Introducing a column trait

And renaming the `Serializer` trait `FastFieldCodec`.

fastfield_codecs/src/blockwise_linear.rs

@@ -300,9 +300,13 @@ impl FastFieldCodec for BlockwiseLinearCodec {
         // If this doesn't overflow the algorithm should be fine
         let theorethical_maximum_offset =
             fastfield_accessor.max_value() - fastfield_accessor.min_value();
-        fastfield_accessor
+        if fastfield_accessor
             .max_value()
-            .checked_add(theorethical_maximum_offset)?;
+            .checked_add(theorethical_maximum_offset)
+            .is_none()
+        {
+            return None;
+        }
 
         let first_val_in_first_block = fastfield_accessor.get_val(0);
         let last_elem_in_first_chunk = CHUNK_SIZE.min(fastfield_accessor.num_vals());

fastfield_codecs/src/column.rs

@@ -1,7 +1,3 @@
-use std::ops::Range;
-
-use crate::ColumnIter;
-
 pub trait Column<T = u64> {
     /// Return the value associated to the given idx.
     ///
@@ -12,18 +8,23 @@ pub trait Column<T = u64> {
     /// May panic if `idx` is greater than the column length.
     fn get_val(&self, idx: u64) -> T;
 
-    /// Returns an iterator over given doc range.
+    /// Fills an output buffer with the fast field values
+    /// associated with the `DocId` going from
+    /// `start` to `start + output.len()`.
+    ///
+    /// Regardless of the type of `Item`, this method works
+    /// - transmuting the output array
+    /// - extracting the `Item`s as if they were `u64`
+    /// - possibly converting the `u64` value to the right type.
     ///
     /// # Panics
     ///
-    /// May panic if `range.end()` is greater than
+    /// May panic if `start + output.len()` is greater than
     /// the segment's `maxdoc`.
-    #[inline]
-    fn get_range(&self, range: Range<u64>) -> ColumnIter<'_, Self, T>
-    where
-        Self: Sized,
-    {
-        ColumnIter::new(self, range)
+    fn get_range(&self, start: u64, output: &mut [T]) {
+        for (out, idx) in output.iter_mut().zip(start..) {
+            *out = self.get_val(idx);
+        }
     }
 
     /// Returns the minimum value for this fast field.

fastfield_codecs/src/lib.rs

@@ -4,9 +4,6 @@ extern crate more_asserts;
 
 use std::io;
 use std::io::Write;
-use std::iter::FusedIterator;
-use std::marker::PhantomData;
-use std::ops::Range;
 
 use common::BinarySerializable;
 use ownedbytes::OwnedBytes;
@@ -19,6 +16,7 @@ mod column;
 
 pub use self::column::Column;
 
+
 #[derive(PartialEq, Eq, PartialOrd, Ord, Debug, Clone, Copy)]
 #[repr(u8)]
 pub enum FastFieldCodecType {
@@ -115,56 +113,6 @@ impl<'a> Column for &'a [u64] {
     }
 }
 
-pub struct ColumnIter<'a, C: Column<I>, I> {
-    column: &'a C,
-    range: Range<u64>,
-    _phantom: PhantomData<I>,
-}
-
-impl<'a, C: Column<I>, I> ColumnIter<'a, C, I> {
-    #[inline]
-    pub fn new(col: &'a C, range: Range<u64>) -> Self {
-        Self {
-            column: col,
-            range,
-            _phantom: PhantomData,
-        }
-    }
-}
-
-impl<'a, C: Column<I>, I> Iterator for ColumnIter<'a, C, I> {
-    type Item = I;
-
-    #[inline]
-    fn next(&mut self) -> Option<Self::Item> {
-        Some(self.column.get_val(self.range.next()?))
-    }
-
-    #[inline]
-    fn fold<Acc, G>(self, init: Acc, mut g: G) -> Acc
-    where
-        G: FnMut(Acc, Self::Item) -> Acc,
-    {
-        self.range
-            .fold(init, move |acc, idx| g(acc, self.column.get_val(idx)))
-    }
-
-    #[inline]
-    fn size_hint(&self) -> (usize, Option<usize>) {
-        let size = (self.range.end - self.range.start) as usize;
-        (size, Some(size))
-    }
-}
-
-impl<'a, C: Column<I>, I> ExactSizeIterator for ColumnIter<'a, C, I> {
-    #[inline]
-    fn len(&self) -> usize {
-        let size = (self.range.end - self.range.start) as usize;
-        size as usize
-    }
-}
-impl<'a, C: Column<I>, I> FusedIterator for ColumnIter<'a, C, I> {}
-
 impl Column for Vec<u64> {
     fn get_val(&self, position: u64) -> u64 {
         self[position as usize]
@@ -248,11 +196,6 @@ mod tests {
         data_and_names.push((vec![5, 50, 3, 13, 1, 1000, 35], "rand small"));
         data_and_names.push((vec![10], "single value"));
 
