Added logging when token is dropped.

* Clippy comments

Clippy complaints that about the cast of &[u32] to a *const __m128i,
because of the lack of alignment constraints.

This commit passes the OutputBuffer object (which enforces proper
    alignment) instead of `&[u32]`.

* Clippy. Block alignment

* Code simplification

* Added comment. Code simplification

* Removed the extraneous freq block len hack.

.travis.yml

@@ -29,7 +29,7 @@ addons:
 matrix:
   include:
     # Android
-    - env: TARGET=aarch64-linux-android
+    - env: TARGET=aarch64-linux-android DISABLE_TESTS=1
     #- env: TARGET=arm-linux-androideabi DISABLE_TESTS=1
     #- env: TARGET=armv7-linux-androideabi DISABLE_TESTS=1
     #- env: TARGET=i686-linux-android DISABLE_TESTS=1
@@ -68,6 +68,11 @@ cache: cargo
 before_cache:
   # Travis can't cache files that are not readable by "others"
   - chmod -R a+r $HOME/.cargo
+  - find ./target/debug -type f -maxdepth 1 -delete
+  - rm -f  ./target/.rustc_info.json
+  - rm -fr ./target/debug/{deps,.fingerprint}/tantivy*
+  - rm -r target/debug/examples/
+  - ls -1 examples/ | sed -e 's/\.rs$//' | xargs -I "{}" find target/* -name "*{}*" -type f -delete
 
 #branches:
 #  only:
@@ -77,4 +82,4 @@ before_cache:
 
 notifications:
   email:
-    on_success: never
+    on_success: never

CHANGELOG.md

@@ -1,3 +1,15 @@
+Tantivy 0.10.0
+====================
+
+
+Minor
+---------
+- Small simplification of the code. 
+Calling .freq() or .doc() when .advance() has never 
+on segment postings should panic from now on.
+- Tokens exceeding `u16::max_value() - 4` chars are discarded silently instead of panicking.
+
+
 Tantivy 0.9.0
 =====================
 *0.9.0 index format is not compatible with the 
@@ -15,6 +27,35 @@ previous index format.*
   for int fields. (@fulmicoton)
 - Added DateTime field (@barrotsteindev)
 - Added IndexReader. By default, index is reloaded automatically upon new commits (@fulmicoton)
+- SIMD linear search within blocks (@fulmicoton)
+
+## How to update ?
+
+tantivy 0.9 brought some API breaking change.
+To update from tantivy 0.8, you will need to go through the following steps.
+
+- `schema::INT_INDEXED` and `schema::INT_STORED`  should be replaced by `schema::INDEXED` and `schema::INT_STORED`.
+- The index now does not hold the pool of searcher anymore. You are required to create an intermediary object called 
+`IndexReader` for this. 
+    
+    ```rust
+    // create the reader. You typically need to create 1 reader for the entire
+    // lifetime of you program.
+    let reader = index.reader()?;
+    
+    // Acquire a searcher (previously `index.searcher()`) is now written:
+    let searcher = reader.searcher();
+    
+    // With the default setting of the reader, you are not required to 
+    // call `index.load_searchers()` anymore.
+    //
+    // The IndexReader will pick up that change automatically, regardless
+    // of whether the update was done in a different process or not.
+    // If this behavior is not wanted, you can create your reader with 
+    // the `ReloadPolicy::Manual`, and manually decide when to reload the index
+    // by calling `reader.reload()?`.
+  
+    ```
 
 
 Tantivy 0.8.2

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tantivy"
-version = "0.9.0"
+version = "0.10.0-dev"
 authors = ["Paul Masurel <paul.masurel@gmail.com>"]
 license = "MIT"
 categories = ["database-implementations", "data-structures"]
@@ -23,7 +23,7 @@ snap = {version="0.2"}
 atomicwrites = {version="0.2.2", optional=true}
 tempfile = "3.0"
 log = "0.4"
-combine = "3"
+combine = ">=3.6.0,<4.0.0"
 tempdir = "0.3"
 serde = "1.0"
 serde_derive = "1.0"

README.md

@@ -17,6 +17,7 @@
 [![](https://sourcerer.io/fame/fulmicoton/tantivy-search/tantivy/images/6)](https://sourcerer.io/fame/fulmicoton/tantivy-search/tantivy/links/6)
 [![](https://sourcerer.io/fame/fulmicoton/tantivy-search/tantivy/images/7)](https://sourcerer.io/fame/fulmicoton/tantivy-search/tantivy/links/7)
 
+[![Become a patron](https://c5.patreon.com/external/logo/become_a_patron_button.png)](https://www.patreon.com/fulmicoton)
 
 
 **Tantivy** is a **full text search engine library** written in rust.
@@ -27,6 +28,14 @@ to build such a search engine.
 
 Tantivy is, in fact, strongly inspired by Lucene's design.
 
+# Benchmark
+
+Tantivy is typically faster than Lucene, but the results will depend on 
+the nature of the queries in your workload.
+
+The following [benchmark](https://tantivy-search.github.io/bench/) break downs 
+performance for different type of queries / collection.
+
 # Features
 
 - Full-text search
@@ -87,6 +96,14 @@ To check out and run tests, you can simply run :
 Some tests will not run with just `cargo test` because of `fail-rs`.
 To run the tests exhaustively, run `./run-tests.sh`. 
 
-# Contribute
+# How can I support this project ?
 
-Send me an email (paul.masurel at gmail.com) if you want to contribute to tantivy.
+There are many ways to support this project. 
+
+- If you use tantivy, tell us about your experience on [gitter](https://gitter.im/tantivy-search/tantivy) or by email (paul.masurel@gmail.com)
+- Report bugs
+- Write a blog post
+- Complete documentation
+- Contribute code (you can join [our gitter](https://gitter.im/tantivy-search/tantivy) )
+- Talk about tantivy around you
+- Drop a word on on [![Say Thanks!](https://img.shields.io/badge/Say%20Thanks-!-1EAEDB.svg)](https://saythanks.io/to/fulmicoton) or even [![Become a patron](https://c5.patreon.com/external/logo/become_a_patron_button.png)](https://www.patreon.com/fulmicoton)

src/common/mod.rs

@@ -13,7 +13,10 @@ pub use self::serialize::{BinarySerializable, FixedSize};
 pub use self::vint::{read_u32_vint, serialize_vint_u32, write_u32_vint, VInt};
 pub use byteorder::LittleEndian as Endianness;
 
-use std::io;
+/// Segment's max doc must be `< MAX_DOC_LIMIT`.
+///
+/// We do not allow segments with more than
+pub const MAX_DOC_LIMIT: u32 = 1 << 31;
 
 /// Computes the number of bits that will be used for bitpacking.
 ///
@@ -52,11 +55,6 @@ pub(crate) fn is_power_of_2(n: usize) -> bool {
     (n > 0) && (n & (n - 1) == 0)
 }
 
-/// Create a default io error given a string.
-pub(crate) fn make_io_err(msg: String) -> io::Error {
-    io::Error::new(io::ErrorKind::Other, msg)
-}
-
 /// Has length trait
 pub trait HasLen {
     /// Return length
@@ -134,4 +132,11 @@ pub(crate) mod test {
         assert_eq!(compute_num_bits(256), 9u8);
         assert_eq!(compute_num_bits(5_000_000_000), 33u8);
     }
+
+    #[test]
+    fn test_max_doc() {
+        // this is the first time I write a unit test for a constant.
+        assert!(((super::MAX_DOC_LIMIT - 1) as i32) >= 0);
+        assert!((super::MAX_DOC_LIMIT as i32) < 0);
+    }
 }

src/core/index.rs

@@ -24,6 +24,7 @@ use schema::Schema;
 use serde_json;
 use std::borrow::BorrowMut;
 use std::fmt;
+#[cfg(feature = "mmap")]
 use std::path::Path;
 use std::sync::Arc;
 use tokenizer::BoxedTokenizer;
@@ -355,10 +356,8 @@ mod tests {
     use directory::RAMDirectory;
     use schema::Field;
     use schema::{Schema, INDEXED, TEXT};
-    use std::path::PathBuf;
     use std::thread;
     use std::time::Duration;
-    use tempdir::TempDir;
     use Index;
     use IndexReader;
     use IndexWriter;
@@ -444,61 +443,69 @@ mod tests {
         test_index_on_commit_reload_policy_aux(field, &mut writer, &reader);
     }
 
