docs {
+            fetch_end = num_docs;
             finished_to_end = true;
         }
 
@@ -116,7 +130,7 @@ impl<T: Send + Sync + PartialOrd + Copy + Debug + 'static> RangeDocSet<T> {
         let doc_buffer: &mut Vec<DocId> = self.loaded_docs.get_cleared_data();
         self.column.get_docids_for_value_range(
             self.value_range.clone(),
-            self.next_fetch_start..end,
+            self.next_fetch_start..fetch_end,
             doc_buffer,
         );
         if let Some(last_doc) = last_doc {
@@ -124,7 +138,7 @@ impl<T: Send + Sync + PartialOrd + Copy + Debug + 'static> RangeDocSet<T> {
                 self.loaded_docs.next();
             }
         }
-        self.next_fetch_start = end;
+        self.next_fetch_start = fetch_end;
 
         finished_to_end
     }
@@ -136,9 +150,6 @@ impl<T: Send + Sync + PartialOrd + Copy + Debug + 'static> DocSet for RangeDocSe
         if let Some(docid) = self.loaded_docs.next() {
             return docid;
         }
-        if self.next_fetch_start >= self.column.num_docs() {
-            return TERMINATED;
-        }
         self.fetch_block();
         self.loaded_docs.current().unwrap_or(TERMINATED)
     }
@@ -174,15 +185,25 @@ impl<T: Send + Sync + PartialOrd + Copy + Debug + 'static> DocSet for RangeDocSe
     }
 
     fn size_hint(&self) -> u32 {
-        self.column.num_docs()
+        // TODO: Implement a better size hint
+        self.column.num_docs() / 10
     }
 
     /// Returns a best-effort hint of the
     /// cost to drive the docset.
     fn cost(&self) -> u64 {
-        // Advancing the docset is relatively expensive since it scans the column.
-        // Keep cost relative to a term query driver; use num_docs as baseline.
-        self.column.num_docs() as u64
+        // Advancing the docset is pretty expensive since it scans the whole column, there is no
+        // index currently (will change with an kd-tree)
+        // Since we use SIMD to scan the fast field range query we lower the cost a little bit,
+        // assuming that we hit 10% of the docs like in size_hint.
+        //
+        // If we would return a cost higher than num_docs, we would never choose ff range query as
+        // the driver in a DocSet, when intersecting a term query with a fast field. But
+        // it's the faster choice when the term query has a lot of docids and the range
+        // query has not.
+        //
+        // Ideally this would take the fast field codec into account
+        (self.column.num_docs() as f64 * 0.8) as u64
     }
 }
 
@@ -236,4 +257,52 @@ mod tests {
         let count = searcher.search(&query, &Count).unwrap();
         assert_eq!(count, 500);
     }
+
+    #[test]
+    fn range_query_no_overlap_optimization() {
+        let mut schema_builder = schema::SchemaBuilder::new();
+        let id_field = schema_builder.add_text_field("id", schema::STRING);
+        let value_field = schema_builder.add_u64_field("value", schema::FAST | schema::INDEXED);
+
+        let dir = RamDirectory::default();
+        let index = IndexBuilder::new()
+            .schema(schema_builder.build())
+            .open_or_create(dir)
+            .unwrap();
+
+        {
+            let mut writer = index.writer(15_000_000).unwrap();
+
+            // Add documents with values in the range [10, 20]
+            for i in 0..100 {
+                let mut doc = TantivyDocument::new();
+                doc.add_text(id_field, format!("doc{i}"));
+                doc.add_u64(value_field, 10 + (i % 11) as u64); // values in range 10-20
+
+                writer.add_document(doc).unwrap();
+            }
+            writer.commit().unwrap();
+        }
+
+        let reader = index.reader().unwrap();
+        let searcher = reader.searcher();
+
+        // Test a range query [100, 200] that has no overlap with data range [10, 20]
+        let query = RangeQuery::new(
+            Bound::Included(Term::from_field_u64(value_field, 100)),
+            Bound::Included(Term::from_field_u64(value_field, 200)),
+        );
+
+        let count = searcher.search(&query, &Count).unwrap();
+        assert_eq!(count, 0); // should return 0 results since there's no overlap
+
+        // Test another non-overlapping range: [0, 5] while data range is [10, 20]
+        let query2 = RangeQuery::new(
+            Bound::Included(Term::from_field_u64(value_field, 0)),
+            Bound::Included(Term::from_field_u64(value_field, 5)),
+        );
+
+        let count2 = searcher.search(&query2, &Count).unwrap();
+        assert_eq!(count2, 0); // should return 0 results since there's no overlap
+    }
 }

src/query/range_query/mod.rs

@@ -20,6 +20,6 @@ pub(crate) fn is_type_valid_for_fastfield_range_query(typ: Type) -> bool {
         | Type::Date
         | Type::Json
         | Type::IpAddr => true,
-        Type::Facet | Type::Bytes | Type::Spatial => false,
+        Type::Facet | Type::Bytes => false,
     }
 }

src/query/range_query/range_query_fastfield.rs

@@ -128,15 +128,12 @@ impl Weight for FastFieldRangeWeight {
                         BoundsRange::new(bounds.lower_bound, bounds.upper_bound),
                     )
                 }
-                Type::Bool
-                | Type::Facet
-                | Type::Bytes
-                | Type::Json
-                | Type::IpAddr
-                | Type::Spatial => Err(crate::TantivyError::InvalidArgument(format!(
-                    "unsupported value bytes type in json term value_bytes {:?}",
-                    term_value.typ()
-                ))),
+                Type::Bool | Type::Facet | Type::Bytes | Type::Json | Type::IpAddr => {
+                    Err(crate::TantivyError::InvalidArgument(format!(
+                        "unsupported value bytes type in json term value_bytes {:?}",
+                        term_value.typ()
+                    )))
+                }
             }
         } else if field_type.is_ip_addr() {
             let parse_ip_from_bytes = |term: &Term| {
@@ -438,7 +435,7 @@ pub(crate) fn maps_to_u64_fastfield(typ: Type) -> bool {
     match typ {
         Type::U64 | Type::I64 | Type::F64 | Type::Bool | Type::Date => true,
         Type::IpAddr => false,
-        Type::Str | Type::Facet | Type::Bytes | Type::Json | Type::Spatial => false,
+        Type::Str | Type::Facet | Type::Bytes | Type::Json => false,
     }
 }
 
@@ -1601,449 +1598,3 @@ pub(crate) mod ip_range_tests {
         Ok(())
     }
 }
-
-#[cfg(all(test, feature = "unstable"))]
-mod bench {
-
-    use rand::rngs::StdRng;
-    use rand::{Rng, SeedableRng};
-    use test::Bencher;
-
-    use super::tests::*;
-    use super::*;
-    use crate::collector::Count;
-    use crate::query::QueryParser;
-    use crate::Index;
-
-    fn get_index_0_to_100() -> Index {
-        let mut rng = StdRng::from_seed([1u8; 32]);
-        let num_vals = 100_000;
-        let docs: Vec<_> = (0..num_vals)
-            .map(|_i| {
-                let id_name = if rng.gen_bool(0.01) {
-                    "veryfew".to_string() // 1%
-                } else if rng.gen_bool(0.1) {
-                    "few".to_string() // 9%
-                } else {
-                    "many".to_string() // 90%
-                };
-                Doc {
-                    id_name,
-                    id: rng.gen_range(0..100),
-                }
-            })
-            .collect();
-
-        create_index_from_docs(&docs, false)
-    }
-
-    fn get_90_percent() -> RangeInclusive<u64> {
-        0..=90
-    }
-
-    fn get_10_percent() -> RangeInclusive<u64> {
-        0..=10
-    }
-
-    fn get_1_percent() -> RangeInclusive<u64> {
-        10..=10
-    }
-
-    fn execute_query(
-        field: &str,
-        id_range: RangeInclusive<u64>,
-        suffix: &str,
-        index: &Index,
-    ) -> usize {
-        let gen_query_inclusive = |from: &u64, to: &u64| {
-            format!(
-                "{}:[{} TO {}] {}",
-                field,
-                &from.to_string(),
-                &to.to_string(),
-                suffix
-            )
-        };
-
-        let query = gen_query_inclusive(id_range.start(), id_range.end());
-        let query_from_text = |text: &str| {
-            QueryParser::for_index(index, vec![])
-                .parse_query(text)
-                .unwrap()
-        };
-        let query = query_from_text(&query);
-        let reader = index.reader().unwrap();
-        let searcher = reader.searcher();
-        searcher.search(&query, &(Count)).unwrap()
-    }
-
-    #[bench]
-    fn bench_id_range_hit_90_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_90_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_10_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_10_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_1_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_1_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_10_percent_intersect_with_10_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_10_percent(), "AND id_name:few", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_1_percent_intersect_with_10_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_1_percent(), "AND id_name:few", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_1_percent_intersect_with_90_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_1_percent(), "AND id_name:many", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_1_percent_intersect_with_1_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_1_percent(), "AND id_name:veryfew", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_10_percent_intersect_with_90_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_10_percent(), "AND id_name:many", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_90_percent_intersect_with_90_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_90_percent(), "AND id_name:many", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_90_percent_intersect_with_10_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_90_percent(), "AND id_name:few", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_90_percent_intersect_with_1_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("id", get_90_percent(), "AND id_name:veryfew", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_90_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_90_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_10_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_10_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_1_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_1_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_10_percent_intersect_with_10_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_10_percent(), "AND id_name:few", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_1_percent_intersect_with_10_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_1_percent(), "AND id_name:few", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_1_percent_intersect_with_90_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_1_percent(), "AND id_name:many", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_1_percent_intersect_with_1_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_1_percent(), "AND id_name:veryfew", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_10_percent_intersect_with_90_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_10_percent(), "AND id_name:many", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_90_percent_intersect_with_90_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_90_percent(), "AND id_name:many", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_90_percent_intersect_with_10_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_90_percent(), "AND id_name:few", &index));
-    }
-
-    #[bench]
-    fn bench_id_range_hit_90_percent_intersect_with_1_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ids", get_90_percent(), "AND id_name:veryfew", &index));
-    }
-}
-
-#[cfg(all(test, feature = "unstable"))]
-mod bench_ip {
-
-    use rand::rngs::StdRng;
-    use rand::{Rng, SeedableRng};
-    use test::Bencher;
-
-    use super::ip_range_tests::*;
-    use super::*;
-    use crate::collector::Count;
-    use crate::query::QueryParser;
-    use crate::Index;
-
-    fn get_index_0_to_100() -> Index {
-        let mut rng = StdRng::from_seed([1u8; 32]);
-        let num_vals = 100_000;
-        let docs: Vec<_> = (0..num_vals)
-            .map(|_i| {
-                let id = if rng.gen_bool(0.01) {
-                    "veryfew".to_string() // 1%
-                } else if rng.gen_bool(0.1) {
-                    "few".to_string() // 9%
-                } else {
-                    "many".to_string() // 90%
-                };
-                Doc {
-                    id,
-                    // Multiply by 1000, so that we create many buckets in the compact space
-                    // The benches depend on this range to select n-percent of elements with the
-                    // methods below.
-                    ip: Ipv6Addr::from_u128(rng.gen_range(0..100) * 1000),
-                }
-            })
-            .collect();
-
-        create_index_from_ip_docs(&docs)
-    }
-
-    fn get_90_percent() -> RangeInclusive<Ipv6Addr> {
-        let start = Ipv6Addr::from_u128(0);
-        let end = Ipv6Addr::from_u128(90 * 1000);
-        start..=end
-    }
-
-    fn get_10_percent() -> RangeInclusive<Ipv6Addr> {
-        let start = Ipv6Addr::from_u128(0);
-        let end = Ipv6Addr::from_u128(10 * 1000);
-        start..=end
-    }
-
-    fn get_1_percent() -> RangeInclusive<Ipv6Addr> {
-        let start = Ipv6Addr::from_u128(10 * 1000);
-        let end = Ipv6Addr::from_u128(10 * 1000);
-        start..=end
-    }
-
-    fn execute_query(
-        field: &str,
-        ip_range: RangeInclusive<Ipv6Addr>,
-        suffix: &str,
-        index: &Index,
-    ) -> usize {
-        let gen_query_inclusive = |from: &Ipv6Addr, to: &Ipv6Addr| {
-            format!(
-                "{}:[{} TO {}] {}",
-                field,
-                &from.to_string(),
-                &to.to_string(),
-                suffix
-            )
-        };
-
-        let query = gen_query_inclusive(ip_range.start(), ip_range.end());
-        let query_from_text = |text: &str| {
-            QueryParser::for_index(index, vec![])
-                .parse_query(text)
-                .unwrap()
-        };
-        let query = query_from_text(&query);
-        let reader = index.reader().unwrap();
-        let searcher = reader.searcher();
-        searcher.search(&query, &(Count)).unwrap()
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_90_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_90_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_10_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_10_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_1_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_1_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_10_percent_intersect_with_10_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_10_percent(), "AND id:few", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_1_percent_intersect_with_10_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_1_percent(), "AND id:few", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_1_percent_intersect_with_90_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_1_percent(), "AND id:many", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_1_percent_intersect_with_1_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_1_percent(), "AND id:veryfew", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_10_percent_intersect_with_90_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_10_percent(), "AND id:many", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_90_percent_intersect_with_90_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_90_percent(), "AND id:many", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_90_percent_intersect_with_10_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_90_percent(), "AND id:few", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_90_percent_intersect_with_1_percent(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ip", get_90_percent(), "AND id:veryfew", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_90_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_90_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_10_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_10_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_1_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_1_percent(), "", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_10_percent_intersect_with_10_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_10_percent(), "AND id:few", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_1_percent_intersect_with_10_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_1_percent(), "AND id:few", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_1_percent_intersect_with_90_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-        bench.iter(|| execute_query("ips", get_1_percent(), "AND id:many", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_1_percent_intersect_with_1_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_1_percent(), "AND id:veryfew", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_10_percent_intersect_with_90_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_10_percent(), "AND id:many", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_90_percent_intersect_with_90_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_90_percent(), "AND id:many", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_90_percent_intersect_with_10_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_90_percent(), "AND id:few", &index));
-    }
-
-    #[bench]
-    fn bench_ip_range_hit_90_percent_intersect_with_1_percent_multi(bench: &mut Bencher) {
-        let index = get_index_0_to_100();
-
-        bench.iter(|| execute_query("ips", get_90_percent(), "AND id:veryfew", &index));
-    }
-}

src/query/reqopt_scorer.rs

@@ -56,6 +56,11 @@ where
         self.req_scorer.seek(target)
     }
 
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        self.score_cache = None;
+        self.req_scorer.seek_into_the_danger_zone(target)
+    }
+
     fn doc(&self) -> DocId {
         self.req_scorer.doc()
     }
@@ -76,6 +81,7 @@ where
     TOptScorer: Scorer,
     TScoreCombiner: ScoreCombiner,
 {
+    #[inline]
     fn score(&mut self) -> Score {
         if let Some(score) = self.score_cache {
             return score;

src/query/score_combiner.rs

@@ -29,6 +29,7 @@ impl ScoreCombiner for DoNothingCombiner {
 
     fn clear(&mut self) {}
 
+    #[inline]
     fn score(&self) -> Score {
         1.0
     }
@@ -49,6 +50,7 @@ impl ScoreCombiner for SumCombiner {
         self.score = 0.0;
     }
 
+    #[inline]
     fn score(&self) -> Score {
         self.score
     }
@@ -86,6 +88,7 @@ impl ScoreCombiner for DisjunctionMaxCombiner {
         self.sum = 0.0;
     }
 
+    #[inline]
     fn score(&self) -> Score {
         self.max + (self.sum - self.max) * self.tie_breaker
     }

src/query/scorer.rs

@@ -18,6 +18,7 @@ pub trait Scorer: downcast_rs::Downcast + DocSet + 'static {
 impl_downcast!(Scorer);
 
 impl Scorer for Box<dyn Scorer> {
+    #[inline]
     fn score(&mut self) -> Score {
         self.deref_mut().score()
     }

src/query/spatial_query.rs

@@ -1,186 +0,0 @@
-//! HUSH
-
-use common::BitSet;
-
-use crate::query::explanation::does_not_match;
-use crate::query::{BitSetDocSet, Explanation, Query, Scorer, Weight};
-use crate::schema::Field;
-use crate::spatial::bkd::{search_intersects, Segment};
-use crate::spatial::point::GeoPoint;
-use crate::spatial::writer::as_point_i32;
-use crate::{DocId, DocSet, Score, TantivyError, TERMINATED};
-
-#[derive(Clone, Copy, Debug)]
-/// HUSH
-pub enum SpatialQueryType {
-    /// HUSH
-    Intersects,
-    //      Within,
-    // Contains,
-}
-
-#[derive(Clone, Copy, Debug)]
-/// HUSH
-pub struct SpatialQuery {
-    field: Field,
-    bounds: [(i32, i32); 2],
-    query_type: SpatialQueryType,
-}
-
-impl SpatialQuery {
-    /// HUSH
-    pub fn new(field: Field, bounds: [GeoPoint; 2], query_type: SpatialQueryType) -> Self {
-        SpatialQuery {
-            field,
-            bounds: [as_point_i32(bounds[0]), as_point_i32(bounds[1])],
-            query_type,
-        }
-    }
-}
-
-impl Query for SpatialQuery {
-    fn weight(
-        &self,
-        _enable_scoring: super::EnableScoring<'_>,
-    ) -> crate::Result<Box<dyn super::Weight>> {
-        Ok(Box::new(SpatialWeight::new(
-            self.field,
-            self.bounds,
-            self.query_type,
-        )))
-    }
-}
-
-pub struct SpatialWeight {
-    field: Field,
-    bounds: [(i32, i32); 2],
-    query_type: SpatialQueryType,
-}
-
-impl SpatialWeight {
-    fn new(field: Field, bounds: [(i32, i32); 2], query_type: SpatialQueryType) -> Self {
-        SpatialWeight {
-            field,
-            bounds,
-            query_type,
-        }
-    }
-}
-
-impl Weight for SpatialWeight {
-    fn scorer(
-        &self,
-        reader: &crate::SegmentReader,
-        boost: crate::Score,
-    ) -> crate::Result<Box<dyn super::Scorer>> {
-        let spatial_reader = reader
-            .spatial_fields()
-            .get_field(self.field)?
-            .ok_or_else(|| TantivyError::SchemaError(format!("No spatial data for field")))?;
-        let block_kd_tree = Segment::new(spatial_reader.get_bytes());
-        match self.query_type {
-            SpatialQueryType::Intersects => {
-                let mut include = BitSet::with_max_value(reader.max_doc());
-                search_intersects(
-                    &block_kd_tree,
-                    block_kd_tree.root_offset,
-                    &[
-                        self.bounds[0].1,
-                        self.bounds[0].0,
-                        self.bounds[1].1,
-                        self.bounds[1].0,
-                    ],
-                    &mut include,
-                )?;
-                Ok(Box::new(SpatialScorer::new(boost, include, None)))
-            }
-        }
-    }
-    fn explain(
-        &self,
-        reader: &crate::SegmentReader,
-        doc: DocId,
-    ) -> crate::Result<super::Explanation> {
-        let mut scorer = self.scorer(reader, 1.0)?;
-        if scorer.seek(doc) != doc {
-            return Err(does_not_match(doc));
-        }
-        let query_type_desc = match self.query_type {
-            SpatialQueryType::Intersects => "SpatialQuery::Intersects",
-        };
-        let score = scorer.score();
-        let mut explanation = Explanation::new(query_type_desc, score);
-        explanation.add_context(format!(
-            "bounds: [({}, {}), ({}, {})]",
-            self.bounds[0].0, self.bounds[0].1, self.bounds[1].0, self.bounds[1].1,
-        ));
-        explanation.add_context(format!("field: {:?}", self.field));
-        Ok(explanation)
-    }
-}
-
-struct SpatialScorer {
-    include: BitSetDocSet,
-    exclude: Option<BitSet>,
-    doc_id: DocId,
-    score: Score,
-}
-
-impl SpatialScorer {
-    pub fn new(score: Score, include: BitSet, exclude: Option<BitSet>) -> Self {
-        let mut scorer = SpatialScorer {
-            include: BitSetDocSet::from(include),
-            exclude,
-            doc_id: 0,
-            score,
-        };
-        scorer.prime();
-        scorer
-    }
-    fn prime(&mut self) {
-        self.doc_id = self.include.doc();
-        while self.exclude() {
-            self.doc_id = self.include.advance();
-        }
-    }
-
-    fn exclude(&self) -> bool {
-        if self.doc_id == TERMINATED {
-            return false;
-        }
-        match &self.exclude {
-            Some(exclude) => exclude.contains(self.doc_id),
-            None => false,
-        }
-    }
-}
-
-impl Scorer for SpatialScorer {
-    fn score(&mut self) -> Score {
-        self.score
-    }
-}
-
-impl DocSet for SpatialScorer {
-    fn advance(&mut self) -> DocId {
-        if self.doc_id == TERMINATED {
-            return TERMINATED;
-        }
-        self.doc_id = self.include.advance();
-        while self.exclude() {
-            self.doc_id = self.include.advance();
-        }
-        self.doc_id
-    }
-
-    fn size_hint(&self) -> u32 {
-        match &self.exclude {
-            Some(exclude) => self.include.size_hint() - exclude.len() as u32,
-            None => self.include.size_hint(),
-        }
-    }
-
-    fn doc(&self) -> DocId {
-        self.doc_id
-    }
-}

src/query/term_query/term_scorer.rs

@@ -98,14 +98,17 @@ impl TermScorer {
 }
 
 impl DocSet for TermScorer {
+    #[inline]
     fn advance(&mut self) -> DocId {
         self.postings.advance()
     }
 