-        data_and_names.push((
-            vec![1572656989877777, 1170935903116329, 720575940379279, 0],
-            "overflow error",
-        ));
-
         data_and_names
     }
 

fastfield_codecs/src/linear.rs

@@ -115,9 +115,9 @@ fn diff(val1: u64, val2: u64) -> f64 {
 #[inline]
 pub fn get_calculated_value(first_val: u64, pos: u64, slope: f32) -> u64 {
     if slope < 0.0 {
-        first_val.saturating_sub((pos as f32 * -slope) as u64)
+        first_val - (pos as f32 * -slope) as u64
     } else {
-        first_val.saturating_add((pos as f32 * slope) as u64)
+        first_val + (pos as f32 * slope) as u64
     }
 }
 
@@ -204,9 +204,13 @@ impl FastFieldCodec for LinearCodec {
         // If this doesn't overflow the algorithm should be fine
         let theorethical_maximum_offset =
             fastfield_accessor.max_value() - fastfield_accessor.min_value();
-        fastfield_accessor
+        if fastfield_accessor
             .max_value()
-            .checked_add(theorethical_maximum_offset)?;
+            .checked_add(theorethical_maximum_offset)
+            .is_none()
+        {
+            return None;
+        }
 
         let first_val = fastfield_accessor.get_val(0);
         let last_val = fastfield_accessor.get_val(fastfield_accessor.num_vals() as u64 - 1);
@@ -310,13 +314,6 @@ mod tests {
 
         create_and_validate(&data, "large amplitude");
     }
-
-    #[test]
-    fn overflow_error_test() {
-        let data = vec![1572656989877777, 1170935903116329, 720575940379279, 0];
-        create_and_validate(&data, "overflow test");
-    }
-
     #[test]
     fn linear_interpol_fast_concave_data() {
         let data = vec![0, 1, 2, 5, 8, 10, 20, 50];

src/aggregation/bucket/term_agg.rs

@@ -242,13 +242,13 @@ impl TermBuckets {
 
     fn increment_bucket(
         &mut self,
-        term_ids: impl Iterator<Item = u64>,
+        term_ids: &[u64],
         doc: DocId,
         sub_aggregation: &AggregationsWithAccessor,
         bucket_count: &BucketCount,
         blueprint: &Option<SegmentAggregationResultsCollector>,
     ) -> crate::Result<()> {
-        for term_id in term_ids {
+        for &term_id in term_ids {
             let entry = self.entries.entry(term_id as u32).or_insert_with(|| {
                 bucket_count.add_count(1);
 