-    #[test]
-    fn test_index_on_commit_reload_policy_mmap() {
-        let schema = throw_away_schema();
-        let field = schema.get_field("num_likes").unwrap();
-        let tempdir = TempDir::new("index").unwrap();
-        let tempdir_path = PathBuf::from(tempdir.path());
-        let index = Index::create_in_dir(&tempdir_path, schema).unwrap();
-        let mut writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-        writer.commit().unwrap();
-        let reader = index
-            .reader_builder()
-            .reload_policy(ReloadPolicy::OnCommit)
-            .try_into()
-            .unwrap();
-        assert_eq!(reader.searcher().num_docs(), 0);
-        test_index_on_commit_reload_policy_aux(field, &mut writer, &reader);
-    }
+    #[cfg(feature = "mmap")]
+    mod mmap_specific {
 
-    #[test]
-    fn test_index_manual_policy_mmap() {
-        let schema = throw_away_schema();
-        let field = schema.get_field("num_likes").unwrap();
-        let index = Index::create_from_tempdir(schema).unwrap();
-        let mut writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
-        writer.commit().unwrap();
-        let reader = index
-            .reader_builder()
-            .reload_policy(ReloadPolicy::Manual)
-            .try_into()
-            .unwrap();
-        assert_eq!(reader.searcher().num_docs(), 0);
-        writer.add_document(doc!(field=>1u64));
-        writer.commit().unwrap();
-        thread::sleep(Duration::from_millis(500));
-        assert_eq!(reader.searcher().num_docs(), 0);
-        reader.reload().unwrap();
-        assert_eq!(reader.searcher().num_docs(), 1);
-    }
+        use super::*;
+        use std::path::PathBuf;
+        use tempdir::TempDir;
 
-    #[test]
-    fn test_index_on_commit_reload_policy_different_directories() {
-        let schema = throw_away_schema();
-        let field = schema.get_field("num_likes").unwrap();
-        let tempdir = TempDir::new("index").unwrap();
-        let tempdir_path = PathBuf::from(tempdir.path());
-        let write_index = Index::create_in_dir(&tempdir_path, schema).unwrap();
-        let read_index = Index::open_in_dir(&tempdir_path).unwrap();
-        let reader = read_index
-            .reader_builder()
-            .reload_policy(ReloadPolicy::OnCommit)
-            .try_into()
-            .unwrap();
-        assert_eq!(reader.searcher().num_docs(), 0);
-        let mut writer = write_index.writer_with_num_threads(1, 3_000_000).unwrap();
-        test_index_on_commit_reload_policy_aux(field, &mut writer, &reader);
+        #[test]
+        fn test_index_on_commit_reload_policy_mmap() {
+            let schema = throw_away_schema();
+            let field = schema.get_field("num_likes").unwrap();
+            let tempdir = TempDir::new("index").unwrap();
+            let tempdir_path = PathBuf::from(tempdir.path());
+            let index = Index::create_in_dir(&tempdir_path, schema).unwrap();
+            let mut writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+            writer.commit().unwrap();
+            let reader = index
+                .reader_builder()
+                .reload_policy(ReloadPolicy::OnCommit)
+                .try_into()
+                .unwrap();
+            assert_eq!(reader.searcher().num_docs(), 0);
+            test_index_on_commit_reload_policy_aux(field, &mut writer, &reader);
+        }
+
+        #[test]
+        fn test_index_manual_policy_mmap() {
+            let schema = throw_away_schema();
+            let field = schema.get_field("num_likes").unwrap();
+            let index = Index::create_from_tempdir(schema).unwrap();
+            let mut writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+            writer.commit().unwrap();
+            let reader = index
+                .reader_builder()
+                .reload_policy(ReloadPolicy::Manual)
+                .try_into()
+                .unwrap();
+            assert_eq!(reader.searcher().num_docs(), 0);
+            writer.add_document(doc!(field=>1u64));
+            writer.commit().unwrap();
+            thread::sleep(Duration::from_millis(500));
+            assert_eq!(reader.searcher().num_docs(), 0);
+            reader.reload().unwrap();
+            assert_eq!(reader.searcher().num_docs(), 1);
+        }
+
+        #[test]
+        fn test_index_on_commit_reload_policy_different_directories() {
+            let schema = throw_away_schema();
+            let field = schema.get_field("num_likes").unwrap();
+            let tempdir = TempDir::new("index").unwrap();
+            let tempdir_path = PathBuf::from(tempdir.path());
+            let write_index = Index::create_in_dir(&tempdir_path, schema).unwrap();
+            let read_index = Index::open_in_dir(&tempdir_path).unwrap();
+            let reader = read_index
+                .reader_builder()
+                .reload_policy(ReloadPolicy::OnCommit)
+                .try_into()
+                .unwrap();
+            assert_eq!(reader.searcher().num_docs(), 0);
+            let mut writer = write_index.writer_with_num_threads(1, 3_000_000).unwrap();
+            test_index_on_commit_reload_policy_aux(field, &mut writer, &reader);
+        }
     }
 
     fn test_index_on_commit_reload_policy_aux(

src/directory/managed_directory.rs

@@ -260,95 +260,98 @@ impl Clone for ManagedDirectory {
 #[cfg(test)]
 mod tests {
 
-    use super::*;
     #[cfg(feature = "mmap")]
-    use directory::MmapDirectory;
-    use std::io::Write;
-    use std::path::Path;
-    use tempdir::TempDir;
+    mod mmap_specific {
 
-    lazy_static! {
-        static ref TEST_PATH1: &'static Path = Path::new("some_path_for_test");
-        static ref TEST_PATH2: &'static Path = Path::new("some_path_for_test2");
-    }
+        use super::super::*;
+        use std::path::Path;
+        use tempdir::TempDir;
 