+    #[inline]
     fn seek(&mut self, target: DocId) -> DocId {
         self.postings.seek(target)
     }
 
+    #[inline]
     fn doc(&self) -> DocId {
         self.postings.doc()
     }
@@ -116,6 +119,7 @@ impl DocSet for TermScorer {
 }
 
 impl Scorer for TermScorer {
+    #[inline]
     fn score(&mut self) -> Score {
         let fieldnorm_id = self.fieldnorm_id();
         let term_freq = self.term_freq();

src/query/union/buffered_union.rs

@@ -15,7 +15,7 @@ const HORIZON: u32 = 64u32 * 64u32;
 // This function is similar except that it does is not unstable, and
 // it does not keep the original vector ordering.
 //
-// Also, it does not "yield" any elements.
+// Elements are dropped and not yielded.
 fn unordered_drain_filter<T, P>(v: &mut Vec<T>, mut predicate: P)
 where P: FnMut(&mut T) -> bool {
     let mut i = 0;
@@ -128,6 +128,7 @@ impl<TScorer: Scorer, TScoreCombiner: ScoreCombiner> BufferedUnionScorer<TScorer
         }
     }
 
+    #[inline]
     fn advance_buffered(&mut self) -> bool {
         while self.bucket_idx < HORIZON_NUM_TINYBITSETS {
             if let Some(val) = self.bitsets[self.bucket_idx].pop_lowest() {
@@ -143,6 +144,12 @@ impl<TScorer: Scorer, TScoreCombiner: ScoreCombiner> BufferedUnionScorer<TScorer
         }
         false
     }
+
+    fn is_in_horizon(&self, target: DocId) -> bool {
+        // wrapping_sub, because target may be < window_start_doc
+        let gap = target.wrapping_sub(self.window_start_doc);
+        gap < HORIZON
+    }
 }
 
 impl<TScorer, TScoreCombiner> DocSet for BufferedUnionScorer<TScorer, TScoreCombiner>
@@ -150,6 +157,7 @@ where
     TScorer: Scorer,
     TScoreCombiner: ScoreCombiner,
 {
+    #[inline]
     fn advance(&mut self) -> DocId {
         if self.advance_buffered() {
             return self.doc;
@@ -217,8 +225,29 @@ where
         }
     }
 
-    // TODO Also implement `count` with deletes efficiently.
+    fn seek_into_the_danger_zone(&mut self, target: DocId) -> bool {
+        if self.is_in_horizon(target) {
+            // Our value is within the buffered horizon and the docset may already have been
+            // processed and removed, so we need to use seek, which uses the regular advance.
+            self.seek(target) == target
+        } else {
+            // The docsets are not in the buffered range, so we can use seek_into_the_danger_zone
+            // of the underlying docsets
+            let is_hit = self
+                .docsets
+                .iter_mut()
+                .any(|docset| docset.seek_into_the_danger_zone(target));
 
+            // The API requires the DocSet to be in a valid state when `seek_into_the_danger_zone`
+            // returns true.
+            if is_hit {
+                self.seek(target);
+            }
+            is_hit
+        }
+    }
+
+    #[inline]
     fn doc(&self) -> DocId {
         self.doc
     }
@@ -231,6 +260,7 @@ where
         self.docsets.iter().map(|docset| docset.cost()).sum()
     }
 
+    // TODO Also implement `count` with deletes efficiently.
     fn count_including_deleted(&mut self) -> u32 {
         if self.doc == TERMINATED {
             return 0;
@@ -259,6 +289,7 @@ where
     TScoreCombiner: ScoreCombiner,
     TScorer: Scorer,
 {
+    #[inline]
     fn score(&mut self) -> Score {
         self.score
     }

src/query/union/simple_union.rs

@@ -92,6 +92,7 @@ impl<TDocSet: DocSet> DocSet for SimpleUnion<TDocSet> {
     }
 
     fn size_hint(&self) -> u32 {
+        // TODO: use estimate_union
         self.docsets
             .iter()
             .map(|docset| docset.size_hint())

src/schema/document/de.rs

@@ -22,7 +22,6 @@ use super::se::BinaryObjectSerializer;
 use super::{OwnedValue, Value};
 use crate::schema::document::type_codes;
 use crate::schema::{Facet, Field};
-use crate::spatial::geometry::Geometry;
 use crate::store::DocStoreVersion;
 use crate::tokenizer::PreTokenizedString;
 
@@ -130,9 +129,6 @@ pub trait ValueDeserializer<'de> {
     /// Attempts to deserialize a pre-tokenized string value from the deserializer.
     fn deserialize_pre_tokenized_string(self) -> Result<PreTokenizedString, DeserializeError>;
 
-    /// HUSH
-    fn deserialize_geometry(self) -> Result<Geometry, DeserializeError>;
-
     /// Attempts to deserialize the value using a given visitor.
     fn deserialize_any<V>(self, visitor: V) -> Result<V::Value, DeserializeError>
     where V: ValueVisitor;
@@ -170,8 +166,6 @@ pub enum ValueType {
     /// A JSON object value. Deprecated.
     #[deprecated(note = "We keep this for backwards compatibility, use Object instead")]
     JSONObject,
-    /// HUSH
-    Geometry,
 }
 
 /// A value visitor for deserializing a document value.
@@ -252,12 +246,6 @@ pub trait ValueVisitor {
         Err(DeserializeError::UnsupportedType(ValueType::PreTokStr))
     }
 
-    #[inline]
-    /// Called when the deserializer visits a geometry value.
-    fn visit_geometry(&self, _val: Geometry) -> Result<Self::Value, DeserializeError> {
-        Err(DeserializeError::UnsupportedType(ValueType::Geometry))
-    }
-
     #[inline]
     /// Called when the deserializer visits an array.
     fn visit_array<'de, A>(&self, _access: A) -> Result<Self::Value, DeserializeError>
@@ -392,7 +380,6 @@ where R: Read
 
                 match ext_type_code {
                     type_codes::TOK_STR_EXT_CODE => ValueType::PreTokStr,
-                    type_codes::GEO_EXT_CODE => ValueType::Geometry,
                     _ => {
                         return Err(DeserializeError::from(io::Error::new(
                             io::ErrorKind::InvalidData,
@@ -508,11 +495,6 @@ where R: Read
             .map_err(DeserializeError::from)
     }
 
-    fn deserialize_geometry(self) -> Result<Geometry, DeserializeError> {
-        self.validate_type(ValueType::Geometry)?;
-        <Geometry as BinarySerializable>::deserialize(self.reader).map_err(DeserializeError::from)
-    }
-
     fn deserialize_any<V>(self, visitor: V) -> Result<V::Value, DeserializeError>
     where V: ValueVisitor {
         match self.value_type {
@@ -557,10 +539,6 @@ where R: Read
                 let val = self.deserialize_pre_tokenized_string()?;
                 visitor.visit_pre_tokenized_string(val)
             }
-            ValueType::Geometry => {
-                let val = self.deserialize_geometry()?;
-                visitor.visit_geometry(val)
-            }
             ValueType::Array => {
                 let access =
                     BinaryArrayDeserializer::from_reader(self.reader, self.doc_store_version)?;

src/schema/document/default_document.rs

@@ -13,7 +13,6 @@ use crate::schema::document::{
 };
 use crate::schema::field_type::ValueParsingError;
 use crate::schema::{Facet, Field, NamedFieldDocument, OwnedValue, Schema};
-use crate::spatial::geometry::Geometry;
 use crate::tokenizer::PreTokenizedString;
 
 #[repr(C, packed)]
@@ -255,7 +254,6 @@ impl CompactDoc {
             }
             ReferenceValueLeaf::IpAddr(num) => write_into(&mut self.node_data, num.to_u128()),
             ReferenceValueLeaf::PreTokStr(pre_tok) => write_into(&mut self.node_data, *pre_tok),
-            ReferenceValueLeaf::Geometry(geometry) => write_into(&mut self.node_data, *geometry),
         };
         ValueAddr { type_id, val_addr }
     }
@@ -466,12 +464,6 @@ impl<'a> CompactDocValue<'a> {
                 .map(Into::into)
                 .map(ReferenceValueLeaf::PreTokStr)
                 .map(Into::into),
-            ValueType::Geometry => self
-                .container
-                .read_from::<Geometry>(addr)
-                .map(Into::into)
-                .map(ReferenceValueLeaf::Geometry)
-                .map(Into::into),
             ValueType::Object => Ok(ReferenceValue::Object(CompactDocObjectIter::new(
                 self.container,
                 addr,
@@ -550,8 +542,6 @@ pub enum ValueType {
     Object = 11,
     /// Pre-tokenized str type,
     Array = 12,
-    /// HUSH
-    Geometry = 13,
 }
 
 impl BinarySerializable for ValueType {
@@ -597,7 +587,6 @@ impl<'a> From<&ReferenceValueLeaf<'a>> for ValueType {
             ReferenceValueLeaf::PreTokStr(_) => ValueType::PreTokStr,
             ReferenceValueLeaf::Facet(_) => ValueType::Facet,
             ReferenceValueLeaf::Bytes(_) => ValueType::Bytes,
-            ReferenceValueLeaf::Geometry(_) => ValueType::Geometry,
         }
     }
 }

src/schema/document/mod.rs

@@ -273,5 +273,4 @@ pub(crate) mod type_codes {
 
     // Extended type codes
     pub const TOK_STR_EXT_CODE: u8 = 0;
-    pub const GEO_EXT_CODE: u8 = 1;
 }

src/schema/document/owned_value.rs

@@ -15,7 +15,6 @@ use crate::schema::document::{
     ValueDeserializer, ValueVisitor,
 };
 use crate::schema::Facet;
-use crate::spatial::geometry::Geometry;
 use crate::tokenizer::PreTokenizedString;
 use crate::DateTime;
 
@@ -50,8 +49,6 @@ pub enum OwnedValue {
     Object(Vec<(String, Self)>),
     /// IpV6 Address. Internally there is no IpV4, it needs to be converted to `Ipv6Addr`.
     IpAddr(Ipv6Addr),
-    /// A GeoRust multi-polygon.
-    Geometry(Geometry),
 }
 
 impl AsRef<OwnedValue> for OwnedValue {
@@ -61,6 +58,31 @@ impl AsRef<OwnedValue> for OwnedValue {
     }
 }
 
+impl OwnedValue {
+    /// Returns a u8 discriminant value for the `OwnedValue` variant.
+    ///
+    /// This can be used to sort `OwnedValue` instances by their type.
+    pub fn discriminant_value(&self) -> u8 {
+        match self {
+            OwnedValue::Null => 0,
+            OwnedValue::Str(_) => 1,
+            OwnedValue::PreTokStr(_) => 2,
+            // It is key to make sure U64, I64, F64 are grouped together in there, otherwise we
+            // might be breaking transivity.
+            OwnedValue::U64(_) => 3,
+            OwnedValue::I64(_) => 4,
+            OwnedValue::F64(_) => 5,
+            OwnedValue::Bool(_) => 6,
+            OwnedValue::Date(_) => 7,
+            OwnedValue::Facet(_) => 8,
+            OwnedValue::Bytes(_) => 9,
+            OwnedValue::Array(_) => 10,
+            OwnedValue::Object(_) => 11,
+            OwnedValue::IpAddr(_) => 12,
+        }
+    }
+}
+
 impl<'a> Value<'a> for &'a OwnedValue {
     type ArrayIter = std::slice::Iter<'a, OwnedValue>;
     type ObjectIter = ObjectMapIter<'a>;
@@ -80,9 +102,6 @@ impl<'a> Value<'a> for &'a OwnedValue {
             OwnedValue::IpAddr(val) => ReferenceValueLeaf::IpAddr(*val).into(),
             OwnedValue::Array(array) => ReferenceValue::Array(array.iter()),
             OwnedValue::Object(object) => ReferenceValue::Object(ObjectMapIter(object.iter())),
-            OwnedValue::Geometry(geometry) => {
-                ReferenceValueLeaf::Geometry(Box::new(geometry.clone())).into()
-            }
         }
     }
 }
@@ -142,10 +161,6 @@ impl ValueDeserialize for OwnedValue {
                 Ok(OwnedValue::PreTokStr(val))
             }
 
-            fn visit_geometry(&self, val: Geometry) -> Result<Self::Value, DeserializeError> {
-                Ok(OwnedValue::Geometry(val))
-            }
-
             fn visit_array<'de, A>(&self, mut access: A) -> Result<Self::Value, DeserializeError>
             where A: ArrayAccess<'de> {
                 let mut elements = Vec::with_capacity(access.size_hint());
@@ -208,7 +223,6 @@ impl serde::Serialize for OwnedValue {
                 }
             }
             OwnedValue::Array(ref array) => array.serialize(serializer),
-            OwnedValue::Geometry(ref geometry) => geometry.to_geojson().serialize(serializer),
         }
     }
 }
@@ -296,7 +310,6 @@ impl<'a, V: Value<'a>> From<ReferenceValue<'a, V>> for OwnedValue {
                 ReferenceValueLeaf::IpAddr(val) => OwnedValue::IpAddr(val),
                 ReferenceValueLeaf::Bool(val) => OwnedValue::Bool(val),
                 ReferenceValueLeaf::PreTokStr(val) => OwnedValue::PreTokStr(*val.clone()),
-                ReferenceValueLeaf::Geometry(val) => OwnedValue::Geometry(*val.clone()),
             },
             ReferenceValue::Array(val) => {
                 OwnedValue::Array(val.map(|v| v.as_value().into()).collect())

src/schema/document/se.rs

@@ -133,10 +133,6 @@ where W: Write
                     self.write_type_code(type_codes::EXT_CODE)?;
                     self.serialize_with_type_code(type_codes::TOK_STR_EXT_CODE, &*val)
                 }
-                ReferenceValueLeaf::Geometry(val) => {
-                    self.write_type_code(type_codes::EXT_CODE)?;
-                    self.serialize_with_type_code(type_codes::GEO_EXT_CODE, &*val)
-                }
             },
             ReferenceValue::Array(elements) => {
                 self.write_type_code(type_codes::ARRAY_CODE)?;

src/schema/document/value.rs

@@ -3,7 +3,6 @@ use std::net::Ipv6Addr;
 
 use common::DateTime;
 
-use crate::spatial::geometry::Geometry;
 use crate::tokenizer::PreTokenizedString;
 
 /// A single field value.
@@ -109,12 +108,6 @@ pub trait Value<'a>: Send + Sync + Debug {
             None
         }
     }
-
-    #[inline]
-    /// HUSH
-    fn as_geometry(&self) -> Option<Box<Geometry>> {
-        self.as_leaf().and_then(|leaf| leaf.into_geometry())
-    }
 }
 
 /// A enum representing a leaf value for tantivy to index.
@@ -143,8 +136,6 @@ pub enum ReferenceValueLeaf<'a> {
     Bool(bool),
     /// Pre-tokenized str type,
     PreTokStr(Box<PreTokenizedString>),
-    /// HUSH
-    Geometry(Box<Geometry>),
 }
 
 impl From<u64> for ReferenceValueLeaf<'_> {
@@ -229,9 +220,6 @@ impl<'a, T: Value<'a> + ?Sized> From<ReferenceValueLeaf<'a>> for ReferenceValue<
             ReferenceValueLeaf::PreTokStr(val) => {
                 ReferenceValue::Leaf(ReferenceValueLeaf::PreTokStr(val))
             }
-            ReferenceValueLeaf::Geometry(val) => {
-                ReferenceValue::Leaf(ReferenceValueLeaf::Geometry(val))
-            }
         }
     }
 }
@@ -343,16 +331,6 @@ impl<'a> ReferenceValueLeaf<'a> {
             None
         }
     }
-
-    #[inline]
-    /// HUSH
-    pub fn into_geometry(self) -> Option<Box<Geometry>> {
-        if let Self::Geometry(val) = self {
-            Some(val)
-        } else {
-            None
-        }
-    }
 }
 
 /// A enum representing a value for tantivy to index.
@@ -470,10 +448,4 @@ where V: Value<'a>
     pub fn is_object(&self) -> bool {
         matches!(self, Self::Object(_))
     }
-
-    #[inline]
-    /// HUSH
-    pub fn into_geometry(self) -> Option<Box<Geometry>> {
-        self.into_leaf().and_then(|leaf| leaf.into_geometry())
-    }
 }

src/schema/field_entry.rs

@@ -1,7 +1,6 @@
 use serde::{Deserialize, Serialize};
 
 use super::ip_options::IpAddrOptions;
-use super::spatial_options::SpatialOptions;
 use crate::schema::bytes_options::BytesOptions;
 use crate::schema::{
     is_valid_field_name, DateOptions, FacetOptions, FieldType, JsonObjectOptions, NumericOptions,
@@ -81,11 +80,6 @@ impl FieldEntry {
         Self::new(field_name, FieldType::JsonObject(json_object_options))
     }
 
-    /// Creates a field entry for a spatial field
-    pub fn new_spatial(field_name: String, spatial_options: SpatialOptions) -> FieldEntry {
-        Self::new(field_name, FieldType::Spatial(spatial_options))
-    }
-
     /// Returns the name of the field
     pub fn name(&self) -> &str {
         &self.name
@@ -135,7 +129,6 @@ impl FieldEntry {
             FieldType::Bytes(ref options) => options.is_stored(),
             FieldType::JsonObject(ref options) => options.is_stored(),
             FieldType::IpAddr(ref options) => options.is_stored(),
-            FieldType::Spatial(ref options) => options.is_stored(),
         }
     }
 }

src/schema/field_type.rs

@@ -9,7 +9,6 @@ use serde_json::Value as JsonValue;
 use thiserror::Error;
 
 use super::ip_options::IpAddrOptions;
-use super::spatial_options::SpatialOptions;
 use super::IntoIpv6Addr;
 use crate::schema::bytes_options::BytesOptions;
 use crate::schema::facet_options::FacetOptions;
@@ -17,7 +16,6 @@ use crate::schema::{
     DateOptions, Facet, IndexRecordOption, JsonObjectOptions, NumericOptions, OwnedValue,
     TextFieldIndexing, TextOptions,
 };
-use crate::spatial::geometry::Geometry;
 use crate::time::format_description::well_known::Rfc3339;
 use crate::time::OffsetDateTime;
 use crate::tokenizer::PreTokenizedString;
@@ -73,8 +71,6 @@ pub enum Type {
     Json = b'j',
     /// IpAddr
     IpAddr = b'p',
-    /// Spatial
-    Spatial = b't',
 }
 
 impl From<ColumnType> for Type {
@@ -143,7 +139,6 @@ impl Type {
             Type::Bytes => "Bytes",
             Type::Json => "Json",
             Type::IpAddr => "IpAddr",
-            Type::Spatial => "Spatial",
         }
     }
 
@@ -194,8 +189,6 @@ pub enum FieldType {
     JsonObject(JsonObjectOptions),
     /// IpAddr field
     IpAddr(IpAddrOptions),
-    /// Spatial field
-    Spatial(SpatialOptions),
 }
 
 impl FieldType {
@@ -212,7 +205,6 @@ impl FieldType {
             FieldType::Bytes(_) => Type::Bytes,
             FieldType::JsonObject(_) => Type::Json,
             FieldType::IpAddr(_) => Type::IpAddr,
-            FieldType::Spatial(_) => Type::Spatial,
         }
     }
 
@@ -249,7 +241,6 @@ impl FieldType {
             FieldType::Bytes(ref bytes_options) => bytes_options.is_indexed(),
             FieldType::JsonObject(ref json_object_options) => json_object_options.is_indexed(),
             FieldType::IpAddr(ref ip_addr_options) => ip_addr_options.is_indexed(),
-            FieldType::Spatial(ref _spatial_options) => true,
         }
     }
 
@@ -287,7 +278,6 @@ impl FieldType {
             FieldType::IpAddr(ref ip_addr_options) => ip_addr_options.is_fast(),
             FieldType::Facet(_) => true,
             FieldType::JsonObject(ref json_object_options) => json_object_options.is_fast(),
-            FieldType::Spatial(_) => false,
         }
     }
 
@@ -307,7 +297,6 @@ impl FieldType {
             FieldType::Bytes(ref bytes_options) => bytes_options.fieldnorms(),
             FieldType::JsonObject(ref _json_object_options) => false,
             FieldType::IpAddr(ref ip_addr_options) => ip_addr_options.fieldnorms(),
-            FieldType::Spatial(_) => false,
         }
     }
 
@@ -359,8 +348,6 @@ impl FieldType {
                     None
                 }
             }
-            FieldType::Spatial(_) => None, /* Geometry types cannot be indexed in the inverted
-                                            * index. */
         }
     }
 
@@ -462,10 +449,6 @@ impl FieldType {
 
                         Ok(OwnedValue::IpAddr(ip_addr.into_ipv6_addr()))
                     }
-                    FieldType::Spatial(_) => Err(ValueParsingError::TypeError {
-                        expected: "spatial field parsing not implemented",
-                        json: JsonValue::String(field_text),
-                    }),
                 }
             }
             JsonValue::Number(field_val_num) => match self {
@@ -525,10 +508,6 @@ impl FieldType {
                     expected: "a string with an ip addr",
                     json: JsonValue::Number(field_val_num),
                 }),
-                FieldType::Spatial(_) => Err(ValueParsingError::TypeError {
-                    expected: "spatial field parsing not implemented",
-                    json: JsonValue::Number(field_val_num),
-                }),
             },
             JsonValue::Object(json_map) => match self {
                 FieldType::Str(_) => {
@@ -544,14 +523,6 @@ impl FieldType {
                     }
                 }
                 FieldType::JsonObject(_) => Ok(OwnedValue::from(json_map)),
-                FieldType::Spatial(_) => Ok(OwnedValue::Geometry(
-                    Geometry::from_geojson(&json_map).map_err(|e| {
-                        ValueParsingError::ParseError {
-                            error: format!("{:?}", e),
-                            json: JsonValue::Object(json_map),
-                        }
-                    })?,
-                )),
                 _ => Err(ValueParsingError::TypeError {
                     expected: self.value_type().name(),
                     json: JsonValue::Object(json_map),

src/schema/flags.rs

@@ -1,6 +1,6 @@
 use std::ops::BitOr;
 