@@ -432,30 +432,39 @@ impl SegmentTermCollector {
             .as_multi()
             .expect("unexpected fast field cardinatility");
         let mut iter = doc.chunks_exact(4);
+        let mut vals1 = vec![];
+        let mut vals2 = vec![];
+        let mut vals3 = vec![];
+        let mut vals4 = vec![];
         for docs in iter.by_ref() {
+            accessor.get_vals(docs[0], &mut vals1);
+            accessor.get_vals(docs[1], &mut vals2);
+            accessor.get_vals(docs[2], &mut vals3);
+            accessor.get_vals(docs[3], &mut vals4);
+
             self.term_buckets.increment_bucket(
-                accessor.get_vals(docs[0]),
+                &vals1,
                 docs[0],
                 &bucket_with_accessor.sub_aggregation,
                 &bucket_with_accessor.bucket_count,
                 &self.blueprint,
             )?;
             self.term_buckets.increment_bucket(
-                accessor.get_vals(docs[1]),
+                &vals2,
                 docs[1],
                 &bucket_with_accessor.sub_aggregation,
                 &bucket_with_accessor.bucket_count,
                 &self.blueprint,
             )?;
             self.term_buckets.increment_bucket(
-                accessor.get_vals(docs[2]),
+                &vals3,
                 docs[2],
                 &bucket_with_accessor.sub_aggregation,
                 &bucket_with_accessor.bucket_count,
                 &self.blueprint,
             )?;
             self.term_buckets.increment_bucket(
-                accessor.get_vals(docs[3]),
+                &vals4,
                 docs[3],
                 &bucket_with_accessor.sub_aggregation,
                 &bucket_with_accessor.bucket_count,
@@ -463,8 +472,10 @@ impl SegmentTermCollector {
             )?;
         }
         for &doc in iter.remainder() {
+            accessor.get_vals(doc, &mut vals1);
+
             self.term_buckets.increment_bucket(
-                accessor.get_vals(doc),
+                &vals1,
                 doc,
                 &bucket_with_accessor.sub_aggregation,
                 &bucket_with_accessor.bucket_count,
@@ -1323,15 +1334,11 @@ mod bench {
             max_bucket_count: 1_000_001u32,
         };
         b.iter(|| {
-            collector
-                .increment_bucket(
-                    vals.iter().cloned(),
-                    0,
-                    &aggregations_with_accessor,
-                    &bucket_count,
-                    &None,
-                )
-                .unwrap();
+            for &val in &vals {
+                collector
+                    .increment_bucket(&[val], 0, &aggregations_with_accessor, &bucket_count, &None)
+                    .unwrap();
+            }
         })
     }
 

src/collector/top_score_collector.rs

@@ -409,7 +409,7 @@ impl TopDocs {
     /// # use tantivy::query::QueryParser;
     /// use tantivy::SegmentReader;
     /// use tantivy::collector::TopDocs;
-    /// use tantivy::fastfield::Column;
+    /// use tantivy::fastfield::FastFieldReader;
     /// use tantivy::schema::Field;
     ///
     /// fn create_schema() -> Schema {
@@ -458,7 +458,7 @@ impl TopDocs {
     ///
     ///             // We can now define our actual scoring function
     ///             move |doc: DocId, original_score: Score| {
-    ///                 let popularity: u64 = popularity_reader.get_val(doc as u64);
+    ///                 let popularity: u64 = popularity_reader.get(doc);
     ///                 // Well.. For the sake of the example we use a simple logarithm
     ///                 // function.
     ///                 let popularity_boost_score = ((2u64 + popularity) as Score).log2();
@@ -517,7 +517,7 @@ impl TopDocs {
     /// use tantivy::SegmentReader;
     /// use tantivy::collector::TopDocs;
     /// use tantivy::schema::Field;
-    /// use fastfield_codecs::Column;
+    /// use tantivy::fastfield::FastFieldReader;
     ///
     /// # fn create_schema() -> Schema {
     /// #    let mut schema_builder = Schema::builder();
@@ -569,8 +569,8 @@ impl TopDocs {
     ///
     ///             // We can now define our actual scoring function
     ///             move |doc: DocId| {
-    ///                 let popularity: u64 = popularity_reader.get_val(doc as u64);
-    ///                 let boosted: u64 = boosted_reader.get_val(doc as u64);
+    ///                 let popularity: u64 = popularity_reader.get(doc);
+    ///                 let boosted: u64 = boosted_reader.get(doc);
     ///                 // Score do not have to be `f64` in tantivy.
     ///                 // Here we return a couple to get lexicographical order
     ///                 // for free.