-    #[test]
-    #[cfg(feature = "mmap")]
-    fn test_managed_directory() {
-        let tempdir = TempDir::new("index").unwrap();
-        let tempdir_path = PathBuf::from(tempdir.path());
-        {
-            let mmap_directory = MmapDirectory::open(&tempdir_path).unwrap();
-            let mut managed_directory = ManagedDirectory::wrap(mmap_directory).unwrap();
+        lazy_static! {
+            static ref TEST_PATH1: &'static Path = Path::new("some_path_for_test");
+            static ref TEST_PATH2: &'static Path = Path::new("some_path_for_test2");
+        }
+
+        use directory::MmapDirectory;
+        use std::io::Write;
+
+        #[test]
+        fn test_managed_directory() {
+            let tempdir = TempDir::new("index").unwrap();
+            let tempdir_path = PathBuf::from(tempdir.path());
             {
-                let mut write_file = managed_directory.open_write(*TEST_PATH1).unwrap();
-                write_file.flush().unwrap();
+                let mmap_directory = MmapDirectory::open(&tempdir_path).unwrap();
+                let mut managed_directory = ManagedDirectory::wrap(mmap_directory).unwrap();
+                {
+                    let mut write_file = managed_directory.open_write(*TEST_PATH1).unwrap();
+                    write_file.flush().unwrap();
+                }
+                {
+                    managed_directory
+                        .atomic_write(*TEST_PATH2, &vec![0u8, 1u8])
+                        .unwrap();
+                }
+                {
+                    assert!(managed_directory.exists(*TEST_PATH1));
+                    assert!(managed_directory.exists(*TEST_PATH2));
+                }
+                {
+                    let living_files: HashSet<PathBuf> =
+                        [TEST_PATH1.to_owned()].into_iter().cloned().collect();
+                    managed_directory.garbage_collect(|| living_files);
+                }
+                {
+                    assert!(managed_directory.exists(*TEST_PATH1));
+                    assert!(!managed_directory.exists(*TEST_PATH2));
+                }
             }
             {
-                managed_directory
-                    .atomic_write(*TEST_PATH2, &vec![0u8, 1u8])
-                    .unwrap();
-            }
-            {
-                assert!(managed_directory.exists(*TEST_PATH1));
-                assert!(managed_directory.exists(*TEST_PATH2));
-            }
-            {
-                let living_files: HashSet<PathBuf> =
-                    [TEST_PATH1.to_owned()].into_iter().cloned().collect();
-                managed_directory.garbage_collect(|| living_files);
-            }
-            {
-                assert!(managed_directory.exists(*TEST_PATH1));
-                assert!(!managed_directory.exists(*TEST_PATH2));
+                let mmap_directory = MmapDirectory::open(&tempdir_path).unwrap();
+                let mut managed_directory = ManagedDirectory::wrap(mmap_directory).unwrap();
+                {
+                    assert!(managed_directory.exists(*TEST_PATH1));
+                    assert!(!managed_directory.exists(*TEST_PATH2));
+                }
+                {
+                    let living_files: HashSet<PathBuf> = HashSet::new();
+                    managed_directory.garbage_collect(|| living_files);
+                }
+                {
+                    assert!(!managed_directory.exists(*TEST_PATH1));
+                    assert!(!managed_directory.exists(*TEST_PATH2));
+                }
             }
         }
-        {
+
+        #[test]
+        fn test_managed_directory_gc_while_mmapped() {
+            let tempdir = TempDir::new("index").unwrap();
+            let tempdir_path = PathBuf::from(tempdir.path());
+            let living_files = HashSet::new();
+
             let mmap_directory = MmapDirectory::open(&tempdir_path).unwrap();
             let mut managed_directory = ManagedDirectory::wrap(mmap_directory).unwrap();
-            {
-                assert!(managed_directory.exists(*TEST_PATH1));
-                assert!(!managed_directory.exists(*TEST_PATH2));
-            }
-            {
-                let living_files: HashSet<PathBuf> = HashSet::new();
-                managed_directory.garbage_collect(|| living_files);
-            }
-            {
-                assert!(!managed_directory.exists(*TEST_PATH1));
-                assert!(!managed_directory.exists(*TEST_PATH2));
-            }
-        }
-    }
-
-    #[test]
-    #[cfg(feature = "mmap ")]
-    fn test_managed_directory_gc_while_mmapped() {
-        let tempdir = TempDir::new("index").unwrap();
-        let tempdir_path = PathBuf::from(tempdir.path());
-        let living_files = HashSet::new();
-
-        let mmap_directory = MmapDirectory::open(&tempdir_path).unwrap();
-        let mut managed_directory = ManagedDirectory::wrap(mmap_directory).unwrap();
-        managed_directory
-            .atomic_write(*TEST_PATH1, &vec![0u8, 1u8])
-            .unwrap();
-        assert!(managed_directory.exists(*TEST_PATH1));
-
-        let _mmap_read = managed_directory.open_read(*TEST_PATH1).unwrap();
-        managed_directory.garbage_collect(|| living_files.clone());
-        if cfg!(target_os = "windows") {
-            // On Windows, gc should try and fail the file as it is mmapped.
+            managed_directory
+                .atomic_write(*TEST_PATH1, &vec![0u8, 1u8])
+                .unwrap();
             assert!(managed_directory.exists(*TEST_PATH1));
-            // unmap should happen here.
-            drop(_mmap_read);
-            // The file should still be in the list of managed file and
-            // eventually be deleted once mmap is released.
-            managed_directory.garbage_collect(|| living_files);
-            assert!(!managed_directory.exists(*TEST_PATH1));
-        } else {
-            assert!(!managed_directory.exists(*TEST_PATH1));
+
+            let _mmap_read = managed_directory.open_read(*TEST_PATH1).unwrap();
+            managed_directory.garbage_collect(|| living_files.clone());
+            if cfg!(target_os = "windows") {
+                // On Windows, gc should try and fail the file as it is mmapped.
+                assert!(managed_directory.exists(*TEST_PATH1));
+                // unmap should happen here.
+                drop(_mmap_read);
+                // The file should still be in the list of managed file and
+                // eventually be deleted once mmap is released.
+                managed_directory.garbage_collect(|| living_files);
+                assert!(!managed_directory.exists(*TEST_PATH1));
+            } else {
+                assert!(!managed_directory.exists(*TEST_PATH1));
+            }
         }
+
     }
 
 }

src/directory/mmap_directory.rs

@@ -6,7 +6,6 @@ use self::notify::RawEvent;
 use self::notify::RecursiveMode;
 use self::notify::Watcher;
 use atomicwrites;
-use common::make_io_err;
 use core::META_FILEPATH;
 use directory::error::LockError;
 use directory::error::{DeleteError, IOError, OpenDirectoryError, OpenReadError, OpenWriteError};
@@ -37,6 +36,11 @@ use std::sync::Weak;
 use std::thread;
 use tempdir::TempDir;
 
+/// Create a default io error given a string.
+pub(crate) fn make_io_err(msg: String) -> io::Error {
+    io::Error::new(io::ErrorKind::Other, msg)
+}
+
 /// Returns None iff the file exists, can be read, but is empty (and hence
 /// cannot be mmapped)
 fn open_mmap(full_path: &Path) -> result::Result<Option<Mmap>, OpenReadError> {

src/docset.rs

@@ -1,4 +1,5 @@
 use common::BitSet;
+use fastfield::DeleteBitSet;
 use std::borrow::Borrow;
 use std::borrow::BorrowMut;
 use std::cmp::Ordering;
@@ -95,9 +96,23 @@ pub trait DocSet {
     }
 
     /// Returns the number documents matching.
-    ///
     /// Calling this method consumes the `DocSet`.
-    fn count(&mut self) -> u32 {
+    fn count(&mut self, delete_bitset: &DeleteBitSet) -> u32 {
+        let mut count = 0u32;
+        while self.advance() {
+            if !delete_bitset.is_deleted(self.doc()) {
+                count += 1u32;
+            }
+        }
+        count
+    }
+
+    /// Returns the count of documents, deleted or not.
+    /// Calling this method consumes the `DocSet`.
+    ///
+    /// Of course, the result is an upper bound of the result
+    /// given by `count()`.
+    fn count_including_deleted(&mut self) -> u32 {
         let mut count = 0u32;
         while self.advance() {
             count += 1u32;
@@ -127,9 +142,14 @@ impl<TDocSet: DocSet + ?Sized> DocSet for Box<TDocSet> {
         unboxed.size_hint()
     }
 
-    fn count(&mut self) -> u32 {
+    fn count(&mut self, delete_bitset: &DeleteBitSet) -> u32 {
         let unboxed: &mut TDocSet = self.borrow_mut();
-        unboxed.count()
+        unboxed.count(delete_bitset)
+    }
+
+    fn count_including_deleted(&mut self) -> u32 {
+        let unboxed: &mut TDocSet = self.borrow_mut();
+        unboxed.count_including_deleted()
     }
 
     fn append_to_bitset(&mut self, bitset: &mut BitSet) {

src/functional_test.rs

@@ -13,7 +13,6 @@ fn check_index_content(searcher: &Searcher, vals: &HashSet<u64>) {
 
 #[test]
 #[ignore]
-#[cfg(feature = "mmap")]
 fn test_indexing() {
     let mut schema_builder = Schema::builder();
 

src/indexer/merger.rs

@@ -1,3 +1,4 @@
+use common::MAX_DOC_LIMIT;
 use core::Segment;
 use core::SegmentReader;
 use core::SerializableSegment;
@@ -23,6 +24,7 @@ use termdict::TermMerger;
 use termdict::TermOrdinal;
 use DocId;
 use Result;
+use TantivyError;
 
 fn compute_total_num_tokens(readers: &[SegmentReader], field: Field) -> u64 {
     let mut total_tokens = 0u64;
@@ -150,6 +152,14 @@ impl IndexMerger {
                 readers.push(reader);
             }
         }
+        if max_doc >= MAX_DOC_LIMIT {
+            let err_msg = format!(
+                "The segment resulting from this merge would have {} docs,\
+                 which exceeds the limit {}.",
+                max_doc, MAX_DOC_LIMIT
+            );
+            return Err(TantivyError::InvalidArgument(err_msg));
+        }
         Ok(IndexMerger {
             schema,
             readers,

src/indexer/segment_updater.rs

@@ -420,6 +420,7 @@ impl SegmentUpdater {
             })
             .collect::<Vec<_>>();
         merge_candidates.extend(committed_merge_candidates.into_iter());
+
         for merge_operation in merge_candidates {
             match self.start_merge_impl(merge_operation) {
                 Ok(merge_future) => {

src/lib.rs

@@ -174,6 +174,7 @@ extern crate downcast_rs;
 #[macro_use]
 extern crate fail;
 