-use crate::schema::{DateOptions, NumericOptions, SpatialOptions, TextOptions};
+use crate::schema::{DateOptions, NumericOptions, TextOptions};
 
 #[derive(Clone)]
 pub struct StoredFlag;
@@ -95,14 +95,6 @@ impl<T: Clone + Into<TextOptions>> BitOr<TextOptions> for SchemaFlagList<T, ()>
     }
 }
 
-impl<T: Clone + Into<SpatialOptions>> BitOr<SpatialOptions> for SchemaFlagList<T, ()> {
-    type Output = SpatialOptions;
-
-    fn bitor(self, rhs: SpatialOptions) -> Self::Output {
-        self.head.into() | rhs
-    }
-}
-
 #[derive(Clone)]
 pub struct SchemaFlagList<Head: Clone, Tail: Clone> {
     pub head: Head,

src/schema/mod.rs

@@ -98,6 +98,10 @@
 //! make it possible to access the value given the doc id rapidly. This is useful if the value
 //! of the field is required during scoring or collection for instance.
 //!
+//! Some queries may leverage Fast fields when run on a field that is not indexed. This can be
+//! handy if that kind of request is infrequent, however note that searching on a Fast field is
+//! generally much slower than searching in an index.
+//!
 //! ```
 //! use tantivy::schema::*;
 //! let mut schema_builder = Schema::builder();
@@ -124,7 +128,6 @@ mod ip_options;
 mod json_object_options;
 mod named_field_document;
 mod numeric_options;
-mod spatial_options;
 mod text_options;
 
 use columnar::ColumnType;
@@ -145,7 +148,6 @@ pub use self::json_object_options::JsonObjectOptions;
 pub use self::named_field_document::NamedFieldDocument;
 pub use self::numeric_options::NumericOptions;
 pub use self::schema::{Schema, SchemaBuilder};
-pub use self::spatial_options::{SpatialOptions, SPATIAL};
 pub use self::term::{Term, ValueBytes};
 pub use self::text_options::{TextFieldIndexing, TextOptions, STRING, TEXT};
 
@@ -170,7 +172,6 @@ pub(crate) fn value_type_to_column_type(typ: Type) -> Option<ColumnType> {
         Type::Bytes => Some(ColumnType::Bytes),
         Type::IpAddr => Some(ColumnType::IpAddr),
         Type::Json => None,
-        Type::Spatial => None,
     }
 }
 

src/schema/schema.rs

@@ -194,16 +194,6 @@ impl SchemaBuilder {
         self.add_field(field_entry)
     }
 
-    /// Adds a spatial entry to the schema in build.
-    pub fn add_spatial_field<T: Into<SpatialOptions>>(
-        &mut self,
-        field_name: &str,
-        field_options: T,
-    ) -> Field {
-        let field_entry = FieldEntry::new_spatial(field_name.to_string(), field_options.into());
-        self.add_field(field_entry)
-    }
-
     /// Adds a field entry to the schema in build.
     pub fn add_field(&mut self, field_entry: FieldEntry) -> Field {
         let field = Field::from_field_id(self.fields.len() as u32);
@@ -218,14 +208,9 @@ impl SchemaBuilder {
     /// Finalize the creation of a `Schema`
     /// This will consume your `SchemaBuilder`
     pub fn build(self) -> Schema {
-        let contains_spatial_field = self
-            .fields
-            .iter()
-            .any(|field_entry| field_entry.field_type().value_type() == Type::Spatial);
         Schema(Arc::new(InnerSchema {
             fields: self.fields,
             fields_map: self.fields_map,
-            contains_spatial_field,
         }))
     }
 }
@@ -233,7 +218,6 @@ impl SchemaBuilder {
 struct InnerSchema {
     fields: Vec<FieldEntry>,
     fields_map: HashMap<String, Field>, // transient
-    contains_spatial_field: bool,
 }
 
 impl PartialEq for InnerSchema {
@@ -384,11 +368,6 @@ impl Schema {
         }
         Some((field, json_path))
     }
-
-    /// Returns true if the schema contains a spatial field.
-    pub(crate) fn contains_spatial_field(&self) -> bool {
-        self.0.contains_spatial_field
-    }
 }
 
 impl Serialize for Schema {
@@ -416,16 +395,16 @@ impl<'de> Deserialize<'de> for Schema {
 
             fn visit_seq<A>(self, mut seq: A) -> Result<Self::Value, A::Error>
             where A: SeqAccess<'de> {
-                let mut schema_builder = SchemaBuilder {
+                let mut schema = SchemaBuilder {
                     fields: Vec::with_capacity(seq.size_hint().unwrap_or(0)),
                     fields_map: HashMap::with_capacity(seq.size_hint().unwrap_or(0)),
                 };
 
                 while let Some(value) = seq.next_element()? {
-                    schema_builder.add_field(value);
+                    schema.add_field(value);
                 }
 
-                Ok(schema_builder.build())
+                Ok(schema.build())
             }
         }
 
@@ -1041,33 +1020,4 @@ mod tests {
             Some((default, "foobar"))
         );
     }
-
-    #[test]
-    fn test_contains_spatial_field() {
-        // No spatial field
-        {
-            let mut schema_builder = Schema::builder();
-            schema_builder.add_text_field("title", TEXT);
-            let schema = schema_builder.build();
-            assert!(!schema.contains_spatial_field());
-
-            // Serialization check
-            let schema_json = serde_json::to_string(&schema).unwrap();
-            let schema_deserialized: Schema = serde_json::from_str(&schema_json).unwrap();
-            assert!(!schema_deserialized.contains_spatial_field());
-        }
-        // With spatial field
-        {
-            let mut schema_builder = Schema::builder();
-            schema_builder.add_text_field("title", TEXT);
-            schema_builder.add_spatial_field("location", SPATIAL);
-            let schema = schema_builder.build();
-            assert!(schema.contains_spatial_field());
-
-            // Serialization check
-            let schema_json = serde_json::to_string(&schema).unwrap();
-            let schema_deserialized: Schema = serde_json::from_str(&schema_json).unwrap();
-            assert!(schema_deserialized.contains_spatial_field());
-        }
-    }
 }

src/schema/spatial_options.rs

@@ -1,53 +0,0 @@
-use std::ops::BitOr;
-
-use serde::{Deserialize, Serialize};
-
-use crate::schema::flags::StoredFlag;
-
-/// Define how a spatial field should be handled by tantivy.
-#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, Default)]
-pub struct SpatialOptions {
-    #[serde(default)]
-    stored: bool,
-}
-
-/// The field will be untokenized and indexed.
-pub const SPATIAL: SpatialOptions = SpatialOptions { stored: false };
-
-impl SpatialOptions {
-    /// Returns true if the geometry is to be stored.
-    #[inline]
-    pub fn is_stored(&self) -> bool {
-        self.stored
-    }
-}
-
-impl<T: Into<SpatialOptions>> BitOr<T> for SpatialOptions {
-    type Output = SpatialOptions;
-
-    fn bitor(self, other: T) -> SpatialOptions {
-        let other = other.into();
-        SpatialOptions {
-            stored: self.stored | other.stored,
-        }
-    }
-}
-
-impl From<StoredFlag> for SpatialOptions {
-    fn from(_: StoredFlag) -> SpatialOptions {
-        SpatialOptions { stored: true }
-    }
-}
-
-// #[cfg(test)]
-// mod tests {
-// use crate::schema::*;
-//
-// #[test]
-// fn test_field_options() {
-// let field_options = STORED | SPATIAL;
-// assert!(field_options.is_stored());
-// let mut schema_builder = Schema::builder();
-// schema_builder.add_spatial_index("where", SPATIAL | STORED);
-// }
-// }

src/schema/term.rs

@@ -1,10 +1,11 @@
-use std::hash::{Hash, Hasher};
+use std::hash::Hash;
 use std::net::Ipv6Addr;
 use std::{fmt, str};
 
 use columnar::MonotonicallyMappableToU128;
 use common::json_path_writer::{JSON_END_OF_PATH, JSON_PATH_SEGMENT_SEP_STR};
 use common::JsonPathWriter;
+use serde::{Deserialize, Serialize};
 
 use super::date_time_options::DATE_TIME_PRECISION_INDEXED;
 use super::{Field, Schema};
@@ -16,23 +17,54 @@ use crate::DateTime;
 /// Term represents the value that the token can take.
 /// It's a serialized representation over different types.
 ///
-/// It actually wraps a `Vec<u8>`. The first 5 bytes are metadata.
-/// 4 bytes are the field id, and the last byte is the type.
-///
-/// The serialized value `ValueBytes` is considered everything after the 4 first bytes (term id).
-#[derive(Clone)]
-pub struct Term<B = Vec<u8>>(B)
-where B: AsRef<[u8]>;
+/// A term is composed of Field and the serialized value bytes.
+/// The serialized value bytes themselves start with a one byte type tag followed by the payload.
+#[derive(Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Serialize, Deserialize)]
+pub struct Term {
+    field: Field,
+    serialized_value_bytes: Vec<u8>,
+}
 
-/// The number of bytes used as metadata by `Term`.
-const TERM_METADATA_LENGTH: usize = 5;
+/// The number of bytes used as metadata when serializing a term.
+const TERM_TYPE_TAG_LEN: usize = 1;
 
 impl Term {
+    /// Takes a serialized term and wraps it as a Term.
+    /// First 4 bytes are the field id
+    #[deprecated(
+        note = "we want to avoid working on the serialized representation directly, replace with \
+                typed API calls (add more if needed) or use serde to serialize/deserialize"
+    )]
+    pub fn wrap(serialized: &[u8]) -> Term {
+        let field_id_bytes: [u8; 4] = serialized[0..4].try_into().unwrap();
+        let field_id = u32::from_be_bytes(field_id_bytes);
+        Term {
+            field: Field::from_field_id(field_id),
+            serialized_value_bytes: serialized[4..].to_vec(),
+        }
+    }
+
+    /// Returns the serialized representation of the term.
+    /// First 4 bytes are the field id
+    #[deprecated(
+        note = "we want to avoid working on the serialized representation directly, replace with \
+                typed API calls (add more if needed) or use serde to serialize/deserialize"
+    )]
+    pub fn serialized_term(&self) -> Vec<u8> {
+        let mut serialized = Vec::with_capacity(4 + self.serialized_value_bytes.len());
+        serialized.extend(self.field.field_id().to_be_bytes().as_ref());
+        serialized.extend_from_slice(&self.serialized_value_bytes);
+        serialized
+    }
+
     /// Create a new Term with a buffer with a given capacity.
     pub fn with_capacity(capacity: usize) -> Term {
-        let mut data = Vec::with_capacity(TERM_METADATA_LENGTH + capacity);
-        data.resize(TERM_METADATA_LENGTH, 0u8);
-        Term(data)
+        let mut data = Vec::with_capacity(TERM_TYPE_TAG_LEN + capacity);
+        data.resize(TERM_TYPE_TAG_LEN, 0u8);
+        Term {
+            field: Field::from_field_id(0u32),
+            serialized_value_bytes: data,
+        }
     }
 
     /// Creates a term from a json path.
@@ -89,7 +121,7 @@ impl Term {
     fn with_bytes_and_field_and_payload(typ: Type, field: Field, bytes: &[u8]) -> Term {
         let mut term = Self::with_capacity(bytes.len());
         term.set_field_and_type(field, typ);
-        term.0.extend_from_slice(bytes);
+        term.serialized_value_bytes.extend_from_slice(bytes);
         term
     }
 
@@ -105,13 +137,13 @@ impl Term {
     /// Sets field and the type.
     pub(crate) fn set_field_and_type(&mut self, field: Field, typ: Type) {
         assert!(self.is_empty());
-        self.0[0..4].clone_from_slice(field.field_id().to_be_bytes().as_ref());
-        self.0[4] = typ.to_code();
+        self.field = field;
+        self.serialized_value_bytes[0] = typ.to_code();
     }
 
     /// Is empty if there are no value bytes.
     pub fn is_empty(&self) -> bool {
-        self.0.len() == TERM_METADATA_LENGTH
+        self.serialized_value_bytes.len() == TERM_TYPE_TAG_LEN
     }
 
     /// Builds a term given a field, and a `Ipv6Addr`-value
@@ -177,7 +209,7 @@ impl Term {
     /// Removes the value_bytes and set the type code.
     pub fn clear_with_type(&mut self, typ: Type) {
         self.truncate_value_bytes(0);
-        self.0[4] = typ.to_code();
+        self.serialized_value_bytes[0] = typ.to_code();
     }
 
     /// Append a type marker + fast value to a term.
@@ -185,9 +217,10 @@ impl Term {
     ///
     /// It will not clear existing bytes.
     pub fn append_type_and_fast_value<T: FastValue>(&mut self, val: T) {
-        self.0.push(T::to_type().to_code());
+        self.serialized_value_bytes.push(T::to_type().to_code());
         let value = val.to_u64();
-        self.0.extend(value.to_be_bytes().as_ref());
+        self.serialized_value_bytes
+            .extend(value.to_be_bytes().as_ref());
     }
 
     /// Append a string type marker + string to a term.
@@ -195,24 +228,25 @@ impl Term {
     ///
     /// It will not clear existing bytes.
     pub fn append_type_and_str(&mut self, val: &str) {
-        self.0.push(Type::Str.to_code());
-        self.0.extend(val.as_bytes().as_ref());
+        self.serialized_value_bytes.push(Type::Str.to_code());
+        self.serialized_value_bytes.extend(val.as_bytes().as_ref());
     }
 
     /// Sets the value of a `Bytes` field.
     pub fn set_bytes(&mut self, bytes: &[u8]) {
         self.truncate_value_bytes(0);
-        self.0.extend(bytes);
+        self.serialized_value_bytes.extend(bytes);
     }
 
     /// Truncates the value bytes of the term. Value and field type stays the same.
     pub fn truncate_value_bytes(&mut self, len: usize) {
-        self.0.truncate(len + TERM_METADATA_LENGTH);
+        self.serialized_value_bytes
+            .truncate(len + TERM_TYPE_TAG_LEN);
     }
 
     /// The length of the bytes.
     pub fn len_bytes(&self) -> usize {
-        self.0.len() - TERM_METADATA_LENGTH
+        self.serialized_value_bytes.len() - TERM_TYPE_TAG_LEN
     }
 
     /// Appends value bytes to the Term.
@@ -220,18 +254,9 @@ impl Term {
     /// This function returns the segment that has just been added.
     #[inline]
     pub fn append_bytes(&mut self, bytes: &[u8]) -> &mut [u8] {
-        let len_before = self.0.len();
-        self.0.extend_from_slice(bytes);
-        &mut self.0[len_before..]
-    }
-}
-
-impl<B> Term<B>
-where B: AsRef<[u8]>
-{
-    /// Wraps a object holding bytes
-    pub fn wrap(data: B) -> Term<B> {
-        Term(data)
+        let len_before = self.serialized_value_bytes.len();
+        self.serialized_value_bytes.extend_from_slice(bytes);
+        &mut self.serialized_value_bytes[len_before..]
     }
 
     /// Return the type of the term.
@@ -241,8 +266,7 @@ where B: AsRef<[u8]>
 
     /// Returns the field.
     pub fn field(&self) -> Field {
-        let field_id_bytes: [u8; 4] = (&self.0.as_ref()[..4]).try_into().unwrap();
-        Field::from_field_id(u32::from_be_bytes(field_id_bytes))
+        self.field
     }
 
     /// Returns the serialized representation of the value.
@@ -252,23 +276,13 @@ where B: AsRef<[u8]>
     /// If the term is a u64, its value is encoded according
     /// to `byteorder::BigEndian`.
     pub fn serialized_value_bytes(&self) -> &[u8] {
-        &self.0.as_ref()[TERM_METADATA_LENGTH..]
+        &self.serialized_value_bytes[TERM_TYPE_TAG_LEN..]
     }
 
     /// Returns the value of the term.
     /// address or JSON path + value. (this does not include the field.)
     pub fn value(&self) -> ValueBytes<&[u8]> {
-        ValueBytes::wrap(&self.0.as_ref()[4..])
-    }
-
-    /// Returns the serialized representation of Term.
-    /// This includes field_id, value type and value.
-    ///
-    /// Do NOT rely on this byte representation in the index.
-    /// This value is likely to change in the future.
-    #[inline]
-    pub fn serialized_term(&self) -> &[u8] {
-        self.0.as_ref()
+        ValueBytes::wrap(self.serialized_value_bytes.as_ref())
     }
 }
 
@@ -452,10 +466,7 @@ where B: AsRef<[u8]>
         }
     }
 
-    /// Returns the serialized representation of Term.
-    ///
-    /// Do NOT rely on this byte representation in the index.
-    /// This value is likely to change in the future.
+    /// Returns the serialized representation of the value bytes including the type tag.
     pub fn as_serialized(&self) -> &[u8] {
         self.0.as_ref()
     }
@@ -503,48 +514,11 @@ where B: AsRef<[u8]>
             Type::IpAddr => {
                 write_opt(f, self.as_ip_addr())?;
             }
-            Type::Spatial => {
-                write!(f, "<spatial term formatting not yet implemented>")?;
-            }
         }
         Ok(())
     }
 }
 
-impl<B> Ord for Term<B>
-where B: AsRef<[u8]>
-{
-    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
-        self.serialized_term().cmp(other.serialized_term())
-    }
-}
-
-impl<B> PartialOrd for Term<B>
-where B: AsRef<[u8]>
-{
-    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
-        Some(self.cmp(other))
-    }
-}
-
-impl<B> PartialEq for Term<B>
-where B: AsRef<[u8]>
-{
-    fn eq(&self, other: &Self) -> bool {
-        self.serialized_term() == other.serialized_term()
-    }
-}
-
-impl<B> Eq for Term<B> where B: AsRef<[u8]> {}
-
-impl<B> Hash for Term<B>
-where B: AsRef<[u8]>
-{
-    fn hash<H: Hasher>(&self, state: &mut H) {
-        self.0.as_ref().hash(state)
-    }
-}
-
 fn write_opt<T: std::fmt::Debug>(f: &mut fmt::Formatter, val_opt: Option<T>) -> fmt::Result {
     if let Some(val) = val_opt {
         write!(f, "{val:?}")?;
@@ -552,13 +526,11 @@ fn write_opt<T: std::fmt::Debug>(f: &mut fmt::Formatter, val_opt: Option<T>) ->
     Ok(())
 }
 
-impl<B> fmt::Debug for Term<B>
-where B: AsRef<[u8]>
-{
+impl fmt::Debug for Term {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        let field_id = self.field().field_id();
+        let field_id = self.field.field_id();
         write!(f, "Term(field={field_id}, ")?;
-        let value_bytes = ValueBytes::wrap(&self.0.as_ref()[4..]);
+        let value_bytes = ValueBytes::wrap(&self.serialized_value_bytes);
         value_bytes.debug_value_bytes(f)?;
         write!(f, ")",)?;
         Ok(())
@@ -581,17 +553,6 @@ mod tests {
         assert_eq!(term.value().as_str(), Some("test"))
     }
 
-    /// Size (in bytes) of the buffer of a fast value (u64, i64, f64, or date) term.
-    /// <field> + <type byte> + <value len>
-    ///
-    /// - <field> is a big endian encoded u32 field id
-    /// - <type_byte>'s most significant bit expresses whether the term is a json term or not The
-    ///   remaining 7 bits are used to encode the type of the value. If this is a JSON term, the
-    ///   type is the type of the leaf of the json.
-    /// - <value> is,  if this is not the json term, a binary representation specific to the type.
-    ///   If it is a JSON Term, then it is prepended with the path that leads to this leaf value.
-    const FAST_VALUE_TERM_LEN: usize = 4 + 1 + 8;
-
     #[test]
     pub fn test_term_u64() {
         let mut schema_builder = Schema::builder();
@@ -599,7 +560,7 @@ mod tests {
         let term = Term::from_field_u64(count_field, 983u64);
         assert_eq!(term.field(), count_field);
         assert_eq!(term.typ(), Type::U64);
-        assert_eq!(term.serialized_term().len(), FAST_VALUE_TERM_LEN);
+        assert_eq!(term.serialized_value_bytes().len(), 8);
         assert_eq!(term.value().as_u64(), Some(983u64))
     }
 
@@ -610,7 +571,7 @@ mod tests {
         let term = Term::from_field_bool(bool_field, true);
         assert_eq!(term.field(), bool_field);
         assert_eq!(term.typ(), Type::Bool);
-        assert_eq!(term.serialized_term().len(), FAST_VALUE_TERM_LEN);
+        assert_eq!(term.serialized_value_bytes().len(), 8);
         assert_eq!(term.value().as_bool(), Some(true))
     }
 }

src/snippet/mod.rs

@@ -483,7 +483,7 @@ mod tests {
 
     use super::{collapse_overlapped_ranges, search_fragments, select_best_fragment_combination};
     use crate::query::QueryParser;
-    use crate::schema::{IndexRecordOption, Schema, TextFieldIndexing, TextOptions, TEXT};
+    use crate::schema::{Schema, TEXT};
     use crate::snippet::SnippetGenerator;
     use crate::tokenizer::{NgramTokenizer, SimpleTokenizer};
     use crate::Index;
@@ -727,8 +727,10 @@ Survey in 2016, 2017, and 2018."#;
         Ok(())
     }
 
+    #[cfg(feature = "stemmer")]
     #[test]
     fn test_snippet_generator() -> crate::Result<()> {
+        use crate::schema::{IndexRecordOption, TextFieldIndexing, TextOptions};
         let mut schema_builder = Schema::builder();
         let text_options = TextOptions::default().set_indexing_options(
             TextFieldIndexing::default()

src/space_usage/mod.rs

@@ -7,13 +7,14 @@
 //! storage-level details into consideration. For example, if your file system block size is 4096
 //! bytes, we can under-count actual resultant space usage by up to 4095 bytes per file.
 