src/fastfield/facet_reader.rs

@@ -76,8 +76,7 @@ impl FacetReader {
 
     /// Return the list of facet ordinals associated to a document.
     pub fn facet_ords(&self, doc: DocId, output: &mut Vec<u64>) {
-        output.clear();
-        output.extend(self.term_ords.get_vals(doc))
+        self.term_ords.get_vals(doc, output);
     }
 }
 

src/fastfield/gcd.rs

@@ -210,7 +210,7 @@ mod tests {
     #[test]
     fn test_fastfield_gcd_i64() -> crate::Result<()> {
         for &code_type in ALL_CODECS {
-            test_fastfield_gcd_i64_with_codec(code_type, 5500)?;
+            test_fastfield_gcd_i64_with_codec(code_type, 5005)?;
         }
         Ok(())
     }
@@ -251,7 +251,7 @@ mod tests {
     #[test]
     fn test_fastfield_gcd_u64() -> crate::Result<()> {
         for &code_type in ALL_CODECS {
-            test_fastfield_gcd_u64_with_codec(code_type, 5500)?;
+            test_fastfield_gcd_u64_with_codec(code_type, 5005)?;
         }
         Ok(())
     }

src/fastfield/mod.rs

@@ -477,7 +477,8 @@ mod tests {
             for (doc, i) in (-100i64..10_000i64).enumerate() {
                 assert_eq!(fast_field_reader.get_val(doc as u64), i);
             }
-            let buffer: Vec<i64> = fast_field_reader.get_range(53..154).collect();
+            let mut buffer = vec![0i64; 100];
+            fast_field_reader.get_range(53, &mut buffer[..]);
             for i in 0..100 {
                 assert_eq!(buffer[i], -100i64 + 53i64 + i as i64);
             }
@@ -606,7 +607,9 @@ mod tests {
         let mut all = vec![];
 
         for doc in docs {
-            all.extend(ff.get_vals(doc));
+            let mut out = vec![];
+            ff.get_vals(doc, &mut out);
+            all.extend(out);
         }
         all
     }
@@ -651,7 +654,8 @@ mod tests {
                 vec![1, 0, 0, 0, 1, 2]
             );
 
-            let out = text_fast_field.get_vals(3u32).collect::<Vec<_>>();
+            let mut out = vec![];
+            text_fast_field.get_vals(3, &mut out);
             assert_eq!(out, vec![0, 1]);
 
             let inverted_index = segment_reader.inverted_index(text_field)?;
@@ -836,20 +840,22 @@ mod tests {
         let fast_fields = segment_reader.fast_fields();
         let date_fast_field = fast_fields.date(date_field).unwrap();
         let dates_fast_field = fast_fields.dates(multi_date_field).unwrap();
+        let mut dates = vec![];
         {
             assert_eq!(date_fast_field.get_val(0).into_timestamp_micros(), 1i64);
-            let dates = dates_fast_field.get_vals(0u32).collect::<Vec<_>>();
+            dates_fast_field.get_vals(0u32, &mut dates);
             assert_eq!(dates.len(), 2);
             assert_eq!(dates[0].into_timestamp_micros(), 2i64);
             assert_eq!(dates[1].into_timestamp_micros(), 3i64);
         }
         {
             assert_eq!(date_fast_field.get_val(1).into_timestamp_micros(), 4i64);
-            assert!(dates_fast_field.get_vals(1u32).next().is_none());
+            dates_fast_field.get_vals(1u32, &mut dates);
+            assert!(dates.is_empty());
         }
         {
             assert_eq!(date_fast_field.get_val(2).into_timestamp_micros(), 0i64);
-            let dates = dates_fast_field.get_vals(2u32).collect::<Vec<_>>();
+            dates_fast_field.get_vals(2u32, &mut dates);
             assert_eq!(dates.len(), 2);
             assert_eq!(dates[0].into_timestamp_micros(), 5i64);
             assert_eq!(dates[1].into_timestamp_micros(), 6i64);
@@ -975,74 +981,13 @@ mod bench {
     use std::collections::HashMap;
     use std::path::Path;
 