+#[cfg(feature = "mmap")]
 #[cfg(test)]
 mod functional_test;
 

src/postings/block_search.rs

@@ -0,0 +1,249 @@
+use postings::compression::AlignedBuffer;
+
+/// This modules define the logic used to search for a doc in a given
+/// block. (at most 128 docs)
+///
+/// Searching within a block is a hotspot when running intersection.
+/// so it was worth defining it in its own module.
+
+#[cfg(target_arch = "x86_64")]
+mod sse2 {
+    use postings::compression::{AlignedBuffer, COMPRESSION_BLOCK_SIZE};
+    use std::arch::x86_64::__m128i as DataType;
+    use std::arch::x86_64::_mm_add_epi32 as op_add;
+    use std::arch::x86_64::_mm_cmplt_epi32 as op_lt;
+    use std::arch::x86_64::_mm_load_si128 as op_load; // requires 128-bits alignment
+    use std::arch::x86_64::_mm_set1_epi32 as set1;
+    use std::arch::x86_64::_mm_setzero_si128 as set0;
+    use std::arch::x86_64::_mm_sub_epi32 as op_sub;
+    use std::arch::x86_64::{_mm_cvtsi128_si32, _mm_shuffle_epi32};
+
+    const MASK1: i32 = 78;
+    const MASK2: i32 = 177;
+
+    /// Performs an exhaustive linear search over the
+    ///
+    /// There is no early exit here. We simply count the
+    /// number of elements that are `< target`.
+    pub(crate) fn linear_search_sse2_128(arr: &AlignedBuffer, target: u32) -> usize {
+        unsafe {
+            let ptr = arr as *const AlignedBuffer as *const DataType;
+            let vkey = set1(target as i32);
+            let mut cnt = set0();
+            // We work over 4 `__m128i` at a time.
+            // A single `__m128i` actual contains 4 `u32`.
+            for i in 0..(COMPRESSION_BLOCK_SIZE as isize) / (4 * 4) {
+                let cmp1 = op_lt(op_load(ptr.offset(i * 4)), vkey);
+                let cmp2 = op_lt(op_load(ptr.offset(i * 4 + 1)), vkey);
+                let cmp3 = op_lt(op_load(ptr.offset(i * 4 + 2)), vkey);
+                let cmp4 = op_lt(op_load(ptr.offset(i * 4 + 3)), vkey);
+                let sum = op_add(op_add(cmp1, cmp2), op_add(cmp3, cmp4));
+                cnt = op_sub(cnt, sum);
+            }
+            cnt = op_add(cnt, _mm_shuffle_epi32(cnt, MASK1));
+            cnt = op_add(cnt, _mm_shuffle_epi32(cnt, MASK2));
+            _mm_cvtsi128_si32(cnt) as usize
+        }
+    }
+
+    #[cfg(test)]
+    mod test {
+        use super::linear_search_sse2_128;
+        use postings::compression::{AlignedBuffer, COMPRESSION_BLOCK_SIZE};
+
+        #[test]
+        fn test_linear_search_sse2_128_u32() {
+            let mut block = [0u32; COMPRESSION_BLOCK_SIZE];
+            for el in 0u32..128u32 {
+                block[el as usize] = el * 2 + 1 << 18;
+            }
+            let target = block[64] + 1;
+            assert_eq!(linear_search_sse2_128(&AlignedBuffer(block), target), 65);
+        }
+    }
+}
+
+/// This `linear search` browser exhaustively through the array.
+/// but the early exit is very difficult to predict.
+///
+/// Coupled with `exponential search` this function is likely
+/// to be called with the same `len`
+fn linear_search(arr: &[u32], target: u32) -> usize {
+    arr.iter().map(|&el| if el < target { 1 } else { 0 }).sum()
+}
+
+fn exponential_search(arr: &[u32], target: u32) -> (usize, usize) {
+    let end = arr.len();
+    let mut begin = 0;
+    for &pivot in &[1, 3, 7, 15, 31, 63] {
+        if pivot >= end {
+            break;
+        }
+        if arr[pivot] > target {
+            return (begin, pivot);
+        }
+        begin = pivot;
+    }
+    (begin, end)
+}
+
+fn galloping(block_docs: &[u32], target: u32) -> usize {
+    let (start, end) = exponential_search(&block_docs, target);
+    start + linear_search(&block_docs[start..end], target)
+}
+
+/// Tantivy may rely on SIMD instructions to search for a specific document within
+/// a given block.
+#[derive(Clone, Copy, PartialEq)]
+pub enum BlockSearcher {
+    #[cfg(target_arch = "x86_64")]
+    SSE2,
+    Scalar,
+}
+
+impl BlockSearcher {
+    /// Search the first index containing an element greater or equal to
+    /// the target.
+    ///
+    /// The results should be equivalent to
+    /// ```ignore
+    /// block[..]
+    //       .iter()
+    //       .take_while(|&&val| val < target)
+    //       .count()
+    /// ```
+    ///
+    /// The `start` argument is just used to hint that the response is
+    /// greater than beyond `start`. The implementation may or may not use
+    /// it for optimization.
+    ///
+    /// # Assumption
+    ///
+    /// The array len is > start.
+    /// The block is sorted
+    /// The target is assumed greater or equal to the `arr[start]`.
+    /// The target is assumed smaller or equal to the last element of the block.
+    ///
+    /// Currently the scalar implementation starts by an exponential search, and
+    /// then operates a linear search in the result subarray.
+    ///
+    /// If SSE2 instructions are available in the `(platform, running CPU)`,
+    /// then we use a different implementation that does an exhaustive linear search over
+    /// the full block whenever the block is full (`len == 128`). It is surprisingly faster, most likely because of the lack
+    /// of branch.
+    pub(crate) fn search_in_block(
+        self,
+        block_docs: &AlignedBuffer,
+        len: usize,
+        start: usize,
+        target: u32,
+    ) -> usize {
+        #[cfg(target_arch = "x86_64")]
+        {
+            use postings::compression::COMPRESSION_BLOCK_SIZE;
+            if self == BlockSearcher::SSE2 && len == COMPRESSION_BLOCK_SIZE {
+                return sse2::linear_search_sse2_128(block_docs, target);
+            }
+        }
+        start + galloping(&block_docs.0[start..len], target)
+    }
+}
+
+impl Default for BlockSearcher {
+    fn default() -> BlockSearcher {
+        #[cfg(target_arch = "x86_64")]
+        {
+            if is_x86_feature_detected!("sse2") {
+                return BlockSearcher::SSE2;
+            }
+        }
+        BlockSearcher::Scalar
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::exponential_search;
+    use super::linear_search;
+    use super::BlockSearcher;
+    use postings::compression::{AlignedBuffer, COMPRESSION_BLOCK_SIZE};
+
+    #[test]
+    fn test_linear_search() {
+        let len: usize = 50;
+        let arr: Vec<u32> = (0..len).map(|el| 1u32 + (el as u32) * 2).collect();
+        for target in 1..*arr.last().unwrap() {
+            let res = linear_search(&arr[..], target);
+            if res > 0 {
+                assert!(arr[res - 1] < target);
+            }
+            if res < len {
+                assert!(arr[res] >= target);
+            }
+        }
+    }
+
+    #[test]
+    fn test_exponentiel_search() {
+        assert_eq!(exponential_search(&[1, 2], 0), (0, 1));
+        assert_eq!(exponential_search(&[1, 2], 1), (0, 1));
+        assert_eq!(
+            exponential_search(&[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11], 7),
+            (3, 7)
+        );
+    }
+
+    fn util_test_search_in_block(block_searcher: BlockSearcher, block: &[u32], target: u32) {
+        let cursor = search_in_block_trivial_but_slow(block, target);
+        assert!(block.len() < COMPRESSION_BLOCK_SIZE);
+        let mut output_buffer = [u32::max_value(); COMPRESSION_BLOCK_SIZE];
+        output_buffer[..block.len()].copy_from_slice(block);
+        for i in 0..cursor {
+            assert_eq!(
+                block_searcher.search_in_block(
+                    &AlignedBuffer(output_buffer),
+                    block.len(),
+                    i,
+                    target
+                ),
+                cursor
+            );
+        }
+    }
+
+    fn util_test_search_in_block_all(block_searcher: BlockSearcher, block: &[u32]) {
+        use std::collections::HashSet;
+        let mut targets = HashSet::new();
+        for (i, val) in block.iter().cloned().enumerate() {
+            if i > 0 {
+                targets.insert(val - 1);
+            }
+            targets.insert(val);
+        }
+        for target in targets {
+            util_test_search_in_block(block_searcher, block, target);
+        }
+    }
+
+    fn search_in_block_trivial_but_slow(block: &[u32], target: u32) -> usize {
+        block.iter().take_while(|&&val| val < target).count()
+    }
+
+    fn test_search_in_block_util(block_searcher: BlockSearcher) {
+        for len in 1u32..128u32 {
+            let v: Vec<u32> = (0..len).map(|i| i * 2).collect();
+            util_test_search_in_block_all(block_searcher, &v[..]);
+        }
+    }
+
+    #[test]
+    fn test_search_in_block_scalar() {
+        test_search_in_block_util(BlockSearcher::Scalar);
+    }
+
+    #[cfg(target_arch = "x86_64")]
+    #[test]
+    fn test_search_in_block_sse2() {
+        test_search_in_block_util(BlockSearcher::SSE2);
+    }
+}

src/postings/compression/mod.rs

@@ -43,9 +43,14 @@ impl BlockEncoder {
     }
 }
 
+/// We ensure that the OutputBuffer is align on 128 bits
+/// in order to run SSE2 linear search on it.
+#[repr(align(128))]
+pub(crate) struct AlignedBuffer(pub [u32; COMPRESSION_BLOCK_SIZE]);
+
 pub struct BlockDecoder {
     bitpacker: BitPacker4x,
-    pub output: [u32; COMPRESSION_BLOCK_SIZE + 1],
+    output: AlignedBuffer,
     pub output_len: usize,
 }
 