-use std::collections::HashMap;
+use std::collections::btree_map::Entry;
+use std::collections::BTreeMap;
 
+use columnar::ColumnSpaceUsage;
 use common::ByteCount;
 use serde::{Deserialize, Serialize};
 
 use crate::index::SegmentComponent;
-use crate::schema::Field;
 
 /// Enum containing any of the possible space usage results for segment components.
 pub enum ComponentSpaceUsage {
@@ -69,7 +70,6 @@ pub struct SegmentSpaceUsage {
     positions: PerFieldSpaceUsage,
     fast_fields: PerFieldSpaceUsage,
     fieldnorms: PerFieldSpaceUsage,
-    spatial: PerFieldSpaceUsage,
 
     store: StoreSpaceUsage,
 
@@ -87,7 +87,6 @@ impl SegmentSpaceUsage {
         positions: PerFieldSpaceUsage,
         fast_fields: PerFieldSpaceUsage,
         fieldnorms: PerFieldSpaceUsage,
-        spatial: PerFieldSpaceUsage,
         store: StoreSpaceUsage,
         deletes: ByteCount,
     ) -> SegmentSpaceUsage {
@@ -96,7 +95,6 @@ impl SegmentSpaceUsage {
             + positions.total()
             + fast_fields.total()
             + fieldnorms.total()
-            + spatial.total()
             + store.total()
             + deletes;
         SegmentSpaceUsage {
@@ -106,7 +104,6 @@ impl SegmentSpaceUsage {
             positions,
             fast_fields,
             fieldnorms,
-            spatial,
             store,
             deletes,
             total,
@@ -125,7 +122,6 @@ impl SegmentSpaceUsage {
             Positions => PerField(self.positions().clone()),
             FastFields => PerField(self.fast_fields().clone()),
             FieldNorms => PerField(self.fieldnorms().clone()),
-            Spatial => PerField(self.spatial().clone()),
             Terms => PerField(self.termdict().clone()),
             SegmentComponent::Store => ComponentSpaceUsage::Store(self.store().clone()),
             SegmentComponent::TempStore => ComponentSpaceUsage::Store(self.store().clone()),
@@ -163,11 +159,6 @@ impl SegmentSpaceUsage {
         &self.fieldnorms
     }
 
-    /// Space usage for field norms
-    pub fn spatial(&self) -> &PerFieldSpaceUsage {
-        &self.spatial
-    }
-
     /// Space usage for stored documents
     pub fn store(&self) -> &StoreSpaceUsage {
         &self.store
@@ -222,17 +213,26 @@ impl StoreSpaceUsage {
 /// Multiple indexes are used to handle variable length things, where
 #[derive(Clone, Debug, Serialize, Deserialize)]
 pub struct PerFieldSpaceUsage {
-    fields: HashMap<Field, FieldUsage>,
+    fields: BTreeMap<String, FieldUsage>,
     total: ByteCount,
 }
 
 impl PerFieldSpaceUsage {
     pub(crate) fn new(fields: Vec<FieldUsage>) -> PerFieldSpaceUsage {
-        let total = fields.iter().map(FieldUsage::total).sum();
-        let field_usage_map: HashMap<Field, FieldUsage> = fields
-            .into_iter()
-            .map(|field_usage| (field_usage.field(), field_usage))
-            .collect();
+        let mut total = ByteCount::default();
+        let mut field_usage_map: BTreeMap<String, FieldUsage> = BTreeMap::new();
+        for field_usage in fields {
+            total += field_usage.total();
+            let field_name = field_usage.field_name().to_string();
+            match field_usage_map.entry(field_name) {
+                Entry::Vacant(entry) => {
+                    entry.insert(field_usage);
+                }
+                Entry::Occupied(mut entry) => {
+                    entry.get_mut().merge(field_usage);
+                }
+            }
+        }
         PerFieldSpaceUsage {
             fields: field_usage_map,
             total,
@@ -240,8 +240,8 @@ impl PerFieldSpaceUsage {
     }
 
     /// Per field space usage
-    pub fn fields(&self) -> impl Iterator<Item = (&Field, &FieldUsage)> {
-        self.fields.iter()
+    pub fn fields(&self) -> impl Iterator<Item = &FieldUsage> {
+        self.fields.values()
     }
 
     /// Bytes used by the represented file
@@ -256,20 +256,23 @@ impl PerFieldSpaceUsage {
 /// See documentation for [`PerFieldSpaceUsage`] for slightly more information.
 #[derive(Clone, Debug, Serialize, Deserialize)]
 pub struct FieldUsage {
-    field: Field,
+    field_name: String,
     num_bytes: ByteCount,
     /// A field can be composed of more than one piece.
     /// These pieces are indexed by arbitrary numbers starting at zero.
     /// `self.num_bytes` includes all of `self.sub_num_bytes`.
     sub_num_bytes: Vec<Option<ByteCount>>,
+    /// Space usage of the column for fast fields, if relevant.
+    column_space_usage: Option<ColumnSpaceUsage>,
 }
 
 impl FieldUsage {
-    pub(crate) fn empty(field: Field) -> FieldUsage {
+    pub(crate) fn empty(field_name: impl Into<String>) -> FieldUsage {
         FieldUsage {
-            field,
+            field_name: field_name.into(),
             num_bytes: Default::default(),
             sub_num_bytes: Vec::new(),
+            column_space_usage: None,
         }
     }
 
@@ -282,9 +285,14 @@ impl FieldUsage {
         self.num_bytes += size
     }
 
+    pub(crate) fn set_column_usage(&mut self, column_space_usage: ColumnSpaceUsage) {
+        self.num_bytes += column_space_usage.total_num_bytes();
+        self.column_space_usage = Some(column_space_usage);
+    }
+
     /// Field
-    pub fn field(&self) -> Field {
-        self.field
+    pub fn field_name(&self) -> &str {
+        &self.field_name
     }
 
     /// Space usage for each index
@@ -292,16 +300,64 @@ impl FieldUsage {
         &self.sub_num_bytes[..]
     }
 
+    /// Returns the number of bytes used by the column payload, if the field is columnar.
+    pub fn column_num_bytes(&self) -> Option<ByteCount> {
+        self.column_space_usage
+            .as_ref()
+            .map(ColumnSpaceUsage::column_num_bytes)
+    }
+
+    /// Returns the number of bytes used by the dictionary for dictionary-encoded columns.
+    pub fn dictionary_num_bytes(&self) -> Option<ByteCount> {
+        self.column_space_usage
+            .as_ref()
+            .and_then(ColumnSpaceUsage::dictionary_num_bytes)
+    }
+
+    /// Returns the space usage of the column, if any.
+    pub fn column_space_usage(&self) -> Option<&ColumnSpaceUsage> {
+        self.column_space_usage.as_ref()
+    }
+
     /// Total bytes used for this field in this context
     pub fn total(&self) -> ByteCount {
         self.num_bytes
     }
+
+    fn merge(&mut self, other: FieldUsage) {
+        assert_eq!(self.field_name, other.field_name);
+        self.num_bytes += other.num_bytes;
+        if other.sub_num_bytes.len() > self.sub_num_bytes.len() {
+            self.sub_num_bytes.resize(other.sub_num_bytes.len(), None);
+        }
+        for (idx, num_bytes_opt) in other.sub_num_bytes.into_iter().enumerate() {
+            if let Some(num_bytes) = num_bytes_opt {
+                match self.sub_num_bytes[idx] {
+                    Some(existing) => self.sub_num_bytes[idx] = Some(existing + num_bytes),
+                    None => self.sub_num_bytes[idx] = Some(num_bytes),
+                }
+            }
+        }
+        self.column_space_usage =
+            merge_column_space_usage(self.column_space_usage.take(), other.column_space_usage);
+    }
+}
+
+fn merge_column_space_usage(
+    left: Option<ColumnSpaceUsage>,
+    right: Option<ColumnSpaceUsage>,
+) -> Option<ColumnSpaceUsage> {
+    match (left, right) {
+        (Some(lhs), Some(rhs)) => Some(lhs.merge(&rhs)),
+        (Some(space), None) | (None, Some(space)) => Some(space),
+        (None, None) => None,
+    }
 }
 
 #[cfg(test)]
 mod test {
     use crate::index::Index;
-    use crate::schema::{Field, Schema, FAST, INDEXED, STORED, TEXT};
+    use crate::schema::{Schema, FAST, INDEXED, STORED, TEXT};
     use crate::space_usage::PerFieldSpaceUsage;
     use crate::{IndexWriter, Term};
 
@@ -317,17 +373,17 @@ mod test {
 
     fn expect_single_field(
         field_space: &PerFieldSpaceUsage,
-        field: &Field,
+        field: &str,
         min_size: u64,
         max_size: u64,
     ) {
         assert!(field_space.total() >= min_size);
         assert!(field_space.total() <= max_size);
         assert_eq!(
-            vec![(field, field_space.total())],
+            vec![(field.to_string(), field_space.total())],
             field_space
                 .fields()
-                .map(|(x, y)| (x, y.total()))
+                .map(|usage| (usage.field_name().to_string(), usage.total()))
                 .collect::<Vec<_>>()
         );
     }
@@ -337,6 +393,7 @@ mod test {
         let mut schema_builder = Schema::builder();
         let name = schema_builder.add_u64_field("name", FAST | INDEXED);
         let schema = schema_builder.build();
+        let field_name = schema.get_field_name(name).to_string();
         let index = Index::create_in_ram(schema);
 
         {
@@ -359,11 +416,11 @@ mod test {
 
         assert_eq!(4, segment.num_docs());
 
-        expect_single_field(segment.termdict(), &name, 1, 512);
-        expect_single_field(segment.postings(), &name, 1, 512);
+        expect_single_field(segment.termdict(), &field_name, 1, 512);
+        expect_single_field(segment.postings(), &field_name, 1, 512);
         assert_eq!(segment.positions().total(), 0);
-        expect_single_field(segment.fast_fields(), &name, 1, 512);
-        expect_single_field(segment.fieldnorms(), &name, 1, 512);
+        expect_single_field(segment.fast_fields(), &field_name, 1, 512);
+        expect_single_field(segment.fieldnorms(), &field_name, 1, 512);
         // TODO: understand why the following fails
         //        assert_eq!(0, segment.store().total());
         assert_eq!(segment.deletes(), 0);
@@ -375,6 +432,7 @@ mod test {
         let mut schema_builder = Schema::builder();
         let name = schema_builder.add_text_field("name", TEXT);
         let schema = schema_builder.build();
+        let field_name = schema.get_field_name(name).to_string();
         let index = Index::create_in_ram(schema);
 
         {
@@ -399,11 +457,11 @@ mod test {
 
         assert_eq!(4, segment.num_docs());
 
-        expect_single_field(segment.termdict(), &name, 1, 512);
-        expect_single_field(segment.postings(), &name, 1, 512);
-        expect_single_field(segment.positions(), &name, 1, 512);
+        expect_single_field(segment.termdict(), &field_name, 1, 512);
+        expect_single_field(segment.postings(), &field_name, 1, 512);
+        expect_single_field(segment.positions(), &field_name, 1, 512);
         assert_eq!(segment.fast_fields().total(), 0);
-        expect_single_field(segment.fieldnorms(), &name, 1, 512);
+        expect_single_field(segment.fieldnorms(), &field_name, 1, 512);
         // TODO: understand why the following fails
         //        assert_eq!(0, segment.store().total());
         assert_eq!(segment.deletes(), 0);
@@ -439,10 +497,15 @@ mod test {
         assert_eq!(4, segment.num_docs());
 
         assert_eq!(segment.termdict().total(), 0);
+        assert!(segment.termdict().fields().next().is_none());
         assert_eq!(segment.postings().total(), 0);
+        assert!(segment.postings().fields().next().is_none());
         assert_eq!(segment.positions().total(), 0);
+        assert!(segment.positions().fields().next().is_none());
         assert_eq!(segment.fast_fields().total(), 0);
+        assert!(segment.fast_fields().fields().next().is_none());
         assert_eq!(segment.fieldnorms().total(), 0);
+        assert!(segment.fieldnorms().fields().next().is_none());
         assert!(segment.store().total() > 0);
         assert!(segment.store().total() < 512);
         assert_eq!(segment.deletes(), 0);
@@ -454,6 +517,7 @@ mod test {
         let mut schema_builder = Schema::builder();
         let name = schema_builder.add_u64_field("name", INDEXED);
         let schema = schema_builder.build();
+        let field_name = schema.get_field_name(name).to_string();
         let index = Index::create_in_ram(schema);
 
         {
@@ -484,11 +548,11 @@ mod test {
 
         assert_eq!(2, segment_space_usage.num_docs());
 
-        expect_single_field(segment_space_usage.termdict(), &name, 1, 512);
-        expect_single_field(segment_space_usage.postings(), &name, 1, 512);
+        expect_single_field(segment_space_usage.termdict(), &field_name, 1, 512);
+        expect_single_field(segment_space_usage.postings(), &field_name, 1, 512);
         assert_eq!(segment_space_usage.positions().total(), 0u64);
         assert_eq!(segment_space_usage.fast_fields().total(), 0u64);
-        expect_single_field(segment_space_usage.fieldnorms(), &name, 1, 512);
+        expect_single_field(segment_space_usage.fieldnorms(), &field_name, 1, 512);
         assert!(segment_space_usage.deletes() > 0);
         Ok(())
     }

src/spatial/bkd.rs

@@ -1,853 +0,0 @@
-//! Block kd-tree spatial indexing for triangulated polygons.
-//!
-//! Implements an immutable bulk-loaded spatial index using recursive median partitioning on
-//! bounding box dimensions. Each leaf stores up to 512 triangles with delta-compressed coordinates
-//! and doc IDs. The tree provides three query types (intersects, within, contains) that use exact
-//! integer arithmetic for geometric predicates and accumulate results in bit sets for efficient
-//! deduplication across leaves.
-//!
-//! The serialized format stores compressed leaf pages followed by the tree structure (leaf and
-//! branch nodes), enabling zero-copy access through memory-mapped segments without upfront
-//! decompression.
-use std::io;
-use std::io::Write;
-
-use common::{BitSet, CountingWriter};
-
-use crate::directory::WritePtr;
-use crate::spatial::delta::{compress, decompress, Compressible};
-use crate::spatial::triangle::Triangle;
-
-#[derive(Clone, Copy)]
-struct SpreadSurvey {
-    min: i32,
-    max: i32,
-}
-
-impl SpreadSurvey {
-    fn survey(&mut self, value: i32) {
-        self.min = self.min.min(value);
-        self.max = self.max.max(value);
-    }
-    fn spread(&self) -> i32 {
-        self.max - self.min
-    }
-}
-
-impl Default for SpreadSurvey {
-    fn default() -> Self {
-        SpreadSurvey {
-            min: i32::MAX,
-            max: i32::MIN,
-        }
-    }
-}
-
-#[derive(Clone, Copy)]
-struct BoundingBoxSurvey {
-    bbox: [i32; 4],
-}
-
-impl BoundingBoxSurvey {
-    fn survey(&mut self, triangle: &Triangle) {
-        self.bbox[0] = triangle.words[0].min(self.bbox[0]);
-        self.bbox[1] = triangle.words[1].min(self.bbox[1]);
-        self.bbox[2] = triangle.words[2].max(self.bbox[2]);
-        self.bbox[3] = triangle.words[3].max(self.bbox[3]);
-    }
-    fn bbox(&self) -> [i32; 4] {
-        self.bbox.clone()
-    }
-}
-
-impl Default for BoundingBoxSurvey {
-    fn default() -> Self {
-        BoundingBoxSurvey {
-            bbox: [i32::MAX, i32::MAX, i32::MIN, i32::MIN],
-        }
-    }
-}
-
-enum BuildNode {
-    Branch {
-        bbox: [i32; 4],
-        left: Box<BuildNode>,
-        right: Box<BuildNode>,
-    },
-    Leaf {
-        bbox: [i32; 4],
-        pos: u64,
-        len: u16,
-    },
-}
-
-struct CompressibleTriangleI32<'a> {
-    triangles: &'a [Triangle],
-    dimension: usize,
-}
-
-impl<'a> CompressibleTriangleI32<'a> {
-    fn new(triangles: &'a [Triangle], dimension: usize) -> Self {
-        CompressibleTriangleI32 {
-            triangles,
-            dimension,
-        }
-    }
-}
-
-impl<'a> Compressible for CompressibleTriangleI32<'a> {
-    type Value = i32;
-    fn len(&self) -> usize {
-        self.triangles.len()
-    }
-    fn get(&self, i: usize) -> i32 {
-        self.triangles[i].words[self.dimension]
-    }
-}
-
-struct CompressibleTriangleDocID<'a> {
-    triangles: &'a [Triangle],
-}
-
-impl<'a> CompressibleTriangleDocID<'a> {
-    fn new(triangles: &'a [Triangle]) -> Self {
-        CompressibleTriangleDocID { triangles }
-    }
-}
-
-impl<'a> Compressible for CompressibleTriangleDocID<'a> {
-    type Value = u32;
-    fn len(&self) -> usize {
-        self.triangles.len()
-    }
-    fn get(&self, i: usize) -> u32 {
-        self.triangles[i].doc_id
-    }
-}
-
-// Leaf pages are first the count of triangles, followed by delta encoded doc_ids, followed by
-// the delta encoded words in order. We will then have the length of the page. We build a tree
-// after the pages with leaf nodes and branch nodes. Leaf nodes will contain the bounding box
-// of the leaf followed position and length of the page. The leaf node is a level of direction
-// to store the position and length of the page in a format that is easy to read directly from
-// the mapping.
-
-// We do not compress the tree nodes. We read them directly from the mapping.
-
-//
-fn write_leaf_pages(
-    triangles: &mut [Triangle],
-    write: &mut CountingWriter<WritePtr>,
-) -> io::Result<BuildNode> {
-    // If less than 512 triangles we are at a leaf, otherwise we still in the inner nodes.
-    if triangles.len() <= 512 {
-        let pos = write.written_bytes();
-        let mut spreads = [SpreadSurvey::default(); 4];
-        let mut bounding_box = BoundingBoxSurvey::default();
-        for triangle in triangles.iter() {
-            for i in 0..4 {
-                spreads[i].survey(triangle.words[i]);
-            }
-            bounding_box.survey(triangle);
-        }
-        let mut max_spread = spreads[0].spread();
-        let mut dimension = 0;
-        for i in 1..4 {
-            let current_spread = spreads[i].spread();
-            if current_spread > max_spread {
-                dimension = i;
-                max_spread = current_spread;
-            }
-        }
-        write.write_all(&(triangles.len() as u16).to_le_bytes())?;
-        triangles.sort_by_key(|t| t.words[dimension]);
-        compress(&CompressibleTriangleDocID::new(triangles), write)?;
-        let compressible = [
-            CompressibleTriangleI32::new(triangles, 0),
-            CompressibleTriangleI32::new(triangles, 1),
-            CompressibleTriangleI32::new(triangles, 2),
-            CompressibleTriangleI32::new(triangles, 3),
-            CompressibleTriangleI32::new(triangles, 4),
-            CompressibleTriangleI32::new(triangles, 5),
-            CompressibleTriangleI32::new(triangles, 6),
-        ];
-        for i in 0..7 {
-            compress(&compressible[i], write)?;
-        }
-        let len = write.written_bytes() - pos;
-        Ok(BuildNode::Leaf {
-            bbox: bounding_box.bbox(),
-            pos,
-            len: len as u16,
-        })
-    } else {
-        let mut spreads = [SpreadSurvey::default(); 4];
-        let mut bounding_box = BoundingBoxSurvey::default();
-        for triangle in triangles.iter() {
-            for i in 0..4 {
-                spreads[i].survey(triangle.words[i]);
-            }
-            bounding_box.survey(triangle);
-        }
-        let mut max_spread = spreads[0].spread();
-        let mut dimension = 0;
-        for i in 0..4 {
-            let current_spread = spreads[i].spread();
-            if current_spread > max_spread {
-                dimension = i;
-                max_spread = current_spread;
-            }
-        }
-        // Partition the triangles.
-        let mid = triangles.len() / 2;
-        triangles.select_nth_unstable_by_key(mid, |t| t.words[dimension]);
-        let partition = triangles[mid].words[dimension];
-        let mut split_point = mid + 1;
-        while split_point < triangles.len() && triangles[split_point].words[dimension] == partition
-        {
-            split_point += 1;
-        }
-        // If we reached the end of triangles then all of the triangles share the partition value
-        // for the dimension. We handle this degeneracy by splitting at the midpoint so that we
-        // won't have a leaf with zero triangles.
-        if split_point == triangles.len() {
-            split_point = mid; // Force split at midpoint index
-        } else {
-            // Our partition does not sort the triangles, it only partitions. We have scan our right
-            // partition to find all the midpoint values and move them to the left partition.
-            let mut reverse = triangles.len() - 1;
-            loop {
-                // Scan backwards looking for the partition value.
-                while triangles[reverse].words[dimension] != partition {
-                    reverse -= 1;
-                }
-                // If we have reached the split point then we are done.
-                if reverse <= split_point {
-                    break;
-                }
-                // Swap the midpoint value with our current split point.
-                triangles.swap(split_point, reverse);
-                // Move the split point up one.
-                split_point += 1;
-                // We know that what was at the split point was not the midpoint value.
-                reverse -= 1;
-            }
-        }
-        // Split into left and write partitions and create child nodes.
-        let (left, right) = triangles.split_at_mut(split_point);
-        let left_node = write_leaf_pages(left, write)?;
-        let right_node = write_leaf_pages(right, write)?;
-        // Return an inner node.
-        Ok(BuildNode::Branch {
-            bbox: bounding_box.bbox(),
-            left: Box::new(left_node),
-            right: Box::new(right_node),
-        })
-    }
-}
-
-fn write_leaf_nodes(node: &BuildNode, write: &mut CountingWriter<WritePtr>) -> io::Result<()> {
-    match node {
-        BuildNode::Branch {
-            bbox: _,
-            left,
-            right,
-        } => {
-            write_leaf_nodes(right, write)?;
-            write_leaf_nodes(left, write)?;
-        }
-        BuildNode::Leaf { bbox, pos, len } => {
-            for &dimension in bbox.iter() {
-                write.write_all(&dimension.to_le_bytes())?;
-            }
-            write.write_all(&pos.to_le_bytes())?;
-            write.write_all(&len.to_le_bytes())?;
-            write.write_all(&[0u8; 6])?;
-        }
-    }
-    Ok(())
-}
-
-fn write_branch_nodes(
-    node: &BuildNode,
-    branch_offset: &mut i32,
-    leaf_offset: &mut i32,
-    write: &mut CountingWriter<WritePtr>,
-) -> io::Result<i32> {
-    match node {
-        BuildNode::Leaf { .. } => {
-            let pos = *leaf_offset;
-            *leaf_offset -= 1;
-            Ok(pos * size_of::<LeafNode>() as i32)
-        }
-        BuildNode::Branch { bbox, left, right } => {
-            let left = write_branch_nodes(left, branch_offset, leaf_offset, write)?;
-            let right = write_branch_nodes(right, branch_offset, leaf_offset, write)?;
-            for &val in bbox {
-                write.write_all(&val.to_le_bytes())?;
-            }
-            write.write_all(&left.to_le_bytes())?;
-            write.write_all(&right.to_le_bytes())?;
-            write.write_all(&[0u8; 8])?;
-            let pos = *branch_offset;
-            *branch_offset += 1;
-            Ok(pos * size_of::<BranchNode>() as i32)
-        }
-    }
-}
-
-const VERSION: u16 = 1u16;
-
-/// Builds and serializes a block kd-tree for spatial indexing of triangles.
-///
-/// Takes a collection of triangles and constructs a complete block kd-tree, writing both the
-/// compressed leaf pages and tree structure to the output. The tree uses recursive median
-/// partitioning on the dimension with maximum spread, storing up to 512 triangles per leaf.
-///
-/// The output format consists of:
-/// - Version header (u16)
-/// - Compressed leaf pages (delta-encoded doc_ids and triangle coordinates)
-/// - 32-byte aligned tree structure (leaf nodes, then branch nodes)
-/// - Footer with triangle count, root offset, and branch position
-///
-/// The `triangles` slice will be reordered during tree construction as partitioning sorts by the
-/// selected dimension at each level.
-pub fn write_block_kd_tree(
-    triangles: &mut [Triangle],
-    write: &mut CountingWriter<WritePtr>,
-) -> io::Result<()> {
-    write.write_all(&VERSION.to_le_bytes())?;
-
-    let tree = write_leaf_pages(triangles, write)?;
-    let current = write.written_bytes();
-    let aligned = current.next_multiple_of(32);
-    let padding = aligned - current;
-    write.write_all(&vec![0u8; padding as usize])?;
-
-    write_leaf_nodes(&tree, write)?;
-    let branch_position = write.written_bytes();
-    let mut branch_offset: i32 = 0;
-    let mut leaf_offset: i32 = -1;
-    let root = write_branch_nodes(&tree, &mut branch_offset, &mut leaf_offset, write)?;
-    write.write_all(&[0u8; 12])?;
-    write.write_all(&triangles.len().to_le_bytes())?;
-    write.write_all(&root.to_le_bytes())?;
-    write.write_all(&branch_position.to_le_bytes())?;
-    Ok(())
-}
-
-fn decompress_leaf(mut data: &[u8]) -> io::Result<Vec<Triangle>> {
-    use common::BinarySerializable;
-    let triangle_count: usize = u16::deserialize(&mut data)? as usize;
-    let mut offset: usize = 0;
-    let mut triangles: Vec<Triangle> = Vec::with_capacity(triangle_count);
-    offset += decompress::<u32, _>(&data[offset..], triangle_count, |_, doc_id| {
-        triangles.push(Triangle::skeleton(doc_id))
-    })?;
-    for i in 0..7 {
-        offset += decompress::<i32, _>(&data[offset..], triangle_count, |j, word| {
-            triangles[j].words[i] = word
-        })?;
-    }
-    Ok(triangles)
-}
-
-#[repr(C)]
-struct BranchNode {
-    bbox: [i32; 4],
-    left: i32,
-    right: i32,
-    pad: [u8; 8],
-}
-
-#[repr(C)]
-struct LeafNode {
-    bbox: [i32; 4],
-    pos: u64,
-    len: u16,
-    pad: [u8; 6],
-}
-
-/// A read-only view into a serialized block kd-tree segment.
-///
-/// Provides access to the tree structure and compressed leaf data through memory-mapped or
-/// buffered byte slices. The segment contains compressed leaf pages followed by the tree structure
-/// (leaf nodes and branch nodes), with a footer containing metadata for locating the root and
-/// interpreting offsets.
-pub struct Segment<'a> {
-    data: &'a [u8],
-    branch_position: u64,
-    /// Offset to the root of the tree, used as the starting point for traversal.
-    pub root_offset: i32,
-}
-
-impl<'a> Segment<'a> {
-    /// Creates a new segment from serialized block kd-tree data.
-    ///
-    /// Reads the footer metadata from the last 12 bytes to locate the tree structure and root
-    /// node.
-    pub fn new(data: &'a [u8]) -> Self {
-        Segment {
-            data,
-            branch_position: u64::from_le_bytes(data[data.len() - 8..].try_into().unwrap()),
-            root_offset: i32::from_le_bytes(
-                data[data.len() - 12..data.len() - 8].try_into().unwrap(),
-            ),
-        }
-    }
-    #[inline(always)]
-    fn bounding_box(&self, offset: i32) -> [i32; 4] {
-        let byte_offset = (self.branch_position as i64 + offset as i64) as usize;
-        let bytes = &self.data[byte_offset..byte_offset + 16];
-        [
-            i32::from_le_bytes(bytes[0..4].try_into().unwrap()),
-            i32::from_le_bytes(bytes[4..8].try_into().unwrap()),
-            i32::from_le_bytes(bytes[8..12].try_into().unwrap()),
-            i32::from_le_bytes(bytes[12..16].try_into().unwrap()),
-        ]
-    }
-    #[inline(always)]
-    fn branch_node(&self, offset: i32) -> BranchNode {
-        let byte_offset = (self.branch_position as i64 + offset as i64) as usize;
-        let bytes = &self.data[byte_offset..byte_offset + 32];
-        BranchNode {
-            bbox: [
-                i32::from_le_bytes(bytes[0..4].try_into().unwrap()),
-                i32::from_le_bytes(bytes[4..8].try_into().unwrap()),
-                i32::from_le_bytes(bytes[8..12].try_into().unwrap()),
-                i32::from_le_bytes(bytes[12..16].try_into().unwrap()),
-            ],
-            left: i32::from_le_bytes(bytes[16..20].try_into().unwrap()),
-            right: i32::from_le_bytes(bytes[20..24].try_into().unwrap()),
-            pad: [0u8; 8],
-        }
-    }
-    #[inline(always)]
-    fn leaf_node(&self, offset: i32) -> LeafNode {
-        let byte_offset = (self.branch_position as i64 + offset as i64) as usize;
-        let bytes = &self.data[byte_offset..byte_offset + 32];
-        LeafNode {
-            bbox: [
-                i32::from_le_bytes(bytes[0..4].try_into().unwrap()),
-                i32::from_le_bytes(bytes[4..8].try_into().unwrap()),
-                i32::from_le_bytes(bytes[8..12].try_into().unwrap()),
-                i32::from_le_bytes(bytes[12..16].try_into().unwrap()),
-            ],
-            pos: u64::from_le_bytes(bytes[16..24].try_into().unwrap()),
-            len: u16::from_le_bytes(bytes[24..26].try_into().unwrap()),
-            pad: [0u8; 6],
-        }
-    }
-    fn leaf_page(&self, leaf_node: &LeafNode) -> &[u8] {
-        &self.data[(leaf_node.pos as usize)..(leaf_node.pos as usize + leaf_node.len as usize)]
-    }
-}
-
-fn collect_all_docs(segment: &Segment, offset: i32, result: &mut BitSet) -> io::Result<()> {
-    if offset < 0 {
-        let leaf_node = segment.leaf_node(offset);
-        let data = segment.leaf_page(&leaf_node);
-        let count = u16::from_le_bytes([data[0], data[1]]) as usize;
-        decompress::<u32, _>(&data[2..], count, |_, doc_id| result.insert(doc_id))?;
-    } else {
-        let branch_node = segment.branch_node(offset);
-        collect_all_docs(segment, branch_node.left, result)?;
-        collect_all_docs(segment, branch_node.right, result)?;
-    }
-    Ok(())
-}
-
-fn bbox_within(bbox: &[i32; 4], query: &[i32; 4]) -> bool {
-    bbox[0] >= query[0] &&  // min_y >= query_min_y
-        bbox[1] >= query[1] &&  // min_x >= query_min_x
-        bbox[2] <= query[2] &&  // max_y <= query_max_y
-        bbox[3] <= query[3] // max_x <= query_max_x
-}
-
-fn bbox_intersects(bbox: &[i32; 4], query: &[i32; 4]) -> bool {
-    !(bbox[2] < query[0] || bbox[0] > query[2] || bbox[3] < query[1] || bbox[1] > query[3])
-}
-
-/// Finds documents with triangles that intersect the query bounding box.
-///
-/// Traverses the tree starting at `offset` (typically `segment.root_offset`), pruning subtrees
-/// whose bounding boxes don't intersect the query. When a node's bbox is entirely within the
-/// query, all its documents are bulk-collected. Otherwise, individual triangles are tested using
-/// exact geometric predicates.
-///
-/// The query is `[min_y, min_x, max_y, max_x]` in integer coordinates. Documents are inserted into
-/// the `result` BitSet, which automatically deduplicates when the same document appears in
-/// multiple leaves.
-pub fn search_intersects(
-    segment: &Segment,
-    offset: i32,
-    query: &[i32; 4],
-    result: &mut BitSet,
-) -> io::Result<()> {
-    let bbox = segment.bounding_box(offset);
-    // bbox doesn't intersect query → skip entire subtree
-    if !bbox_intersects(&bbox, query) {
-    }
-    // bbox entirely within query → all triangles intersect
-    else if bbox_within(&bbox, query) {
-        collect_all_docs(segment, offset, result)?;
-    } else if offset < 0 {
-        // bbox crosses query → test each triangle
-        let leaf_node = segment.leaf_node(offset);
-        let triangles = decompress_leaf(segment.leaf_page(&leaf_node))?;
-        for triangle in &triangles {
-            if triangle_intersects(triangle, query) {
-                result.insert(triangle.doc_id); // BitSet deduplicates
-            }
-        }
-    } else {
-        let branch_node = segment.branch_node(offset);
-        // bbox crosses query → must check children
-        search_intersects(segment, branch_node.left, query, result)?;
-        search_intersects(segment, branch_node.right, query, result)?;
-    }
-    Ok(())
-}
-
-#[expect(clippy::too_many_arguments)]
-fn line_intersects_line(
-    x1: i32,
-    y1: i32,
-    x2: i32,
-    y2: i32,
-    x3: i32,
-    y3: i32,
-    x4: i32,
-    y4: i32,
-) -> bool {
-    // Cast to i128 to prevent overflow in coordinate arithmetic
-    let x1 = x1 as i128;
-    let y1 = y1 as i128;
-    let x2 = x2 as i128;
-    let y2 = y2 as i128;
-    let x3 = x3 as i128;
-    let y3 = y3 as i128;
-    let x4 = x4 as i128;
-    let y4 = y4 as i128;
-
-    // Proper segment-segment intersection test
-    let d = (x1 - x2) * (y3 - y4) - (y1 - y2) * (x3 - x4);
-    if d == 0 {
-        // parallel
-        return false;
-    }
-
-    let t = (x1 - x3) * (y3 - y4) - (y1 - y3) * (x3 - x4);
-    let u = -((x1 - x2) * (y1 - y3) - (y1 - y2) * (x1 - x3));
-
-    if d > 0 {
-        t >= 0 && t <= d && u >= 0 && u <= d
-    } else {
-        t <= 0 && t >= d && u <= 0 && u >= d
-    }
-}
-
-fn edge_intersects_bbox(x1: i32, y1: i32, x2: i32, y2: i32, bbox: &[i32; 4]) -> bool {
-    // Test against all 4 rectangle edges, bottom, right, top, left.
-    line_intersects_line(x1, y1, x2, y2, bbox[1], bbox[0], bbox[3], bbox[0])
-        || line_intersects_line(x1, y1, x2, y2, bbox[3], bbox[0], bbox[3], bbox[2])
-        || line_intersects_line(x1, y1, x2, y2, bbox[3], bbox[2], bbox[1], bbox[2])
-        || line_intersects_line(x1, y1, x2, y2, bbox[1], bbox[2], bbox[1], bbox[0])
-}
-
-fn edge_crosses_bbox(x1: i32, y1: i32, x2: i32, y2: i32, bbox: &[i32; 4]) -> bool {
-    // Edge has endpoint outside while other is inside (crosses boundary)
-    let p1_inside = y1 >= bbox[0] && x1 >= bbox[1] && y1 <= bbox[2] && x1 <= bbox[3];
-    let p2_inside = y2 >= bbox[0] && x2 >= bbox[1] && y2 <= bbox[2] && x2 <= bbox[3];
-    p1_inside != p2_inside
-}
-
-fn triangle_within(triangle: &Triangle, query: &[i32; 4]) -> bool {
-    let tri_bbox = &triangle.words[0..4];
-
-    // Triangle bbox entirely within query → WITHIN
-    if tri_bbox[0] >= query[0]
-        && tri_bbox[1] >= query[1]
-        && tri_bbox[2] <= query[2]
-        && tri_bbox[3] <= query[3]
-    {
-        return true;
-    }
-
-    // Triangle bbox entirely outside → NOT WITHIN
-    if tri_bbox[2] < query[0]
-        || tri_bbox[3] < query[1]
-        || tri_bbox[0] > query[2]
-        || tri_bbox[1] > query[3]
-    {
-        return false;
-    }
-
-    // Decode vertices.
-    let ([ay, ax, by, bx, cy, cx], [ab, bc, ca]) = triangle.decode();
-
-    // Check each edge - if boundary edge crosses query bbox, NOT WITHIN
-    if ab && edge_crosses_bbox(ax, ay, bx, by, query) {
-        return false;
-    }
-    if bc && edge_crosses_bbox(bx, by, cx, cy, query) {
-        return false;
-    }
-    if ca && edge_crosses_bbox(cx, cy, ax, ay, query) {
-        return false;
-    }
-
-    // No boundary edges cross out
-    true
-}
-
-#[expect(clippy::too_many_arguments)]
-fn point_in_triangle(
-    px: i32,
-    py: i32,
-    ax: i32,
-    ay: i32,
-    bx: i32,
-    by: i32,
-    cx: i32,
-    cy: i32,
-) -> bool {
-    let v0x = (cx - ax) as i128;
-    let v0y = (cy - ay) as i128;
-    let v1x = (bx - ax) as i128;
-    let v1y = (by - ay) as i128;
-    let v2x = (px - ax) as i128;
-    let v2y = (py - ay) as i128;
-
-    let dot00 = v0x * v0x + v0y * v0y;
-    let dot01 = v0x * v1x + v0y * v1y;
-    let dot02 = v0x * v2x + v0y * v2y;
-    let dot11 = v1x * v1x + v1y * v1y;
-    let dot12 = v1x * v2x + v1y * v2y;
-
-    let denom = dot00 * dot11 - dot01 * dot01;
-    if denom == 0 {
-        return false;
-    }
-
-    let u = dot11 * dot02 - dot01 * dot12;
-    let v = dot00 * dot12 - dot01 * dot02;
-
-    u >= 0 && v >= 0 && u + v <= denom
-}
-
-fn triangle_intersects(triangle: &Triangle, query: &[i32; 4]) -> bool {
-    let tri_bbox = &triangle.words[0..4];
-
-    // Quick reject: bboxes don't overlap
-    if tri_bbox[2] < query[0]
-        || tri_bbox[3] < query[1]
-        || tri_bbox[0] > query[2]
-        || tri_bbox[1] > query[3]
-    {
-        return false;
-    }
-
-    let ([ay, ax, by, bx, cy, cx], _) = triangle.decode();
-
-    // Any triangle vertex inside rectangle?
-    if (ax >= query[1] && ax <= query[3] && ay >= query[0] && ay <= query[2])
-        || (bx >= query[1] && bx <= query[3] && by >= query[0] && by <= query[2])
-        || (cx >= query[1] && cx <= query[3] && cy >= query[0] && cy <= query[2])
-    {
-        return true;
-    }
-
-    // Any rectangle corner inside triangle?
-    let corners = [
-        (query[1], query[0]), // min_x, min_y
-        (query[3], query[0]), // max_x, min_y
-        (query[3], query[2]), // max_x, max_y
-        (query[1], query[2]), // min_x, max_y
-    ];
-    for (x, y) in corners {
-        if point_in_triangle(x, y, ax, ay, bx, by, cx, cy) {
-            return true;
-        }
-    }
-
-    // Any triangle edge intersect rectangle edges?
-    edge_intersects_bbox(ax, ay, bx, by, query)
-        || edge_intersects_bbox(bx, by, cx, cy, query)
-        || edge_intersects_bbox(cx, cy, ax, ay, query)
-}
-
-/// Finds documents where all triangles are within the query bounding box.
-///
-/// Traverses the tree starting at `offset` (typically `segment.root_offset`), testing each
-/// triangle to determine if it lies entirely within the query bounds. Uses two `BitSet` instances
-/// to track state: `result` accumulates candidate documents, while `excluded` marks documents that
-/// have at least one triangle extending outside the query.
-///
-/// The query is `[min_y, min_x, max_y, max_x]` in integer coordinates. The final result is
-/// documents in `result` that are NOT in `excluded` - the caller must compute this difference.
-pub fn search_within(
-    segment: &Segment,
-    offset: i32,
-    query: &[i32; 4], // [min_y, min_x, max_y, max_x]
-    result: &mut BitSet,
-    excluded: &mut BitSet,
-) -> io::Result<()> {
-    let bbox = segment.bounding_box(offset);
-    if !bbox_intersects(&bbox, query) {
-    } else if offset < 0 {
-        let leaf_node = segment.leaf_node(offset);
-        // bbox crosses query → test each triangle
-        let triangles = decompress_leaf(segment.leaf_page(&leaf_node))?;
-        for triangle in &triangles {
-            if triangle_intersects(triangle, query) {
-                if excluded.contains(triangle.doc_id) {
-                    continue; // Already excluded
-                }
-                if triangle_within(triangle, query) {
-                    result.insert(triangle.doc_id);
-                } else {
-                    excluded.insert(triangle.doc_id);
-                }
-            }
-        }
-    } else {
-        let branch_node = segment.branch_node(offset);
-        search_within(segment, branch_node.left, query, result, excluded)?;
-        search_within(segment, branch_node.right, query, result, excluded)?;
-    }
-    Ok(())
-}
-
-enum ContainsRelation {
-    CANDIDATE, // Query might be contained
-    NOTWITHIN, // Query definitely not contained
-    DISJOINT,  // Triangle doesn't overlap query
-}
-
-fn triangle_contains_relation(triangle: &Triangle, query: &[i32; 4]) -> ContainsRelation {
-    let tri_bbox = &triangle.words[0..4];
-
-    if query[2] < tri_bbox[0]
-        || query[3] < tri_bbox[1]
-        || query[0] > tri_bbox[2]
-        || query[1] > tri_bbox[3]
-    {
-        return ContainsRelation::DISJOINT;
-    }
-
-    let ([ay, ax, by, bx, cy, cx], [ab, bc, ca]) = triangle.decode();
-
-    let corners = [
-        (query[1], query[0]),
-        (query[3], query[0]),
-        (query[3], query[2]),
-        (query[1], query[2]),
-    ];
-
-    let mut any_corner_inside = false;
-    for &(qx, qy) in &corners {
-        if point_in_triangle(qx, qy, ax, ay, bx, by, cx, cy) {
-            any_corner_inside = true;
-            break;
-        }
-    }
-
-    let ab_intersects = edge_intersects_bbox(ax, ay, bx, by, query);
-    let bc_intersects = edge_intersects_bbox(bx, by, cx, cy, query);
-    let ca_intersects = edge_intersects_bbox(cx, cy, ax, ay, query);
-
-    if (ab && edge_crosses_bbox(ax, ay, bx, by, query))
-        || (bc && edge_crosses_bbox(bx, by, cx, cy, query))
-        || (ca && edge_crosses_bbox(cx, cy, ax, ay, query))
-    {
-        return ContainsRelation::NOTWITHIN;
-    }
-
-    if any_corner_inside || ab_intersects || bc_intersects || ca_intersects {
-        return ContainsRelation::CANDIDATE;
-    }
-
-    ContainsRelation::DISJOINT
-}
-
-/// Finds documents whose polygons contain the query bounding box.
-///
-/// Traverses the tree starting at `offset` (typically `segment.root_offset`), testing each
-/// triangle using three-state logic: `CANDIDATE` (query might be contained), `NOTWITHIN` (boundary
-/// edge crosses query), or `DISJOINT` (no overlap). Only boundary edges are tested for crossing -
-/// internal tessellation edges are ignored.
-///
-/// The query is `[min_y, min_x, max_y, max_x]` in integer coordinates. Uses two `BitSet`
-/// instances: `result` accumulates candidates, `excluded` marks documents with disqualifying
-/// boundary crossings. The final result is documents in `result` that are NOT in `excluded`.
-pub fn search_contains(
-    segment: &Segment,
-    offset: i32,
-    query: &[i32; 4],
-    result: &mut BitSet,
-    excluded: &mut BitSet,
-) -> io::Result<()> {
-    let bbox = segment.bounding_box(offset);
-    if !bbox_intersects(&bbox, query) {
-    } else if offset < 0 {
-        let leaf_node = segment.leaf_node(offset);
-        // bbox crosses query → test each triangle
-        let triangles = decompress_leaf(segment.leaf_page(&leaf_node))?;
-        for triangle in &triangles {
-            if triangle_intersects(triangle, query) {
-                let doc_id = triangle.doc_id;
-                if excluded.contains(doc_id) {
-                    continue;
-                }
-                match triangle_contains_relation(triangle, query) {
-                    ContainsRelation::CANDIDATE => result.insert(doc_id),
-                    ContainsRelation::NOTWITHIN => excluded.insert(doc_id),
-                    ContainsRelation::DISJOINT => {}
-                }
-            }
-        }
-    } else {
-        let branch_node = segment.branch_node(offset);
-        search_contains(segment, branch_node.left, query, result, excluded)?;
-        search_contains(segment, branch_node.right, query, result, excluded)?;
-    }
-    Ok(())
-}
-
-/// HUSH
-pub struct LeafPageIterator<'a> {
-    segment: &'a Segment<'a>,
-    descent_stack: Vec<i32>,
-}
-
-impl<'a> LeafPageIterator<'a> {
-    /// HUSH
-    pub fn new(segment: &'a Segment<'a>) -> Self {
-        Self {
-            segment,
-            descent_stack: vec![segment.root_offset],
-        }
-    }
-}
-
-impl<'a> Iterator for LeafPageIterator<'a> {
-    type Item = io::Result<Vec<Triangle>>;
-
-    fn next(&mut self) -> Option<Self::Item> {
-        let offset = self.descent_stack.pop()?;
-        if offset < 0 {
-            let leaf_node = self.segment.leaf_node(offset);
-            let leaf_page = self.segment.leaf_page(&leaf_node);
-            match decompress_leaf(&leaf_page) {
-                Ok(triangles) => Some(Ok(triangles)),
-                Err(e) => Some(Err(e)),
-            }
-        } else {
-            let branch_node = self.segment.branch_node(offset);
-            self.descent_stack.push(branch_node.right);
-            self.descent_stack.push(branch_node.left);
-            self.next()
-        }
-    }
-}