@@ -55,11 +60,9 @@ impl BlockDecoder {
     }
 
     pub fn with_val(val: u32) -> BlockDecoder {
-        let mut output = [val; COMPRESSION_BLOCK_SIZE + 1];
-        output[COMPRESSION_BLOCK_SIZE] = 0u32;
         BlockDecoder {
             bitpacker: BitPacker4x::new(),
-            output,
+            output: AlignedBuffer([val; COMPRESSION_BLOCK_SIZE]),
             output_len: 0,
         }
     }
@@ -72,23 +75,28 @@ impl BlockDecoder {
     ) -> usize {
         self.output_len = COMPRESSION_BLOCK_SIZE;
         self.bitpacker
-            .decompress_sorted(offset, &compressed_data, &mut self.output, num_bits)
+            .decompress_sorted(offset, &compressed_data, &mut self.output.0, num_bits)
     }
 
     pub fn uncompress_block_unsorted(&mut self, compressed_data: &[u8], num_bits: u8) -> usize {
         self.output_len = COMPRESSION_BLOCK_SIZE;
         self.bitpacker
-            .decompress(&compressed_data, &mut self.output, num_bits)
+            .decompress(&compressed_data, &mut self.output.0, num_bits)
     }
 
     #[inline]
     pub fn output_array(&self) -> &[u32] {
-        &self.output[..self.output_len]
+        &self.output.0[..self.output_len]
+    }
+
+    #[inline]
+    pub(crate) fn output_aligned(&self) -> (&AlignedBuffer, usize) {
+        (&self.output, self.output_len)
     }
 
     #[inline]
     pub fn output(&self, idx: usize) -> u32 {
-        self.output[idx]
+        self.output.0[idx]
     }
 }
 
@@ -159,12 +167,12 @@ impl VIntDecoder for BlockDecoder {
         num_els: usize,
     ) -> usize {
         self.output_len = num_els;
-        vint::uncompress_sorted(compressed_data, &mut self.output[..num_els], offset)
+        vint::uncompress_sorted(compressed_data, &mut self.output.0[..num_els], offset)
     }
 
     fn uncompress_vint_unsorted<'a>(&mut self, compressed_data: &'a [u8], num_els: usize) -> usize {
         self.output_len = num_els;
-        vint::uncompress_unsorted(compressed_data, &mut self.output[..num_els])
+        vint::uncompress_unsorted(compressed_data, &mut self.output.0[..num_els])
     }
 }
 

src/postings/mod.rs

@@ -2,6 +2,7 @@
 Postings module (also called inverted index)
 */
 
+mod block_search;
 pub(crate) mod compression;
 /// Postings module
 ///
@@ -16,6 +17,8 @@ mod skip;
 mod stacker;
 mod term_info;
 
+pub(crate) use self::block_search::BlockSearcher;
+
 pub(crate) use self::postings_writer::MultiFieldPostingsWriter;
 pub use self::serializer::{FieldSerializer, InvertedIndexSerializer};
 
@@ -52,13 +55,15 @@ pub mod tests {
     use fieldnorm::FieldNormReader;
     use indexer::operation::AddOperation;
     use indexer::SegmentWriter;
+    use merge_policy::NoMergePolicy;
     use query::Scorer;
     use rand::rngs::StdRng;
     use rand::{Rng, SeedableRng};
-    use schema::Field;
-    use schema::IndexRecordOption;
     use schema::{Document, Schema, Term, INDEXED, STRING, TEXT};
+    use schema::{Field, TextOptions};
+    use schema::{IndexRecordOption, TextFieldIndexing};
     use std::iter;
+    use tokenizer::{SimpleTokenizer, MAX_TOKEN_LEN};
     use DocId;
     use Score;
 
@@ -104,9 +109,7 @@ pub mod tests {
         let searcher = index.reader().unwrap().searcher();
         let inverted_index = searcher.segment_reader(0u32).inverted_index(title);
         let term = Term::from_field_text(title, "abc");
-
         let mut positions = Vec::new();
-
         {
             let mut postings = inverted_index
                 .read_postings(&term, IndexRecordOption::WithFreqsAndPositions)
@@ -159,6 +162,52 @@ pub mod tests {
         }
     }
 
+    #[test]
+    pub fn test_drop_token_that_are_too_long() {
+        let ok_token_text: String = iter::repeat('A').take(MAX_TOKEN_LEN).collect();
+        let mut exceeding_token_text: String = iter::repeat('A').take(MAX_TOKEN_LEN + 1).collect();
+        exceeding_token_text.push_str(" hello");
+        let mut schema_builder = Schema::builder();
+        let text_options = TextOptions::default().set_indexing_options(
+            TextFieldIndexing::default()
+                .set_index_option(IndexRecordOption::WithFreqsAndPositions)
+                .set_tokenizer("simple_no_truncation"),
+        );
+        let text_field = schema_builder.add_text_field("text", text_options);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema.clone());
+        index
+            .tokenizers()
+            .register("simple_no_truncation", SimpleTokenizer);
+        let reader = index.reader().unwrap();
+        let mut index_writer = index.writer_with_num_threads(1, 3_000_000).unwrap();
+        index_writer.set_merge_policy(Box::new(NoMergePolicy));
+        {
+            index_writer.add_document(doc!(text_field=>exceeding_token_text));
+            index_writer.commit().unwrap();
+            reader.reload().unwrap();
+            let searcher = reader.searcher();
+            let segment_reader = searcher.segment_reader(0u32);
+            let inverted_index = segment_reader.inverted_index(text_field);
+            assert_eq!(inverted_index.terms().num_terms(), 1);
+            let mut bytes = vec![];
+            assert!(inverted_index.terms().ord_to_term(0, &mut bytes));
+            assert_eq!(&bytes, b"hello");
+        }
+        {
+            index_writer.add_document(doc!(text_field=>ok_token_text.clone()));
+            index_writer.commit().unwrap();
+            reader.reload().unwrap();
+            let searcher = reader.searcher();
+            let segment_reader = searcher.segment_reader(1u32);
+            let inverted_index = segment_reader.inverted_index(text_field);
+            assert_eq!(inverted_index.terms().num_terms(), 1);
+            let mut bytes = vec![];
+            assert!(inverted_index.terms().ord_to_term(0, &mut bytes));
+            assert_eq!(&bytes[..], ok_token_text.as_bytes());
+        }
+    }
+
     #[test]
     pub fn test_position_and_fieldnorm1() {
         let mut positions = Vec::new();

src/postings/postings_writer.rs

@@ -12,8 +12,8 @@ use std::io;
 use std::marker::PhantomData;
 use std::ops::DerefMut;
 use termdict::TermOrdinal;
-use tokenizer::Token;
 use tokenizer::TokenStream;
+use tokenizer::{Token, MAX_TOKEN_LEN};
 use DocId;
 use Result;
 
@@ -210,8 +210,18 @@ pub trait PostingsWriter {
     ) -> u32 {
         let mut term = Term::for_field(field);
         let mut sink = |token: &Token| {
-            term.set_text(token.text.as_str());
-            self.subscribe(term_index, doc_id, token.position as u32, &term, heap);
+            // We skip all tokens with a len greater than u16.
+            if token.text.len() <= MAX_TOKEN_LEN {
+                term.set_text(token.text.as_str());
+                self.subscribe(term_index, doc_id, token.position as u32, &term, heap);
+            } else {
+                info!(
+                    "A token exceeding MAX_TOKEN_LEN ({}>{}) was dropped. Search for \
+                     MAX_TOKEN_LEN in the documentation for more information.",
+                    token.text.len(),
+                    MAX_TOKEN_LEN
+                );
+            }
         };
         token_stream.process(&mut sink)
     }

src/postings/segment_postings.rs

@@ -4,9 +4,10 @@ use common::{BinarySerializable, VInt};
 use docset::{DocSet, SkipResult};
 use owned_read::OwnedRead;
 use positions::PositionReader;
-use postings::compression::compressed_block_size;
+use postings::compression::{compressed_block_size, AlignedBuffer};
 use postings::compression::{BlockDecoder, VIntDecoder, COMPRESSION_BLOCK_SIZE};
 use postings::serializer::PostingsSerializer;
+use postings::BlockSearcher;
 use postings::FreqReadingOption;
 use postings::Postings;
 use postings::SkipReader;
@@ -60,6 +61,7 @@ pub struct SegmentPostings {
     block_cursor: BlockSegmentPostings,
     cur: usize,
     position_computer: Option<PositionComputer>,
+    block_searcher: BlockSearcher,
 }
 
 impl SegmentPostings {
@@ -70,6 +72,7 @@ impl SegmentPostings {
             block_cursor: empty_block_cursor,
             cur: COMPRESSION_BLOCK_SIZE,
             position_computer: None,
+            block_searcher: BlockSearcher::default(),
         }
     }
 
@@ -117,42 +120,33 @@ impl SegmentPostings {
             block_cursor: segment_block_postings,
             cur: COMPRESSION_BLOCK_SIZE, // cursor within the block
             position_computer: positions_stream_opt.map(PositionComputer::new),
+            block_searcher: BlockSearcher::default(),
         }
     }
 }
 
-fn linear_search(arr: &[u32], target: u32) -> usize {
-    arr.iter().map(|&el| if el < target { 1 } else { 0 }).sum()
-}
-
-fn exponential_search(arr: &[u32], target: u32) -> (usize, usize) {
-    let end = arr.len();
-    let mut begin = 0;
-    for &pivot in &[1, 3, 7, 15, 31, 63] {
-        if pivot >= end {
-            break;
-        }
-        if arr[pivot] > target {
-            return (begin, pivot);
-        }
-        begin = pivot;
-    }
-    (begin, end)
-}
-
-/// Search the first index containing an element greater or equal to the target.
-///
-/// # Assumption
-///
-/// The array is assumed non empty.
-/// The target is assumed greater or equal to the first element.
-/// The target is assumed smaller or equal to the last element.
-fn search_within_block(block_docs: &[u32], target: u32) -> usize {
-    let (start, end) = exponential_search(block_docs, target);
-    start + linear_search(&block_docs[start..end], target)
-}
-
 impl DocSet for SegmentPostings {
+    // goes to the next element.
+    // next needs to be called a first time to point to the correct element.
+    #[inline]
+    fn advance(&mut self) -> bool {
+        if self.position_computer.is_some() && self.cur < COMPRESSION_BLOCK_SIZE {
+            let term_freq = self.term_freq() as usize;
+            if let Some(position_computer) = self.position_computer.as_mut() {
+                position_computer.add_skip(term_freq);
+            }
+        }
+        self.cur += 1;
+        if self.cur >= self.block_cursor.block_len() {
+            self.cur = 0;
+            if !self.block_cursor.advance() {
+                self.cur = COMPRESSION_BLOCK_SIZE;
+                return false;
+            }
+        }
+        true
+    }
+
     fn skip_next(&mut self, target: DocId) -> SkipResult {
         if !self.advance() {
             return SkipResult::End;
@@ -175,7 +169,6 @@ impl DocSet for SegmentPostings {
 
         // skip blocks until one that might contain the target
         // check if we need to go to the next block
-        let need_positions = self.position_computer.is_some();
         let mut sum_freqs_skipped: u32 = 0;
         if !self
             .block_cursor
@@ -189,7 +182,7 @@ impl DocSet for SegmentPostings {
             // we are not in the right block.
             //
             // First compute all of the freqs skipped from the current block.
-            if need_positions {
+            if self.position_computer.is_some() {
                 sum_freqs_skipped = self.block_cursor.freqs()[self.cur..].iter().sum();
                 match self.block_cursor.skip_to(target) {
                     BlockSegmentPostingsSkipResult::Success(block_skip_freqs) => {
@@ -208,25 +201,21 @@ impl DocSet for SegmentPostings {
             self.cur = 0;
         }
 
-        // we're in the right block now, start with an exponential search
-        let block_docs = self.block_cursor.docs();
-        let new_cur = self
-            .cur
-            .wrapping_add(search_within_block(&block_docs[self.cur..], target));
+        let cur = self.cur;
 
-        if need_positions {
-            sum_freqs_skipped += self.block_cursor.freqs()[self.cur..new_cur]
-                .iter()
-                .sum::<u32>();
-            self.position_computer
-                .as_mut()
-                .unwrap()
-                .add_skip(sum_freqs_skipped as usize);
+        // we're in the right block now, start with an exponential search
+        let (output, len) = self.block_cursor.docs_aligned();
+        let new_cur = self
+            .block_searcher
+            .search_in_block(&output, len, cur, target);
+        if let Some(position_computer) = self.position_computer.as_mut() {
+            sum_freqs_skipped += self.block_cursor.freqs()[cur..new_cur].iter().sum::<u32>();
+            position_computer.add_skip(sum_freqs_skipped as usize);
         }
         self.cur = new_cur;
 
         // `doc` is now the first element >= `target`
-        let doc = block_docs[new_cur];
+        let doc = output.0[new_cur];
         debug_assert!(doc >= target);
         if doc == target {
             SkipResult::Reached
@@ -235,40 +224,25 @@ impl DocSet for SegmentPostings {
         }
     }
 
-    // goes to the next element.
-    // next needs to be called a first time to point to the correct element.
-    #[inline]
-    fn advance(&mut self) -> bool {
-        if self.position_computer.is_some() {
-            let term_freq = self.term_freq() as usize;
-            self.position_computer.as_mut().unwrap().add_skip(term_freq);
-        }
-        self.cur += 1;
-        if self.cur >= self.block_cursor.block_len() {
-            self.cur = 0;
-            if !self.block_cursor.advance() {
-                self.cur = COMPRESSION_BLOCK_SIZE;
-                return false;
-            }
-        }
-        true
-    }
-
-    fn size_hint(&self) -> u32 {
-        self.len() as u32
-    }
-
     /// Return the current document's `DocId`.
+    ///
+    /// # Panics
+    ///
+    /// Will panics if called without having called advance before.
     #[inline]
     fn doc(&self) -> DocId {
         let docs = self.block_cursor.docs();
         debug_assert!(
             self.cur < docs.len(),
-            "Have you forgotten to call `.advance()` at least once before calling .doc()."
+            "Have you forgotten to call `.advance()` at least once before calling `.doc()`                                      ."
         );
         docs[self.cur]
     }
 
+    fn size_hint(&self) -> u32 {
+        self.len() as u32
+    }
+
     fn append_to_bitset(&mut self, bitset: &mut BitSet) {
         // finish the current block
         if self.advance() {
@@ -292,17 +266,33 @@ impl HasLen for SegmentPostings {
 }
 
 impl Postings for SegmentPostings {
+    /// Returns the frequency associated to the current document.
+    /// If the schema is set up so that no frequency have been encoded,
+    /// this method should always return 1.
+    ///
+    /// # Panics
+    ///
+    /// Will panics if called without having called advance before.
     fn term_freq(&self) -> u32 {
+        debug_assert!(
+            // Here we do not use the len of `freqs()`
+            // because it is actually ok to request for the freq of doc
+            // even if no frequency were encoded for the field.
+            //
+            // In that case we hit the block just as if the frequency had been
+            // decoded. The block is simply prefilled by the value 1.
+            self.cur < COMPRESSION_BLOCK_SIZE,
+            "Have you forgotten to call `.advance()` at least once before calling \
+             `.term_freq()`."
+        );
         self.block_cursor.freq(self.cur)
     }
 
     fn positions_with_offset(&mut self, offset: u32, output: &mut Vec<u32>) {
-        if self.position_computer.is_some() {
-            output.resize(self.term_freq() as usize, 0u32);
-            self.position_computer
-                .as_mut()
-                .unwrap()
-                .positions_with_offset(offset, &mut output[..])
+        let term_freq = self.term_freq() as usize;
+        if let Some(position_comp) = self.position_computer.as_mut() {
+            output.resize(term_freq, 0u32);
+            position_comp.positions_with_offset(offset, &mut output[..]);
         } else {
             output.clear();
         }
@@ -424,6 +414,10 @@ impl BlockSegmentPostings {
         self.doc_decoder.output_array()
     }
 