src/spatial/delta.rs

@@ -1,300 +0,0 @@
-//! Delta compression for block kd-tree leaves.
-//!
-//! Delta compression with dimension-major bit-packing for block kd-tree leaves. Each leaf contains
-//! ≤512 triangles sorted by the split dimension (the dimension with maximum spread chosen during
-//! tree construction). We store all 512 values for dimension 0, then all for dimension 1, etc.,
-//! enabling tight bit-packing per dimension and better cache locality during decode.
-//!
-//! The split dimension is already optimal for compression. Since triangles in a leaf are spatially
-//! clustered, sorting by the max-spread dimension naturally orders them by proximity in all
-//! dimensions. Testing multiple sort orders would be wasted effort.
-//!
-//! Our encoding uses ~214 units/meter for latitude, ~107 units/meter for longitude (millimeter
-//! precision). A quarter-acre lot (32m × 32m) spans ~6,850 units across 512 sorted triangles = avg
-//! delta ~13 units = 4 bits. A baseball field (100m × 100m) is ~42 unit deltas = 6 bits. Even
-//! Russia-sized polygons (1000 km) average ~418,000 unit deltas = 19 bits. Time will tell if these
-//! numbers are anything to go by in practice.
-//!
-//! Our format for use with leaf-page triangles: First a count of triangles in the page, then the
-//! delta encoded doc_ids followed by delta encoding of each series of the triangle dimensions,
-//! followed by delta encoding of the flags. Creates eight parallel arrays from which triangles can
-//! be reconstructed.
-//!
-//! Note: Tantivy also has delta encoding in `sstable/src/delta.rs`, but that's for string
-//! dictionary compression (prefix sharing + vint deltas). This module uses bit-packing with zigzag
-//! encoding, which is optimal for our signed i32 spatial coordinates with small deltas. It uses
-//! the same basic algorithm to compress u32 doc_ids.
-use std::io::{self, Write};
-
-fn zigzag_encode(x: i32) -> u32 {
-    ((x << 1) ^ (x >> 31)) as u32
-}
-
-fn zigzag_decode(x: u32) -> i32 {
-    ((x >> 1) ^ (0u32.wrapping_sub(x & 1))) as i32
-}
-
-/// Trait for reading values by index during compression.
-///
-/// The `Compressible` trait allows `compress()` to work with two different data sources,
-/// `Vec<Triangle>` when indexing and memory mapped `Triangle` when merging. The compress function
-/// reads values on-demand via `get()`, computing deltas and bit-packing without intermediate
-/// allocations.
-pub trait Compressible {
-    /// The type of the values being compressed.
-    type Value: Copy;
-    /// Returns the number of values in this source.
-    fn len(&self) -> usize;
-    /// Returns the value at the given index.
-    fn get(&self, i: usize) -> Self::Value;
-}
-
-/// Operations for types that can be delta-encoded and bit-packed into four-byte words.
-pub trait DeltaEncoder: Copy {
-    /// Computes a zigzag-encoded delta between two values.
-    fn compute_delta(current: Self, previous: Self) -> u32;
-    /// Converts a value to little-endian bytes for storage.
-    fn to_le_bytes(value: Self) -> [u8; 4];
-}
-
-impl DeltaEncoder for i32 {
-    fn compute_delta(current: Self, previous: Self) -> u32 {
-        zigzag_encode(current.wrapping_sub(previous))
-    }
-    fn to_le_bytes(value: Self) -> [u8; 4] {
-        value.to_le_bytes()
-    }
-}
-
-// Delta encoding for u32 values using wrapping arithmetic and zigzag encoding.
-//
-// This handles arbitrary u32 document IDs that may be non-sequential or widely spaced. The
-// strategy uses wrapping subtraction followed by zigzag encoding:
-//
-// 1. wrapping_sub computes the difference modulo 2^32, producing a u32 result
-// 2. Cast to i32 reinterprets the bit pattern as signed (two's complement)
-// 3. zigzag_encode maps signed values to unsigned for efficient bit-packing:
-//    - Positive deltas (0, 1, 2...) encode to even numbers (0, 2, 4...)
-//    - Negative deltas (-1, -2, -3...) encode to odd numbers (1, 3, 5...)
-//
-// Example with large jump (doc_id 0 → 4,000,000,000):
-//   delta = 4_000_000_000u32.wrapping_sub(0) = 4_000_000_000u32
-//   as i32 = -294,967,296 (bit pattern preserved via two's complement)
-//   zigzag_encode(-294,967,296) = some u32 value
-//
-// During decompression, zigzag_decode returns the signed i32 delta, which is cast back to u32 and
-// added with wrapping_add. The bit pattern round-trips correctly because wrapping_add and
-// wrapping_sub are mathematical inverses modulo 2^32, making this encoding symmetric for the full
-// u32 range.
-impl DeltaEncoder for u32 {
-    fn compute_delta(current: Self, previous: Self) -> u32 {
-        zigzag_encode(current.wrapping_sub(previous) as i32)
-    }
-    fn to_le_bytes(value: Self) -> [u8; 4] {
-        value.to_le_bytes()
-    }
-}
-
-/// Compresses values from a `Compressible` source using delta encoding and bit-packing.
-///
-/// Computes signed deltas between consecutive values, zigzag encodes them, and determines the
-/// minimum bit width needed to represent all deltas. Writes a header (1 byte for bit width +
-/// 4 bytes for first value in little-endian), then bit-packs the remaining deltas.
-pub fn compress<T, W>(compressible: &T, write: &mut W) -> io::Result<()>
-where
-    T: Compressible,
-    T::Value: DeltaEncoder,
-    W: Write,
-{
-    let mut max_delta = 0u32;
-    for i in 1..compressible.len() {
-        let delta = T::Value::compute_delta(compressible.get(i), compressible.get(i - 1));
-        max_delta = max_delta.max(delta);
-    }
-    let bits = if max_delta == 0 {
-        0u32
-    } else {
-        32 - max_delta.leading_zeros() as u32
-    };
-    let mask = if bits == 32 {
-        u32::MAX
-    } else {
-        (1u32 << bits) - 1
-    };
-    write.write_all(&[bits as u8])?;
-    write.write_all(&T::Value::to_le_bytes(compressible.get(0)))?;
-    let mut buffer = 0u64;
-    let mut buffer_bits = 0u32;
-    for i in 1..compressible.len() {
-        let delta = T::Value::compute_delta(compressible.get(i), compressible.get(i - 1));
-        let value = delta & mask;
-        buffer = (buffer << bits) | (value as u64);
-        buffer_bits += bits;
-        while buffer_bits >= 8 {
-            buffer_bits -= 8;
-            write.write_all(&[(buffer >> buffer_bits) as u8])?;
-        }
-    }
-    if buffer_bits > 0 {
-        write.write_all(&[(buffer << (8 - buffer_bits)) as u8])?;
-    }
-    Ok(())
-}
-
-/// Operations needed to decompress delta-encoded values back to their original form.
-pub trait DeltaDecoder: Copy + Sized {
-    /// Converts from little-endian bytes to a value.
-    fn from_le_bytes(bytes: [u8; 4]) -> Self;
-    /// Applies a zigzag-decoded delta to reconstruct the next value.
-    fn apply_delta(value: Self, delta: u32) -> Self;
-}
-
-impl DeltaDecoder for i32 {
-    fn from_le_bytes(bytes: [u8; 4]) -> Self {
-        i32::from_le_bytes(bytes)
-    }
-    fn apply_delta(value: Self, delta: u32) -> Self {
-        value.wrapping_add(zigzag_decode(delta))
-    }
-}
-
-impl DeltaDecoder for u32 {
-    fn from_le_bytes(bytes: [u8; 4]) -> Self {
-        u32::from_le_bytes(bytes)
-    }
-    fn apply_delta(value: Self, delta: u32) -> Self {
-        value.wrapping_add(zigzag_decode(delta) as u32)
-    }
-}
-
-/// Decompresses bit-packed delta-encoded values from a byte slice.
-///
-/// Reads the header to get bit width and first value, then unpacks the bit-packed deltas, applies
-/// zigzag decoding, and reconstructs the original values by accumulating deltas.
-///
-/// Returns the count of bytes read from `data`.
-pub fn decompress<T: DeltaDecoder, F>(
-    data: &[u8],
-    count: usize,
-    mut process: F,
-) -> io::Result<usize>
-where
-    F: FnMut(usize, T),
-{
-    if data.len() < 5 {
-        return Err(io::Error::new(
-            io::ErrorKind::UnexpectedEof,
-            "truncated header",
-        ));
-    }
-    let bits = data[0] as u32;
-    let first = T::from_le_bytes([data[1], data[2], data[3], data[4]]);
-    process(0, first);
-    let mut offset = 5;
-    if bits == 0 {
-        // All deltas are zero - all values same as first
-        for i in 1..count {
-            process(i, first);
-        }
-        return Ok(offset);
-    }
-    let mut buffer = 0u64;
-    let mut buffer_bits = 0u32;
-    let mut prev = first;
-    for i in 1..count {
-        // Refill buffer with bytes
-        while buffer_bits < bits {
-            if offset >= data.len() {
-                return Err(io::Error::new(
-                    io::ErrorKind::UnexpectedEof,
-                    format!("expected {} values but only decoded {}", count, i - 1),
-                ));
-            }
-            buffer = (buffer << 8) | (data[offset] as u64);
-            offset += 1;
-            buffer_bits += 8;
-        }
-        if buffer_bits >= bits {
-            // Extract packed value
-            buffer_bits -= bits;
-            let encoded = ((buffer >> buffer_bits) & ((1u64 << bits) - 1)) as u32;
-            let value = T::apply_delta(prev, encoded);
-            process(i, value);
-            prev = value;
-        } else {
-            break;
-        }
-    }
-    Ok(offset)
-}
-
-#[cfg(test)]
-mod test {
-    use super::*;
-
-    pub struct CompressibleI32Vec {
-        vec: Vec<i32>,
-    }
-
-    impl CompressibleI32Vec {
-        fn new(vec: Vec<i32>) -> Self {
-            CompressibleI32Vec { vec }
-        }
-    }
-
-    impl Compressible for CompressibleI32Vec {
-        type Value = i32;
-        fn len(&self) -> usize {
-            return self.vec.len();
-        }
-        fn get(&self, i: usize) -> i32 {
-            return self.vec[i];
-        }
-    }
-
-    #[test]
-    fn test_spatial_delta_compress_decompress() {
-        let values = vec![
-            100000, 99975, 100050, 99980, 100100, 100025, 99950, 100150, 100075, 99925, 100200,
-            100100,
-        ];
-        let compressible = CompressibleI32Vec::new(values.clone());
-        let mut buffer = Vec::new();
-        compress(&compressible, &mut buffer).unwrap();
-        let mut vec = Vec::new();
-        decompress::<i32, _>(&buffer, values.len(), |_, value| vec.push(value)).unwrap();
-        assert_eq!(vec, values);
-    }
-
-    #[test]
-    fn test_spatial_delta_bad_header() {
-        let mut vec = Vec::new();
-        let result = decompress::<i32, _>(&[1, 2], 1, |_, value| vec.push(value));
-        assert!(result.is_err());
-    }
-
-    #[test]
-    fn test_spatial_delta_insufficient_data() {
-        let mut vec = Vec::new();
-        let result = decompress::<i32, _>(&[5, 0, 0, 0, 1], 12, |_, value| vec.push(value));
-        assert!(result.is_err());
-    }
-
-    #[test]
-    fn test_spatial_delta_single_item() {
-        let mut vec = Vec::new();
-        decompress::<i32, _>(&[5, 1, 0, 0, 0], 1, |_, value| vec.push(value)).unwrap();
-        assert_eq!(vec[0], 1);
-    }
-
-    #[test]
-    fn test_spatial_delta_zero_length_delta() {
-        let values = vec![1, 1, 1];
-        let compressible = CompressibleI32Vec::new(values.clone());
-        let mut buffer = Vec::new();
-        compress(&compressible, &mut buffer).unwrap();
-        let mut vec = Vec::new();
-        decompress::<i32, _>(&buffer, values.len(), |_, value| vec.push(value)).unwrap();
-        assert_eq!(vec, values);
-    }
-}