+    pub(crate) fn docs_aligned(&self) -> (&AlignedBuffer, usize) {
+        self.doc_decoder.output_aligned()
+    }
+
     /// Return the document at index `idx` of the block.
     #[inline]
     pub fn doc(&self, idx: usize) -> u32 {
@@ -614,16 +608,13 @@ impl<'b> Streamer<'b> for BlockSegmentPostings {
 
 #[cfg(test)]
 mod tests {
-
-    use super::exponential_search;
-    use super::linear_search;
-    use super::search_within_block;
     use super::BlockSegmentPostings;
     use super::BlockSegmentPostingsSkipResult;
     use super::SegmentPostings;
     use common::HasLen;
     use core::Index;
     use docset::DocSet;
+    use postings::postings::Postings;
     use schema::IndexRecordOption;
     use schema::Schema;
     use schema::Term;
@@ -632,21 +623,6 @@ mod tests {
     use DocId;
     use SkipResult;
 
-    #[test]
-    fn test_linear_search() {
-        let len: usize = 50;
-        let arr: Vec<u32> = (0..len).map(|el| 1u32 + (el as u32) * 2).collect();
-        for target in 1..*arr.last().unwrap() {
-            let res = linear_search(&arr[..], target);
-            if res > 0 {
-                assert!(arr[res - 1] < target);
-            }
-            if res < len {
-                assert!(arr[res] >= target);
-            }
-        }
-    }
-
     #[test]
     fn test_empty_segment_postings() {
         let mut postings = SegmentPostings::empty();
@@ -655,6 +631,18 @@ mod tests {
         assert_eq!(postings.len(), 0);
     }
 
+    #[test]
+    #[should_panic(expected = "Have you forgotten to call `.advance()`")]
+    fn test_panic_if_doc_called_before_advance() {
+        SegmentPostings::empty().doc();
+    }
+
+    #[test]
+    #[should_panic(expected = "Have you forgotten to call `.advance()`")]
+    fn test_panic_if_freq_called_before_advance() {
+        SegmentPostings::empty().term_freq();
+    }
+
     #[test]
     fn test_empty_block_segment_postings() {
         let mut postings = BlockSegmentPostings::empty();
@@ -662,56 +650,6 @@ mod tests {
         assert_eq!(postings.doc_freq(), 0);
     }
 
-    fn search_within_block_trivial_but_slow(block: &[u32], target: u32) -> usize {
-        block
-            .iter()
-            .cloned()
-            .enumerate()
-            .filter(|&(_, ref val)| *val >= target)
-            .next()
-            .unwrap()
-            .0
-    }
-
-    #[test]
-    fn test_exponentiel_search() {
-        assert_eq!(exponential_search(&[1, 2], 0), (0, 1));
-        assert_eq!(exponential_search(&[1, 2], 1), (0, 1));
-        assert_eq!(
-            exponential_search(&[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11], 7),
-            (3, 7)
-        );
-    }
-
-    fn util_test_search_within_block(block: &[u32], target: u32) {
-        assert_eq!(
-            search_within_block(block, target),
-            search_within_block_trivial_but_slow(block, target)
-        );
-    }
-
-    fn util_test_search_within_block_all(block: &[u32]) {
-        use std::collections::HashSet;
-        let mut targets = HashSet::new();
-        for (i, val) in block.iter().cloned().enumerate() {
-            if i > 0 {
-                targets.insert(val - 1);
-            }
-            targets.insert(val);
-        }
-        for target in targets {
-            util_test_search_within_block(block, target);
-        }
-    }
-
-    #[test]
-    fn test_search_within_block() {
-        for len in 1u32..128u32 {
-            let v: Vec<u32> = (0..len).map(|i| i * 2).collect();
-            util_test_search_within_block_all(&v[..]);
-        }
-    }
-
     #[test]
     fn test_block_segment_postings() {
         let mut block_segments = build_block_postings(&(0..100_000).collect::<Vec<u32>>());

src/postings/serializer.rs

@@ -14,7 +14,7 @@ use termdict::{TermDictionaryBuilder, TermOrdinal};
 use DocId;
 use Result;
 
-/// `PostingsSerializer` is in charge of serializing
+/// `InvertedIndexSerializer` is in charge of serializing
 /// postings on disk, in the
 /// * `.idx` (inverted index)
 /// * `.pos` (positions file)
@@ -54,7 +54,7 @@ pub struct InvertedIndexSerializer {
 }
 
 impl InvertedIndexSerializer {
-    /// Open a new `PostingsSerializer` for the given segment
+    /// Open a new `InvertedIndexSerializer` for the given segment
     fn create(
         terms_write: CompositeWrite<WritePtr>,
         postings_write: CompositeWrite<WritePtr>,

src/query/intersection.rs

@@ -14,41 +14,35 @@ use Score;
 /// specialized implementation if the two
 /// shortest scorers are `TermScorer`s.
 pub fn intersect_scorers(mut scorers: Vec<Box<Scorer>>) -> Box<Scorer> {
+    if scorers.is_empty() {
+        return Box::new(EmptyScorer);
+    }
+    if scorers.len() == 1 {
+        return scorers.pop().unwrap();
+    }
+    // We know that we have at least 2 elements.
     let num_docsets = scorers.len();
     scorers.sort_by(|left, right| right.size_hint().cmp(&left.size_hint()));
-    let rarest_opt = scorers.pop();
-    let second_rarest_opt = scorers.pop();
+    let left = scorers.pop().unwrap();
+    let right = scorers.pop().unwrap();
     scorers.reverse();
-    match (rarest_opt, second_rarest_opt) {
-        (None, None) => Box::new(EmptyScorer),
-        (Some(single_docset), None) => single_docset,
-        (Some(left), Some(right)) => {
-            {
-                let all_term_scorers = [&left, &right]
-                    .iter()
-                    .all(|&scorer| scorer.is::<TermScorer>());
-                if all_term_scorers {
-                    let left = *(left.downcast::<TermScorer>().map_err(|_| ()).unwrap());
-                    let right = *(right.downcast::<TermScorer>().map_err(|_| ()).unwrap());
-                    return Box::new(Intersection {
-                        left,
-                        right,
-                        others: scorers,
-                        num_docsets,
-                    });
-                }
-            }
-            Box::new(Intersection {
-                left,
-                right,
-                others: scorers,
-                num_docsets,
-            })
-        }
-        _ => {
-            unreachable!();
-        }
+    let all_term_scorers = [&left, &right]
+        .iter()
+        .all(|&scorer| scorer.is::<TermScorer>());
+    if all_term_scorers {
+        return Box::new(Intersection {
+            left: *(left.downcast::<TermScorer>().map_err(|_| ()).unwrap()),
+            right: *(right.downcast::<TermScorer>().map_err(|_| ()).unwrap()),
+            others: scorers,
+            num_docsets,
+        });
     }
+    Box::new(Intersection {
+        left,
+        right,
+        others: scorers,
+        num_docsets,
+    })
 }
 
 /// Creates a `DocSet` that iterator through the intersection of two `DocSet`s.
@@ -124,7 +118,6 @@ impl<TDocSet: DocSet, TOtherDocSet: DocSet> DocSet for Intersection<TDocSet, TOt
                         return false;
                     }
                 }
-
                 match left.skip_next(candidate) {
                     SkipResult::Reached => {
                         break;
@@ -140,35 +133,36 @@ impl<TDocSet: DocSet, TOtherDocSet: DocSet> DocSet for Intersection<TDocSet, TOt
             }
             // test the remaining scorers;
             for (ord, docset) in self.others.iter_mut().enumerate() {
-                if ord != other_candidate_ord {
-                    // `candidate_ord` is already at the
-                    // right position.
-                    //
-                    // Calling `skip_next` would advance this docset
-                    // and miss it.
-                    match docset.skip_next(candidate) {
-                        SkipResult::Reached => {}
-                        SkipResult::OverStep => {
-                            // this is not in the intersection,
-                            // let's update our candidate.
-                            candidate = docset.doc();
-                            match left.skip_next(candidate) {
-                                SkipResult::Reached => {
-                                    other_candidate_ord = ord;
-                                }
-                                SkipResult::OverStep => {
-                                    candidate = left.doc();
-                                    other_candidate_ord = usize::max_value();
-                                }
-                                SkipResult::End => {
-                                    return false;
-                                }
+                if ord == other_candidate_ord {
+                    continue;
+                }
+                // `candidate_ord` is already at the
+                // right position.
+                //
+                // Calling `skip_next` would advance this docset
+                // and miss it.
+                match docset.skip_next(candidate) {
+                    SkipResult::Reached => {}
+                    SkipResult::OverStep => {
+                        // this is not in the intersection,
+                        // let's update our candidate.
+                        candidate = docset.doc();
+                        match left.skip_next(candidate) {
+                            SkipResult::Reached => {
+                                other_candidate_ord = ord;
+                            }
+                            SkipResult::OverStep => {
+                                candidate = left.doc();
+                                other_candidate_ord = usize::max_value();
+                            }
+                            SkipResult::End => {
+                                return false;
                             }
-                            continue 'outer;
-                        }
-                        SkipResult::End => {
-                            return false;
                         }
+                        continue 'outer;
+                    }
+                    SkipResult::End => {
+                        return false;
                     }
                 }
             }

src/query/term_query/mod.rs

@@ -98,4 +98,20 @@ mod tests {
         }
     }
 
+    #[test]
+    fn test_term_query_count_when_there_are_deletes() {
+        let mut schema_builder = Schema::builder();
+        let text_field = schema_builder.add_text_field("text", TEXT);
+        let schema = schema_builder.build();
+        let index = Index::create_in_ram(schema);
+        let mut index_writer = index.writer_with_num_threads(1, 5_000_000).unwrap();
+        index_writer.add_document(doc!(text_field=>"a b"));
+        index_writer.add_document(doc!(text_field=>"a c"));
+        index_writer.delete_term(Term::from_field_text(text_field, "b"));
+        index_writer.commit().unwrap();
+        let term_a = Term::from_field_text(text_field, "a");
+        let term_query = TermQuery::new(term_a, IndexRecordOption::Basic);
+        let reader = index.reader().unwrap();
+        assert_eq!(term_query.count(&*reader.searcher()).unwrap(), 1);
+    }
 }

src/query/term_query/term_weight.rs

@@ -39,15 +39,15 @@ impl Weight for TermWeight {
     }
 
     fn count(&self, reader: &SegmentReader) -> Result<u32> {
-        if reader.num_deleted_docs() == 0 {
+        if let Some(delete_bitset) = reader.delete_bitset() {
+            Ok(self.scorer(reader)?.count(delete_bitset))
+        } else {
             let field = self.term.field();
             Ok(reader
                 .inverted_index(field)
                 .get_term_info(&self.term)
                 .map(|term_info| term_info.doc_freq)
                 .unwrap_or(0))
-        } else {
-            Ok(self.scorer(reader)?.count())
         }
     }
 }

src/query/union.rs

@@ -145,7 +145,7 @@ where
         }
     }
 
-    fn count(&mut self) -> u32 {
+    fn count_including_deleted(&mut self) -> u32 {
         let mut count = self.bitsets[self.cursor..HORIZON_NUM_TINYBITSETS]
             .iter()
             .map(|bitset| bitset.len())
@@ -163,6 +163,8 @@ where
         count
     }
 
+    // TODO implement `count` efficiently.
+
     fn skip_next(&mut self, target: DocId) -> SkipResult {
         if !self.advance() {
             return SkipResult::End;
@@ -300,7 +302,7 @@ mod tests {
             count += 1;
         }
         assert!(!union_expected.advance());
-        assert_eq!(count, make_union().count());
+        assert_eq!(count, make_union().count_including_deleted());
     }
 
     #[test]

src/query/weight.rs

@@ -13,6 +13,11 @@ pub trait Weight: Send + Sync + 'static {
 
     /// Returns the number documents within the given `SegmentReader`.
     fn count(&self, reader: &SegmentReader) -> Result<u32> {
-        Ok(self.scorer(reader)?.count())
+        let mut scorer = self.scorer(reader)?;
+        if let Some(delete_bitset) = reader.delete_bitset() {
+            Ok(scorer.count(delete_bitset))
+        } else {
+            Ok(scorer.count_including_deleted())
+        }
     }
 }

src/tokenizer/mod.rs

@@ -97,6 +97,8 @@
 //! If you built your schema programmatically, a complete example
 //! could like this for instance.
 //!
+//! Note that tokens with a len greater or equal to [`MAX_TOKEN_LEN`](./constant.MAX_TOKEN_LEN.html).
+//!
 //! # Example
 //!
 //! ```
@@ -157,6 +159,13 @@ pub use self::tokenizer::BoxedTokenizer;
 pub use self::tokenizer::{Token, TokenFilter, TokenStream, Tokenizer};
 pub use self::tokenizer_manager::TokenizerManager;
 
+/// Maximum authorized len (in bytes) for a token.
+///
+/// Tokenizer are in charge of not emitting tokens larger than this value.
+/// Currently, if a faulty tokenizer implementation emits tokens with a length larger than
+/// `2^16 - 1 - 4`, the token will simply be ignored downstream.
+pub const MAX_TOKEN_LEN: usize = u16::max_value() as usize - 4;
+
 #[cfg(test)]
 pub mod tests {
     use super::{
@@ -228,27 +237,27 @@ pub mod tests {
     fn test_non_en_tokenizer() {
         let tokenizer_manager = TokenizerManager::default();
         tokenizer_manager.register(
-            "es_stem",
+            "el_stem",
             SimpleTokenizer
                 .filter(RemoveLongFilter::limit(40))
                 .filter(LowerCaser)
-                .filter(Stemmer::new(Language::Spanish)),
+                .filter(Stemmer::new(Language::Greek)),
         );
-        let en_tokenizer = tokenizer_manager.get("es_stem").unwrap();
+        let en_tokenizer = tokenizer_manager.get("el_stem").unwrap();
         let mut tokens: Vec<Token> = vec![];
         {
             let mut add_token = |token: &Token| {
                 tokens.push(token.clone());
             };
             en_tokenizer
-                .token_stream("Hola, feliz contribuyente!")
+                .token_stream("Καλημέρα, χαρούμενε φορολογούμενε!")
                 .process(&mut add_token);
         }
 
         assert_eq!(tokens.len(), 3);
-        assert_token(&tokens[0], 0, "hola", 0, 4);
-        assert_token(&tokens[1], 1, "feliz", 6, 11);
-        assert_token(&tokens[2], 2, "contribuyent", 12, 25);
+        assert_token(&tokens[0], 0, "καλημερ", 0, 16);
+        assert_token(&tokens[1], 1, "χαρουμεν", 18, 36);
+        assert_token(&tokens[2], 2, "φορολογουμεν", 37, 63);
     }
 
     #[test]

src/tokenizer/stemmer.rs

@@ -2,7 +2,6 @@
 
 use super::{Token, TokenFilter, TokenStream};
 use rust_stemmers::{self, Algorithm};
-use std::sync::Arc;
 
 /// Available stemmer languages.
 #[derive(Debug, Serialize, Deserialize, Eq, PartialEq, Copy, Clone)]
@@ -57,14 +56,14 @@ impl Language {
 /// Tokens are expected to be lowercased beforehand.
 #[derive(Clone)]
 pub struct Stemmer {
-    stemmer_algorithm: Arc<Algorithm>,
+    stemmer_algorithm: Algorithm,
 }
 
 impl Stemmer {
     /// Creates a new Stemmer `TokenFilter` for a given language algorithm.
     pub fn new(language: Language) -> Stemmer {
         Stemmer {
-            stemmer_algorithm: Arc::new(language.algorithm()),
+            stemmer_algorithm: language.algorithm(),
         }
     }
 }
@@ -83,7 +82,7 @@ where
     type ResultTokenStream = StemmerTokenStream<TailTokenStream>;
 
     fn transform(&self, token_stream: TailTokenStream) -> Self::ResultTokenStream {
-        let inner_stemmer = rust_stemmers::Stemmer::create(Algorithm::English);
+        let inner_stemmer = rust_stemmers::Stemmer::create(self.stemmer_algorithm);
         StemmerTokenStream::wrap(inner_stemmer, token_stream)
     }
 }