src/spatial/geometry.rs

@@ -1,490 +0,0 @@
-//! HUSH
-
-use std::io::{self, Read, Write};
-
-use common::{BinarySerializable, VInt};
-use serde_json::{json, Map, Value};
-
-use crate::spatial::point::GeoPoint;
-use crate::spatial::xor::{compress_f64, decompress_f64};
-
-/// HUSH
-#[derive(Debug)]
-pub enum GeometryError {
-    /// HUSH
-    MissingType,
-    /// HUSH
-    MissingField(String), // "expected array", "wrong nesting depth", etc
-    /// HUSH
-    UnsupportedType(String),
-    /// HUSH
-    InvalidCoordinate(String), // Can report the actual bad value
-    /// HUSH
-    InvalidStructure(String), // "expected array", "wrong nesting depth", etc
-}
-
-/// HUSH
-#[derive(Debug, Clone, PartialEq)]
-pub enum Geometry {
-    /// HUSH
-    Point(GeoPoint),
-    /// HUSH
-    MultiPoint(Vec<GeoPoint>),
-    /// HUSH
-    LineString(Vec<GeoPoint>),
-    /// HUSH
-    MultiLineString(Vec<Vec<GeoPoint>>),
-    /// HUSH
-    Polygon(Vec<Vec<GeoPoint>>),
-    /// HUSH
-    MultiPolygon(Vec<Vec<Vec<GeoPoint>>>),
-    /// HUSH
-    GeometryCollection(Vec<Self>),
-}
-
-impl Geometry {
-    /// HUSH
-    pub fn from_geojson(object: &Map<String, Value>) -> Result<Self, GeometryError> {
-        let geometry_type = object
-            .get("type")
-            .and_then(|v| v.as_str())
-            .ok_or(GeometryError::MissingType)?;
-        match geometry_type {
-            "Point" => {
-                let coordinates = get_coordinates(object)?;
-                let point = to_point(coordinates)?;
-                Ok(Geometry::Point(point))
-            }
-            "MultiPoint" => {
-                let coordinates = get_coordinates(object)?;
-                let multi_point = to_line_string(coordinates)?;
-                Ok(Geometry::MultiPoint(multi_point))
-            }
-            "LineString" => {
-                let coordinates = get_coordinates(object)?;
-                let line_string = to_line_string(coordinates)?;
-                if line_string.len() < 2 {
-                    return Err(GeometryError::InvalidStructure(
-                        "a line string contains at least 2 points".to_string(),
-                    ));
-                }
-                Ok(Geometry::LineString(line_string))
-            }
-            "MultiLineString" => {
-                let coordinates = get_coordinates(object)?;
-                let multi_line_string = to_multi_line_string(coordinates)?;
-                for line_string in &multi_line_string {
-                    if line_string.len() < 2 {
-                        return Err(GeometryError::InvalidStructure(
-                            "a line string contains at least 2 points".to_string(),
-                        ));
-                    }
-                }
-                Ok(Geometry::MultiLineString(multi_line_string))
-            }
-            "Polygon" => {
-                let coordinates = get_coordinates(object)?;
-                let polygon = to_multi_line_string(coordinates)?;
-                for ring in &polygon {
-                    if ring.len() < 3 {
-                        return Err(GeometryError::InvalidStructure(
-                            "a polygon ring contains at least 3 points".to_string(),
-                        ));
-                    }
-                }
-                Ok(Geometry::Polygon(polygon))
-            }
-            "MultiPolygon" => {
-                let mut result = Vec::new();
-                let multi_polygons = get_coordinates(object)?;
-                let multi_polygons =
-                    multi_polygons
-                        .as_array()
-                        .ok_or(GeometryError::InvalidStructure(
-                            "expected an array of polygons".to_string(),
-                        ))?;
-                for polygon in multi_polygons {
-                    let polygon = to_multi_line_string(polygon)?;
-                    for ring in &polygon {
-                        if ring.len() < 3 {
-                            return Err(GeometryError::InvalidStructure(
-                                "a polygon ring contains at least 3 points".to_string(),
-                            ));
-                        }
-                    }
-                    result.push(polygon);
-                }
-                Ok(Geometry::MultiPolygon(result))
-            }
-            "GeometriesCollection" => {
-                let geometries = object
-                    .get("geometries")
-                    .ok_or(GeometryError::MissingField("geometries".to_string()))?;
-                let geometries = geometries
-                    .as_array()
-                    .ok_or(GeometryError::InvalidStructure(
-                        "geometries is not an array".to_string(),
-                    ))?;
-                let mut result = Vec::new();
-                for geometry in geometries {
-                    let object = geometry.as_object().ok_or(GeometryError::InvalidStructure(
-                        "geometry is not an object".to_string(),
-                    ))?;
-                    result.push(Geometry::from_geojson(object)?);
-                }
-                Ok(Geometry::GeometryCollection(result))
-            }
-            _ => Err(GeometryError::UnsupportedType(geometry_type.to_string())),
-        }
-    }
-
-    /// Serialize the geometry to GeoJSON format.
-    /// https://fr.wikipedia.org/wiki/GeoJSON
-    pub fn to_geojson(&self) -> Map<String, Value> {
-        let mut map = Map::new();
-        match self {
-            Geometry::Point(point) => {
-                map.insert("type".to_string(), Value::String("Point".to_string()));
-                let coords = json!([point.lon, point.lat]);
-                map.insert("coordinates".to_string(), coords);
-            }
-            Geometry::MultiPoint(points) => {
-                map.insert("type".to_string(), Value::String("MultiPoint".to_string()));
-                let coords: Vec<Value> = points.iter().map(|p| json!([p.lon, p.lat])).collect();
-                map.insert("coordinates".to_string(), Value::Array(coords));
-            }
-            Geometry::LineString(line) => {
-                map.insert("type".to_string(), Value::String("LineString".to_string()));
-                let coords: Vec<Value> = line.iter().map(|p| json!([p.lon, p.lat])).collect();
-                map.insert("coordinates".to_string(), Value::Array(coords));
-            }
-            Geometry::MultiLineString(lines) => {
-                map.insert(
-                    "type".to_string(),
-                    Value::String("MultiLineString".to_string()),
-                );
-                let coords: Vec<Value> = lines
-                    .iter()
-                    .map(|line| Value::Array(line.iter().map(|p| json!([p.lon, p.lat])).collect()))
-                    .collect();
-                map.insert("coordinates".to_string(), Value::Array(coords));
-            }
-            Geometry::Polygon(rings) => {
-                map.insert("type".to_string(), Value::String("Polygon".to_string()));
-                let coords: Vec<Value> = rings
-                    .iter()
-                    .map(|ring| Value::Array(ring.iter().map(|p| json!([p.lon, p.lat])).collect()))
-                    .collect();
-                map.insert("coordinates".to_string(), Value::Array(coords));
-            }
-            Geometry::MultiPolygon(polygons) => {
-                map.insert(
-                    "type".to_string(),
-                    Value::String("MultiPolygon".to_string()),
-                );
-                let coords: Vec<Value> = polygons
-                    .iter()
-                    .map(|polygon| {
-                        Value::Array(
-                            polygon
-                                .iter()
-                                .map(|ring| {
-                                    Value::Array(
-                                        ring.iter().map(|p| json!([p.lon, p.lat])).collect(),
-                                    )
-                                })
-                                .collect(),
-                        )
-                    })
-                    .collect();
-                map.insert("coordinates".to_string(), Value::Array(coords));
-            }
-            Geometry::GeometryCollection(geometries) => {
-                map.insert(
-                    "type".to_string(),
-                    Value::String("GeometryCollection".to_string()),
-                );
-                let geoms: Vec<Value> = geometries
-                    .iter()
-                    .map(|g| Value::Object(g.to_geojson()))
-                    .collect();
-                map.insert("geometries".to_string(), Value::Array(geoms));
-            }
-        }
-        map
-    }
-}
-
-fn get_coordinates(object: &Map<String, Value>) -> Result<&Value, GeometryError> {
-    let coordinates = object
-        .get("coordinates")
-        .ok_or(GeometryError::MissingField("coordinates".to_string()))?;
-    Ok(coordinates)
-}
-
-fn to_point(value: &Value) -> Result<GeoPoint, GeometryError> {
-    let lonlat = value.as_array().ok_or(GeometryError::InvalidStructure(
-        "expected 2 element array pair of lon/lat".to_string(),
-    ))?;
-    if lonlat.len() != 2 {
-        return Err(GeometryError::InvalidStructure(
-            "expected 2 element array pair of lon/lat".to_string(),
-        ));
-    }
-    let lon = lonlat[0].as_f64().ok_or(GeometryError::InvalidCoordinate(
-        "longitude must be f64".to_string(),
-    ))?;
-    if !lon.is_finite() || !(-180.0..=180.0).contains(&lon) {
-        return Err(GeometryError::InvalidCoordinate(format!(
-            "invalid longitude: {}",
-            lon
-        )));
-    }
-    let lat = lonlat[1].as_f64().ok_or(GeometryError::InvalidCoordinate(
-        "latitude must be f64".to_string(),
-    ))?;
-    if !lat.is_finite() || !(-90.0..=90.0).contains(&lat) {
-        return Err(GeometryError::InvalidCoordinate(format!(
-            "invalid latitude: {}",
-            lat
-        )));
-    }
-    Ok(GeoPoint { lon, lat })
-}
-
-fn to_line_string(value: &Value) -> Result<Vec<GeoPoint>, GeometryError> {
-    let mut result = Vec::new();
-    let coordinates = value.as_array().ok_or(GeometryError::InvalidStructure(
-        "expected an array of lon/lat arrays".to_string(),
-    ))?;
-    for coordinate in coordinates {
-        result.push(to_point(coordinate)?);
-    }
-    Ok(result)
-}
-
-fn to_multi_line_string(value: &Value) -> Result<Vec<Vec<GeoPoint>>, GeometryError> {
-    let mut result = Vec::new();
-    let coordinates = value.as_array().ok_or(GeometryError::InvalidStructure(
-        "expected an array of an array of lon/lat arrays".to_string(),
-    ))?;
-    for coordinate in coordinates {
-        result.push(to_line_string(coordinate)?);
-    }
-    Ok(result)
-}
-
-impl BinarySerializable for Geometry {
-    fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
-        match self {
-            Geometry::Point(point) => {
-                0u8.serialize(writer)?;
-                point.lon.serialize(writer)?;
-                point.lat.serialize(writer)?;
-                Ok(())
-            }
-            Geometry::MultiPoint(points) => {
-                1u8.serialize(writer)?;
-                serialize_line_string(points, writer)
-            }
-            Geometry::LineString(line_string) => {
-                2u8.serialize(writer)?;
-                serialize_line_string(line_string, writer)
-            }
-            Geometry::MultiLineString(multi_line_string) => {
-                3u8.serialize(writer)?;
-                serialize_polygon(&multi_line_string[..], writer)
-            }
-            Geometry::Polygon(polygon) => {
-                4u8.serialize(writer)?;
-                serialize_polygon(polygon, writer)
-            }
-            Geometry::MultiPolygon(multi_polygon) => {
-                5u8.serialize(writer)?;
-                BinarySerializable::serialize(&VInt(multi_polygon.len() as u64), writer)?;
-                for polygon in multi_polygon {
-                    BinarySerializable::serialize(&VInt(polygon.len() as u64), writer)?;
-                    for ring in polygon {
-                        BinarySerializable::serialize(&VInt(ring.len() as u64), writer)?;
-                    }
-                }
-                let mut lon = Vec::new();
-                let mut lat = Vec::new();
-                for polygon in multi_polygon {
-                    for ring in polygon {
-                        for point in ring {
-                            lon.push(point.lon);
-                            lat.push(point.lat);
-                        }
-                    }
-                }
-                let lon = compress_f64(&lon);
-                let lat = compress_f64(&lat);
-                VInt(lon.len() as u64).serialize(writer)?;
-                writer.write_all(&lon)?;
-                VInt(lat.len() as u64).serialize(writer)?;
-                writer.write_all(&lat)?;
-                Ok(())
-            }
-            Geometry::GeometryCollection(geometries) => {
-                6u8.serialize(writer)?;
-                BinarySerializable::serialize(&VInt(geometries.len() as u64), writer)?;
-                for geometry in geometries {
-                    geometry.serialize(writer)?;
-                }
-                Ok(())
-            }
-        }
-    }
-
-    fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
-        let discriminant: u8 = BinarySerializable::deserialize(reader)?;
-        match discriminant {
-            0 => {
-                let lon = BinarySerializable::deserialize(reader)?;
-                let lat = BinarySerializable::deserialize(reader)?;
-                Ok(Geometry::Point(GeoPoint { lon, lat }))
-            }
-            1 => Ok(Geometry::MultiPoint(deserialize_line_string(reader)?)),
-            2 => Ok(Geometry::LineString(deserialize_line_string(reader)?)),
-            3 => Ok(Geometry::MultiLineString(deserialize_polygon(reader)?)),
-            4 => Ok(Geometry::Polygon(deserialize_polygon(reader)?)),
-            5 => {
-                let polygon_count = VInt::deserialize(reader)?.0 as usize;
-                let mut polygons = Vec::new();
-                let mut count = 0;
-                for _ in 0..polygon_count {
-                    let ring_count = VInt::deserialize(reader)?.0 as usize;
-                    let mut rings = Vec::new();
-                    for _ in 0..ring_count {
-                        let point_count = VInt::deserialize(reader)?.0 as usize;
-                        rings.push(point_count);
-                        count += point_count;
-                    }
-                    polygons.push(rings);
-                }
-                let lon_bytes: Vec<u8> = BinarySerializable::deserialize(reader)?;
-                let lat_bytes: Vec<u8> = BinarySerializable::deserialize(reader)?;
-                let lon = decompress_f64(&lon_bytes, count);
-                let lat = decompress_f64(&lat_bytes, count);
-                let mut multi_polygon = Vec::new();
-                let mut offset = 0;
-                for rings in polygons {
-                    let mut polygon = Vec::new();
-                    for point_count in rings {
-                        let mut ring = Vec::new();
-                        for _ in 0..point_count {
-                            ring.push(GeoPoint {
-                                lon: lon[offset],
-                                lat: lat[offset],
-                            });
-                            offset += 1;
-                        }
-                        polygon.push(ring);
-                    }
-                    multi_polygon.push(polygon);
-                }
-                Ok(Geometry::MultiPolygon(multi_polygon))
-            }
-            6 => {
-                let geometry_count = VInt::deserialize(reader)?.0 as usize;
-                let mut geometries = Vec::new();
-                for _ in 0..geometry_count {
-                    geometries.push(Geometry::deserialize(reader)?);
-                }
-                Ok(Geometry::GeometryCollection(geometries))
-            }
-            _ => Err(io::Error::new(
-                io::ErrorKind::InvalidData,
-                "invalid geometry type",
-            )),
-        }
-    }
-}
-
-fn serialize_line_string<W: Write + ?Sized>(line: &[GeoPoint], writer: &mut W) -> io::Result<()> {
-    BinarySerializable::serialize(&VInt(line.len() as u64), writer)?;
-    let mut lon = Vec::new();
-    let mut lat = Vec::new();
-    for point in line {
-        lon.push(point.lon);
-        lat.push(point.lat);
-    }
-    let lon = compress_f64(&lon);
-    let lat = compress_f64(&lat);
-    VInt(lon.len() as u64).serialize(writer)?;
-    writer.write_all(&lon)?;
-    VInt(lat.len() as u64).serialize(writer)?;
-    writer.write_all(&lat)?;
-    Ok(())
-}
-
-fn serialize_polygon<W: Write + ?Sized>(
-    line_string: &[Vec<GeoPoint>],
-    writer: &mut W,
-) -> io::Result<()> {
-    BinarySerializable::serialize(&VInt(line_string.len() as u64), writer)?;
-    for ring in line_string {
-        BinarySerializable::serialize(&VInt(ring.len() as u64), writer)?;
-    }
-    let mut lon: Vec<f64> = Vec::new();
-    let mut lat: Vec<f64> = Vec::new();
-    for ring in line_string {
-        for point in ring {
-            lon.push(point.lon);
-            lat.push(point.lat);
-        }
-    }
-    let lon: Vec<u8> = compress_f64(&lon);
-    let lat: Vec<u8> = compress_f64(&lat);
-    VInt(lon.len() as u64).serialize(writer)?;
-    writer.write_all(&lon)?;
-    VInt(lat.len() as u64).serialize(writer)?;
-    writer.write_all(&lat)?;
-    Ok(())
-}
-
-fn deserialize_line_string<R: Read>(reader: &mut R) -> io::Result<Vec<GeoPoint>> {
-    let point_count = VInt::deserialize(reader)?.0 as usize;
-    let lon_bytes: Vec<u8> = BinarySerializable::deserialize(reader)?;
-    let lat_bytes: Vec<u8> = BinarySerializable::deserialize(reader)?;
-    let lon: Vec<f64> = decompress_f64(&lon_bytes, point_count);
-    let lat: Vec<f64> = decompress_f64(&lat_bytes, point_count);
-    let mut line_string: Vec<GeoPoint> = Vec::new();
-    for offset in 0..point_count {
-        line_string.push(GeoPoint {
-            lon: lon[offset],
-            lat: lat[offset],
-        });
-    }
-    Ok(line_string)
-}
-
-fn deserialize_polygon<R: Read>(reader: &mut R) -> io::Result<Vec<Vec<GeoPoint>>> {
-    let ring_count = VInt::deserialize(reader)?.0 as usize;
-    let mut rings = Vec::new();
-    let mut count = 0;
-    for _ in 0..ring_count {
-        let point_count = VInt::deserialize(reader)?.0 as usize;
-        rings.push(point_count);
-        count += point_count;
-    }
-    let lon_bytes: Vec<u8> = BinarySerializable::deserialize(reader)?;
-    let lat_bytes: Vec<u8> = BinarySerializable::deserialize(reader)?;
-    let lon: Vec<f64> = decompress_f64(&lon_bytes, count);
-    let lat: Vec<f64> = decompress_f64(&lat_bytes, count);
-    let mut polygon: Vec<Vec<GeoPoint>> = Vec::new();
-    let mut offset = 0;
-    for point_count in rings {
-        let mut ring = Vec::new();
-        for _ in 0..point_count {
-            ring.push(GeoPoint {
-                lon: lon[offset],
-                lat: lat[offset],
-            });
-            offset += 1;
-        }
-        polygon.push(ring);
-    }
-    Ok(polygon)
-}

src/spatial/mod.rs

@@ -1,12 +0,0 @@
-//! Spatial module (implements a block kd-tree)
-
-pub mod bkd;
-pub mod delta;
-pub mod geometry;
-pub mod point;
-pub mod radix_select;
-pub mod reader;
-pub mod serializer;
-pub mod triangle;
-pub mod writer;
-pub mod xor;

src/spatial/point.rs

@@ -1,8 +0,0 @@
-/// A point in the geographical coordinate system.
-#[derive(Debug, Clone, Copy, PartialEq)]
-pub struct GeoPoint {
-    /// Longitude
-    pub lon: f64,
-    /// Latitude
-    pub lat: f64,
-}

src/spatial/radix_select.rs

@@ -1,122 +0,0 @@
-//! Radix selection for block kd-tree tree partitioning.
-//!
-//! Implements byte-wise histogram selection to find median values without comparisons, enabling
-//! efficient partitioning of spatial data during block kd-tree construction. Processes values
-//! through multiple passes, building histograms for each byte position after a common prefix,
-//! avoiding the need to sort or compare elements directly.
-
-/// Performs radix selection to find the median value without comparisons by building byte-wise
-/// histograms.
-pub struct RadixSelect {
-    histogram: [usize; 256],
-    prefix: Vec<u8>,
-    offset: usize,
-    nth: usize,
-}
-
-impl RadixSelect {
-    /// Creates a new radix selector for finding the nth element among values with a common prefix.
-    ///
-    /// The offset specifies how many matching elements appeared in previous buckets (from earlier
-    /// passes). The nth parameter is 0-indexed, so pass 31 to find the 32nd element (median of
-    /// 64).
-    pub fn new(prefix: Vec<u8>, offset: usize, nth: usize) -> Self {
-        RadixSelect {
-            histogram: [0; 256],
-            prefix,
-            offset,
-            nth,
-        }
-    }
-    /// Updates the histogram with a value if it matches the current prefix.
-    ///
-    /// Values that don't start with the prefix are ignored. For matching values, increments the
-    /// count for the byte at position `prefix.len()`.
-    pub fn update(&mut self, value: i32) {
-        let bytes = value.to_be_bytes();
-        if !bytes.starts_with(&self.prefix) {
-            return;
-        }
-        let byte = bytes[self.prefix.len()];
-        self.histogram[byte as usize] += 1;
-    }
-    /// Finds which bucket contains the nth element and returns the bucket value and offset.
-    ///
-    /// Returns a tuple of `(bucket_byte, count_before)` where bucket_byte is the value of the byte
-    /// that contains the nth element, and count_before is the number of elements in earlier
-    /// buckets (becomes the offset for the next pass).
-    pub fn nth(&self) -> (u8, usize) {
-        let mut count = self.offset;
-        for (bucket, &frequency) in self.histogram.iter().enumerate() {
-            if count + frequency > self.nth as usize {
-                return (bucket as u8, count);
-            }
-            count += frequency;
-        }
-        panic!("nth element {} not found in histogram", self.nth);
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn radix_selection() {
-        let dimensions = [
-            (
-                vec![
-                    0x10101010, 0x10101011, 0x10101012, 0x10101013, 0x10101014, 0x10101015,
-                    0x10101016, 0x10101017, 0x10101018, 0x10101019, 0x1010101A, 0x1010101B,
-                    0x1010101C, 0x1010101D, 0x1010101E, 0x1010101F, 0x10101020, 0x10101021,
-                    0x10101022, 0x10101023, 0x10101024, 0x10101025, 0x10101026, 0x10101027,
-                    0x10101028, 0x10101029, 0x1010102A, 0x1010102B, 0x1010102C, 0x1010102D,
-                    0x1010102E, 0x1010102F, 0x10101030, 0x10101031, 0x10101032, 0x10101033,
-                    0x10101034, 0x10101035, 0x10101036, 0x10101037, 0x10101038, 0x10101039,
-                    0x1010103A, 0x1010103B, 0x1010103C, 0x1010103D, 0x1010103E, 0x1010103F,
-                    0x10101040, 0x10101041, 0x10101042, 0x10101043, 0x10101044, 0x10101045,
-                    0x10101046, 0x10101047, 0x10101048, 0x10101049, 0x1010104A, 0x1010104B,
-                    0x1010104C, 0x1010104D, 0x1010104E, 0x1010104F,
-                ],
-                [(0x10, 0), (0x10, 0), (0x10, 0), (0x2F, 31)],
-            ),
-            (
-                vec![
-                    0x10101010, 0x10101011, 0x10101012, 0x10101013, 0x10101014, 0x10101015,
-                    0x10101016, 0x10101017, 0x10101018, 0x10101019, 0x1010101A, 0x1010101B,
-                    0x1010101C, 0x1010101D, 0x1010101E, 0x1010101F, 0x10101020, 0x10101021,
-                    0x10101022, 0x10101023, 0x10101024, 0x20101025, 0x20201026, 0x20301027,
-                    0x20401028, 0x20501029, 0x2060102A, 0x2070102B, 0x2080102C, 0x2090102D,
-                    0x20A0102E, 0x20B0102F, 0x20C01030, 0x20D01031, 0x20E01032, 0x20F01033,
-                    0x20F11034, 0x20F21035, 0x20F31036, 0x20F41037, 0x20F51038, 0x20F61039,
-                    0x3010103A, 0x3010103B, 0x3010103C, 0x3010103D, 0x3010103E, 0x3010103F,
-                    0x30101040, 0x30101041, 0x30101042, 0x30101043, 0x30101044, 0x30101045,
-                    0x30101046, 0x30101047, 0x30101048, 0x30101049, 0x3010104A, 0x3010104B,
-                    0x3010104C, 0x3010104D, 0x3010104E, 0x3010104F,
-                ],
-                [(0x20, 21), (0xB0, 31), (0x10, 31), (0x2F, 31)],
-            ),
-        ];
-        for (numbers, expected) in dimensions {
-            let mut offset = 0;
-            let mut prefix = Vec::new();
-            for i in 0..4 {
-                let mut radix_select = RadixSelect::new(prefix.clone(), offset, 31);
-                for &number in &numbers {
-                    radix_select.update(number);
-                }
-                let (byte, count) = radix_select.nth();
-                if i != 3 {
-                    assert_eq!(expected[i].0, byte);
-                    assert_eq!(expected[i].1, count);
-                }
-                prefix.push(byte);
-                offset = count;
-            }
-            let mut sorted = numbers.clone();
-            sorted.sort();
-            let radix_result = i32::from_be_bytes(prefix.as_slice().try_into().unwrap());
-            assert_eq!(radix_result, sorted[31]);
-        }
-    }
-}

src/spatial/reader.rs

@@ -1,70 +0,0 @@
-//! HUSH
-use std::io;
-use std::sync::Arc;
-
-use common::file_slice::FileSlice;
-use common::OwnedBytes;
-
-use crate::directory::CompositeFile;
-use crate::schema::Field;
-use crate::space_usage::PerFieldSpaceUsage;
-
-#[derive(Clone)]
-pub struct SpatialReaders {
-    data: Arc<CompositeFile>,
-}
-
-impl SpatialReaders {
-    pub fn empty() -> SpatialReaders {
-        SpatialReaders {
-            data: Arc::new(CompositeFile::empty()),
-        }
-    }
-
-    /// Creates a field norm reader.
-    pub fn open(file: FileSlice) -> crate::Result<SpatialReaders> {
-        let data = CompositeFile::open(&file)?;
-        Ok(SpatialReaders {
-            data: Arc::new(data),
-        })
-    }
-
-    /// Returns the FieldNormReader for a specific field.
-    pub fn get_field(&self, field: Field) -> crate::Result<Option<SpatialReader>> {
-        if let Some(file) = self.data.open_read(field) {
-            let spatial_reader = SpatialReader::open(file)?;
-            Ok(Some(spatial_reader))
-        } else {
-            Ok(None)
-        }
-    }
-
-    /// Return a break down of the space usage per field.
-    pub fn space_usage(&self) -> PerFieldSpaceUsage {
-        self.data.space_usage()
-    }
-
-    /// Returns a handle to inner file
-    pub fn get_inner_file(&self) -> Arc<CompositeFile> {
-        self.data.clone()
-    }
-}
-
-/// HUSH
-#[derive(Clone)]
-pub struct SpatialReader {
-    data: OwnedBytes,
-}
-
-impl SpatialReader {
-    /// Opens the spatial reader from a `FileSlice`. Returns `None` if the file is empty (no
-    /// spatial fields indexed.)
-    pub fn open(spatial_file: FileSlice) -> io::Result<SpatialReader> {
-        let data = spatial_file.read_bytes()?;
-        Ok(SpatialReader { data })
-    }
-    /// HUSH
-    pub fn get_bytes(&self) -> &[u8] {
-        self.data.as_ref()
-    }
-}

src/spatial/serializer.rs

@@ -1,37 +0,0 @@
-//! HUSH
-use std::io;
-use std::io::Write;
-
-use crate::directory::{CompositeWrite, WritePtr};
-use crate::schema::Field;
-use crate::spatial::bkd::write_block_kd_tree;
-use crate::spatial::triangle::Triangle;
-
-/// The fieldnorms serializer is in charge of
-/// the serialization of field norms for all fields.
-pub struct SpatialSerializer {
-    composite_write: CompositeWrite,
-}
-
-impl SpatialSerializer {
-    /// Create a composite file from the write pointer.
-    pub fn from_write(write: WritePtr) -> io::Result<SpatialSerializer> {
-        // just making room for the pointer to header.
-        let composite_write = CompositeWrite::wrap(write);
-        Ok(SpatialSerializer { composite_write })
-    }
-
-    /// Serialize the given field
-    pub fn serialize_field(&mut self, field: Field, triangles: &mut [Triangle]) -> io::Result<()> {
-        let write = self.composite_write.for_field(field);
-        write_block_kd_tree(triangles, write)?;
-        write.flush()?;
-        Ok(())
-    }
-
-    /// Clean up, flush, and close.
-    pub fn close(self) -> io::Result<()> {
-        self.composite_write.close()?;
-        Ok(())
-    }
-}

src/spatial/triangle.rs

@@ -1,515 +0,0 @@
-//! A triangle encoding with bounding box in the first four words for efficient spatial pruning.
-//!
-//! Encodes triangles with the bounding box in the first four words, enabling efficient spatial
-//! pruning during tree traversal without reconstructing the full triangle. The remaining words
-//! contain an additional vertex and packed reconstruction metadata, allowing exact triangle
-//! recovery when needed.
-
-use i_triangle::advanced::delaunay::IntDelaunay;
-use i_triangle::i_overlay::i_float::int::point::IntPoint;
-
-use crate::DocId;
-
-const MINY_MINX_MAXY_MAXX_Y_X: i32 = 0;
-const MINY_MINX_Y_X_MAXY_MAXX: i32 = 1;
-const MAXY_MINX_Y_X_MINY_MAXX: i32 = 2;
-const MAXY_MINX_MINY_MAXX_Y_X: i32 = 3;
-const Y_MINX_MINY_X_MAXY_MAXX: i32 = 4;
-const Y_MINX_MINY_MAXX_MAXY_X: i32 = 5;
-const MAXY_MINX_MINY_X_Y_MAXX: i32 = 6;
-const MINY_MINX_Y_MAXX_MAXY_X: i32 = 7;
-
-/// Converts geographic coordinates (WGS84 lat/lon) to integer spatial coordinates.
-///
-/// Maps the full globe to the i32 range using linear scaling:
-/// - Latitude: -90° to +90° → -2³¹ to +2³¹-1
-/// - Longitude: -180° to +180° → -2³¹ to +2³¹-1
-///
-/// Provides approximately 214 units/meter for latitude and 107 units/meter for longitude, giving
-/// millimeter-level precision. Uses `floor()` to ensure consistent quantization.
-///
-/// Returns `(y, x)` where y=latitude coordinate, x=longitude coordinate.
-pub fn latlon_to_point(lat: f64, lon: f64) -> (i32, i32) {
-    let y = (lat / (180.0 / (1i64 << 32) as f64)).floor() as i32;
-    let x = (lon / (360.0 / (1i64 << 32) as f64)).floor() as i32;
-    (y, x)
-}
-
-/// Creates a bounding box from two lat/lon corner coordinates.
-///
-/// Takes two arbitrary corner points and produces a normalized bounding box in the internal
-/// coordinate system. Automatically computes min/max for each dimension.
-///
-/// Returns `[min_y, min_x, max_y, max_x]` matching the internal storage format used throughout the
-/// block kd-tree and triangle encoding.
-pub fn latlon_to_bbox(lat1: f64, lon1: f64, lat2: f64, lon2: f64) -> [i32; 4] {
-    let (y1, x1) = latlon_to_point(lat1, lon1);
-    let (y2, x2) = latlon_to_point(lat2, lon2);
-    [y1.min(y2), x1.min(x2), y1.max(y2), x1.max(x2)]
-}
-
-/// A triangle encoded with bounding box in the first four words for efficient spatial pruning.
-///
-/// Encodes the bounding box, one vertex, boundary edge flags, and a reconstruction code that
-/// together allow exact triangle recovery while optimizing for spatial query performance. Finally,
-/// it contains the document id.
-#[repr(C)]
-#[derive(Debug)]
-pub struct Triangle {
-    /// The bounding box, one vertex, followed by a packed integer containing boundary edge flags
-    /// and a reconstruction code.
-    pub words: [i32; 7],
-    /// The id of the document associated with this triangle.
-    pub doc_id: DocId,
-}
-
-impl Triangle {
-    /// Encodes a triangle with the bounding box in the first four words for efficient spatial
-    /// pruning.
-    ///
-    /// Takes three vertices as `[y0, x0, y1, x1, y2, x2]` and edge boundary flags `[ab, bc, ca]`
-    /// indicating which edges are polygon boundaries. Returns a triangle struct with the bounding
-    /// box in the first four words as `[min_y, min_x, max_y, max_x]`. When decoded, the vertex
-    /// order may differ from the original input to `new()` due to normalized rotation.
-    ///
-    /// The edge boundary flags are here to express whether an edge is part of the boundaries
-    /// in the tesselation of the larger polygon it belongs to.
-    pub fn new(doc_id: u32, triangle: [i32; 6], boundaries: [bool; 3]) -> Self {
-        let mut ay = triangle[0];
-        let mut ax = triangle[1];
-        let mut by = triangle[2];
-        let mut bx = triangle[3];
-        let mut cy = triangle[4];
-        let mut cx = triangle[5];
-        let mut ab = boundaries[0];
-        let mut bc = boundaries[1];
-        let mut ca = boundaries[2];
-        // rotate edges and place minX at the beginning
-        if bx < ax || cx < ax {
-            let temp_x = ax;
-            let temp_y = ay;
-            let temp_boundary = ab;
-            if bx < cx {
-                ax = bx;
-                ay = by;
-                ab = bc;
-                bx = cx;
-                by = cy;
-                bc = ca;
-                cx = temp_x;
-                cy = temp_y;
-                ca = temp_boundary;
-            } else {
-                ax = cx;
-                ay = cy;
-                ab = ca;
-                cx = bx;
-                cy = by;
-                ca = bc;
-                bx = temp_x;
-                by = temp_y;
-                bc = temp_boundary;
-            }
-        } else if ax == bx && ax == cx {
-            // degenerated case, all points with same longitude
-            // we need to prevent that ax is in the middle (not part of the MBS)
-            if by < ay || cy < ay {
-                let temp_x = ax;
-                let temp_y = ay;
-                let temp_boundary = ab;
-                if by < cy {
-                    ax = bx;
-                    ay = by;
-                    ab = bc;
-                    bx = cx;
-                    by = cy;
-                    bc = ca;
-                    cx = temp_x;
-                    cy = temp_y;
-                    ca = temp_boundary;
-                } else {
-                    ax = cx;
-                    ay = cy;
-                    ab = ca;
-                    cx = bx;
-                    cy = by;
-                    ca = bc;
-                    bx = temp_x;
-                    by = temp_y;
-                    bc = temp_boundary;
-                }
-            }
-        }
-        // change orientation if clockwise (CW)
-        if !is_counter_clockwise(
-            IntPoint { y: ay, x: ax },
-            IntPoint { y: by, x: bx },
-            IntPoint { y: cy, x: cx },
-        ) {
-            // To change the orientation, we simply swap B and C.
-            let temp_x = bx;
-            let temp_y = by;
-            let temp_boundary = ab;
-            // ax and ay do not change, ab becomes bc
-            ab = ca;
-            bx = cx;
-            by = cy;
-            // bc does not change, ca becomes ab
-            cx = temp_x;
-            cy = temp_y;
-            ca = temp_boundary;
-        }
-        let min_x = ax;
-        let min_y = ay.min(by).min(cy);
-        let max_x = ax.max(bx).max(cx);
-        let max_y = ay.max(by).max(cy);
-        let (y, x, code) = if min_y == ay {
-            if max_y == by && max_x == bx {
-                (cy, cx, MINY_MINX_MAXY_MAXX_Y_X)
-            } else if max_y == cy && max_x == cx {
-                (by, bx, MINY_MINX_Y_X_MAXY_MAXX)
-            } else {
-                (by, cx, MINY_MINX_Y_MAXX_MAXY_X)
-            }
-        } else if max_y == ay {
-            if min_y == by && max_x == bx {
-                (cy, cx, MAXY_MINX_MINY_MAXX_Y_X)
-            } else if min_y == cy && max_x == cx {
-                (by, bx, MAXY_MINX_Y_X_MINY_MAXX)
-            } else {
-                (cy, bx, MAXY_MINX_MINY_X_Y_MAXX)
-            }
-        } else if max_x == bx && min_y == by {
-            (ay, cx, Y_MINX_MINY_MAXX_MAXY_X)
-        } else if max_x == cx && max_y == cy {
-            (ay, bx, Y_MINX_MINY_X_MAXY_MAXX)
-        } else {
-            panic!("Could not encode the provided triangle");
-        };
-        let boundaries_bits = (ab as i32) | ((bc as i32) << 1) | ((ca as i32) << 2);
-        let packed = code | (boundaries_bits << 3);
-        Triangle {
-            words: [min_y, min_x, max_y, max_x, y, x, packed],
-            doc_id: doc_id,
-        }
-    }
-
-    /// Builds a degenerated triangle degenerating for a single point.
-    /// All vertices are that point, and all vertices are boundaries.
-    pub fn from_point(doc_id: DocId, point_x: i32, point_y: i32) -> Triangle {
-        Triangle::new(
-            doc_id,
-            [point_y, point_x, point_y, point_x, point_y, point_x],
-            [true, true, true],
-        )
-    }
-
-    /// Builds a degenerated triangle for a segment.
-    /// Line segment AB is represented as the triangle ABA.
-    pub fn from_line_segment(doc_id: DocId, a_x: i32, a_y: i32, b_x: i32, b_y: i32) -> Triangle {
-        Triangle::new(doc_id, [a_y, a_x, b_y, b_x, a_y, a_x], [true, true, true])
-    }
-
-    /// Create a triangle with only the doc_id and the words initialized to zero.
-    ///
-    /// The doc_id and words in the field are delta-compressed as a series with the doc_id
-    /// serialized first. When we reconstruct the triangle we can first reconstruct skeleton
-    /// triangles with the doc_id series, then populate the words directly from the decompression
-    /// as we decompress each series.
-    ///
-    /// An immutable constructor would require that we decompress first into parallel `Vec`
-    /// instances, then loop through the count of triangles building triangles using a constructor
-    /// that takes all eight field values at once. This saves a copy, the triangle is the
-    /// decompression destination.
-    pub fn skeleton(doc_id: u32) -> Self {
-        Triangle {
-            doc_id: doc_id,
-            words: [0i32; 7],
-        }
-    }
-
-    /// Decodes the triangle back to vertex coordinates and boundary flags.
-    ///
-    /// Returns vertices as `[y0, x0, y1, x1, y2, x2]` in CCW order and boundary flags `[ab, bc,
-    /// ca]`. The vertex order may differ from the original input to `new()` due to normalized CCW
-    /// rotation.
-    pub fn decode(&self) -> ([i32; 6], [bool; 3]) {
-        let packed = self.words[6];
-        let code = packed & 7; // Lower 3 bits
-        let boundaries = [
-            (packed & (1 << 3)) != 0, // bit 3 = ab
-            (packed & (1 << 4)) != 0, // bit 4 = bc
-            (packed & (1 << 5)) != 0, // bit 5 = ca
-        ];
-        let (ay, ax, by, bx, cy, cx) = match code {
-            MINY_MINX_MAXY_MAXX_Y_X => (
-                self.words[0],
-                self.words[1],
-                self.words[2],
-                self.words[3],
-                self.words[4],
-                self.words[5],
-            ),
-            MINY_MINX_Y_X_MAXY_MAXX => (
-                self.words[0],
-                self.words[1],
-                self.words[4],
-                self.words[5],
-                self.words[2],
-                self.words[3],
-            ),
-            MAXY_MINX_Y_X_MINY_MAXX => (
-                self.words[2],
-                self.words[1],
-                self.words[4],
-                self.words[5],
-                self.words[0],
-                self.words[3],
-            ),
-            MAXY_MINX_MINY_MAXX_Y_X => (
-                self.words[2],
-                self.words[1],
-                self.words[0],
-                self.words[3],
-                self.words[4],
-                self.words[5],
-            ),
-            Y_MINX_MINY_X_MAXY_MAXX => (
-                self.words[4],
-                self.words[1],
-                self.words[0],
-                self.words[5],
-                self.words[2],
-                self.words[3],
-            ),
-            Y_MINX_MINY_MAXX_MAXY_X => (
-                self.words[4],
-                self.words[1],
-                self.words[0],
-                self.words[3],
-                self.words[2],
-                self.words[5],
-            ),
-            MAXY_MINX_MINY_X_Y_MAXX => (
-                self.words[2],
-                self.words[1],
-                self.words[0],
-                self.words[5],
-                self.words[4],
-                self.words[3],
-            ),
-            MINY_MINX_Y_MAXX_MAXY_X => (
-                self.words[0],
-                self.words[1],
-                self.words[4],
-                self.words[3],
-                self.words[2],
-                self.words[5],
-            ),
-            _ => panic!("Could not decode the provided triangle"),
-        };
-        ([ay, ax, by, bx, cy, cx], boundaries)
-    }
-
-    /// Returns the bounding box coordinates of the encoded triangle.
-    ///
-    /// Provides access to the bounding box `[min_y, min_x, max_y, max_x]` stored in the first four
-    /// words of the structure. The bounding box is stored first for efficient spatial pruning,
-    /// determining whether it is necessary to decode the triangle for precise intersection or
-    /// containment tests.
-    pub fn bbox(&self) -> &[i32] {
-        &self.words[..4]
-    }
-}
-
-/// Encodes the triangles of a Delaunay triangulation into block kd-tree triangles.
-///
-/// Takes the output of a Delaunay triangulation from `i_triangle` and encodes each triangle into
-/// the normalized triangle used by the block kd-tree. Each triangle includes its bounding box,
-/// vertex coordinates, and boundary edge flags that distinguish original polygon edges from
-/// internal tessellation edges.
-///
-/// The boundary edge information provided by the `i_triangle` Delaunay triangulation is essential
-/// for CONTAINS and WITHIN queries to work correctly.
-pub fn delaunay_to_triangles(doc_id: u32, delaunay: &IntDelaunay, triangles: &mut Vec<Triangle>) {
-    for triangle in delaunay.triangles.iter() {
-        let bounds = [
-            triangle.neighbors[0] == usize::MAX,
-            triangle.neighbors[1] == usize::MAX,
-            triangle.neighbors[2] == usize::MAX,
-        ];
-        let v0 = &delaunay.points[triangle.vertices[0].index];
-        let v1 = &delaunay.points[triangle.vertices[1].index];
-        let v2 = &delaunay.points[triangle.vertices[2].index];
-        triangles.push(Triangle::new(
-            doc_id,
-            [v0.y, v0.x, v1.y, v1.x, v2.y, v2.x],
-            bounds,
-        ))
-    }
-}
-
-/// Returns true if the path A -> B -> C is Counter-Clockwise (CCW) or collinear.
-/// Returns false if it is Clockwise (CW).
-#[inline(always)]
-fn is_counter_clockwise(a: IntPoint, b: IntPoint, c: IntPoint) -> bool {
-    // We calculate the 2D cross product (determinant) of vectors AB and AC.
-    // Formula: (bx - ax)(cy - ay) - (by - ay)(cx - ax)
-
-    // We cast to i64 to prevent overflow, as multiplying two i32s can exceed i32::MAX.
-    let val = (b.x as i64 - a.x as i64) * (c.y as i64 - a.y as i64)
-        - (b.y as i64 - a.y as i64) * (c.x as i64 - a.x as i64);
-
-    // If the result is positive, the triangle is CCW.
-    // If negative, it is CW.
-    // If zero, the points are collinear (we return true in that case).
-    val >= 0
-}
-
-#[cfg(test)]
-mod tests {
-    use i_triangle::i_overlay::i_float::int::point::IntPoint;
-    use i_triangle::int::triangulatable::IntTriangulatable;
-
-    use super::*;
-
-    #[test]
-    fn encode_triangle() {
-        let test_cases = [
-            ([1, 1, 3, 2, 2, 4], [true, false, false]),
-            ([1, 1, 2, 4, 3, 2], [false, false, true]),
-            ([2, 4, 1, 1, 3, 2], [false, true, false]),
-            ([2, 4, 3, 2, 1, 1], [false, true, false]),
-            ([3, 2, 1, 1, 2, 4], [true, false, false]),
-            ([3, 2, 2, 4, 1, 1], [false, false, true]),
-        ];
-        let ccw_coords = [1, 1, 2, 4, 3, 2];
-        let ccw_bounds = [false, false, true];
-        for (coords, bounds) in test_cases {
-            let triangle = Triangle::new(1, coords, bounds);
-            let (decoded_coords, decoded_bounds) = triangle.decode();
-            assert_eq!(decoded_coords, ccw_coords);
-            assert_eq!(decoded_bounds, ccw_bounds);
-        }
-    }
-
-    #[test]
-    fn test_cw_triangle_boundary_and_coord_flip() {
-        // 1. Define distinct coordinates for a Clockwise triangle
-        // Visual layout:
-        // A(50,40): Top Center-ish
-        // B(10,60): Bottom Right
-        // C(20,10): Bottom Left (Has the Minimum X=10)
-        // Path A->B->C is Clockwise.
-        let input_coords = [
-            50, 40, // A (y, x)
-            10, 60, // B
-            20, 10, // C
-        ];
-
-        // 2. Define Boundaries [ab, bc, ca]
-        // We set BC=true and CA=false.
-        // The bug (ab=bc) would erroneously put 'true' into the first slot.
-        // The fix (ab=ca) should put 'false' into the first slot.
-        let input_bounds = [false, true, false];
-
-        // 3. Encode
-        let triangle = Triangle::new(1, input_coords, input_bounds);
-        let (decoded_coords, decoded_bounds) = triangle.decode();
-
-        // 4. Expected Coordinates
-        // The internal logic detects CW, swaps B/C to make it CCW:
-        //    A(50,40) -> C(20,10) -> B(10,60)
-        // Then it rotates to put Min-X first.
-        // Min X is 10 (Vertex C).
-        // Final Sequence: C -> B -> A
-        let expected_coords = [
-            20, 10, // C
-            10, 60, // B
-            50, 40, // A
-        ];
-
-        // 5. Expected Boundaries
-        // After Flip (A->C->B):
-        //    Edge AC (was CA) = false
-        //    Edge CB (was BC) = true
-        //    Edge BA (was AB) = false
-        //    Unrotated: [false, true, false]
-        // After Rotation (shifting to start at C):
-        //    Shift left by 1: [true, false, false]
-        let expected_bounds = [true, false, false];
-
-        assert_eq!(
-            decoded_coords, expected_coords,
-            "Coordinates did not decode as expected"
-        );
-        assert_eq!(
-            decoded_bounds, expected_bounds,
-            "Boundary flags were incorrect (likely swap bug)"
-        );
-    }
-
-    #[test]
-    fn degenerate_triangle() {
-        let test_cases = [
-            (
-                [1, 1, 2, 1, 3, 1],
-                [true, false, false],
-                [1, 1, 2, 1, 3, 1],
-                [true, false, false],
-            ),
-            (
-                [2, 1, 1, 1, 3, 1],
-                [true, false, false],
-                [1, 1, 3, 1, 2, 1],
-                [false, false, true],
-            ),
-            (
-                [2, 1, 3, 1, 1, 1],
-                [false, false, true],
-                [1, 1, 2, 1, 3, 1],
-                [true, false, false],
-            ),
-        ];
-        for (coords, bounds, ccw_coords, ccw_bounds) in test_cases {
-            let triangle = Triangle::new(1, coords, bounds);
-            let (decoded_coords, decoded_bounds) = triangle.decode();
-            assert_eq!(decoded_coords, ccw_coords);
-            assert_eq!(decoded_bounds, ccw_bounds);
-        }
-    }
-
-    #[test]
-    fn decode_triangle() {
-        // distinct values for each coordinate to catch transposition
-        let test_cases = [
-            [11, 10, 60, 80, 41, 40],
-            [1, 0, 11, 20, 31, 30],
-            [30, 0, 11, 10, 1, 20],
-            [30, 0, 1, 20, 21, 11],
-            [20, 0, 1, 30, 41, 40],
-            [20, 0, 1, 30, 31, 10],
-            [30, 0, 1, 10, 11, 20],
-            [1, 0, 10, 20, 21, 11],
-        ];
-        for coords in test_cases {
-            let triangle = Triangle::new(1, coords, [true, true, true]);
-            let (decoded_coords, _) = triangle.decode();
-            assert_eq!(decoded_coords, coords);
-        }
-    }
-
-    #[test]
-    fn triangulate_box() {
-        let i_polygon = vec![vec![
-            IntPoint::new(0, 0),
-            IntPoint::new(10, 0),
-            IntPoint::new(10, 10),
-            IntPoint::new(0, 10),
-        ]];
-        let mut triangles = Vec::new();
-        let delaunay = i_polygon.triangulate().into_delaunay();
-        delaunay_to_triangles(1, &delaunay, &mut triangles);
-        assert_eq!(triangles.len(), 2);
-    }
-}