-    use fastfield_codecs::Column;
     use test::{self, Bencher};
 
     use super::tests::{generate_permutation, FIELD, SCHEMA};
     use super::*;
     use crate::directory::{CompositeFile, Directory, RamDirectory, WritePtr};
     use crate::fastfield::tests::generate_permutation_gcd;
-    use crate::schema::{NumericOptions, Schema};
-    use crate::Document;
-
-    fn multi_values(num_docs: usize, vals_per_doc: usize) -> Vec<Vec<u64>> {
-        let mut vals = vec![];
-        for _i in 0..num_docs {
-            let mut block = vec![];
-            for j in 0..vals_per_doc {
-                block.push(j as u64);
-            }
-            vals.push(block);
-        }
-
-        vals
-    }
-
-    #[bench]
-    fn bench_multi_value_fflookup(b: &mut Bencher) {
-        let num_docs = 100_000;
-
-        let path = Path::new("test");
-        let directory: RamDirectory = RamDirectory::create();
-        {
-            let options = NumericOptions::default().set_fast(Cardinality::MultiValues);
-            let mut schema_builder = Schema::builder();
-            let field = schema_builder.add_u64_field("field", options);
-            let schema = schema_builder.build();
-
-            let write: WritePtr = directory.open_write(Path::new("test")).unwrap();
-            let mut serializer = CompositeFastFieldSerializer::from_write(write).unwrap();
-            let mut fast_field_writers = FastFieldsWriter::from_schema(&schema);
-            for block in &multi_values(num_docs, 3) {
-                let mut doc = Document::new();
-                for val in block {
-                    doc.add_u64(field, *val);
-                }
-                fast_field_writers.add_document(&doc);
-            }
-            fast_field_writers
-                .serialize(&mut serializer, &HashMap::new(), None)
-                .unwrap();
-            serializer.close().unwrap();
-        }
-        let file = directory.open_read(&path).unwrap();
-        {
-            let fast_fields_composite = CompositeFile::open(&file).unwrap();
-            let data_idx = fast_fields_composite.open_read_with_idx(*FIELD, 0).unwrap();
-            let idx_reader = DynamicFastFieldReader::<u64>::open(data_idx).unwrap();
-
-            let data_vals = fast_fields_composite.open_read_with_idx(*FIELD, 1).unwrap();
-            let vals_reader = DynamicFastFieldReader::<u64>::open(data_vals).unwrap();
-            let fast_field_reader = MultiValuedFastFieldReader::open(idx_reader, vals_reader);
-            b.iter(|| {
-                let mut sum = 0u64;
-                for i in 0u32..num_docs as u32 {
-                    sum += fast_field_reader.get_vals(i).sum::<u64>();
-                }
-                sum
-            });
-        }
-    }
+    use crate::fastfield::FastFieldReader;
 
     #[bench]
     fn bench_intfastfield_linear_veclookup(b: &mut Bencher) {
@@ -1097,7 +1042,7 @@ mod bench {
                 let n = test::black_box(7000u32);
                 let mut a = 0u64;
                 for i in (0u32..n / 7).map(|val| val * 7) {
-                    a ^= fast_field_reader.get_val(i as u64);
+                    a ^= fast_field_reader.get(i);
                 }
                 a
             });
@@ -1129,8 +1074,8 @@ mod bench {
 
             b.iter(|| {
                 let mut a = 0u32;
-                for i in 0u64..permutation.len() as u64 {
-                    a = fast_field_reader.get_val(i) as u32;
+                for i in 0u32..permutation.len() as u32 {
+                    a = fast_field_reader.get(i) as u32;
                 }
                 a
             });
@@ -1163,7 +1108,7 @@ mod bench {
             b.iter(|| {
                 let mut a = 0u32;
                 for i in 0u32..permutation.len() as u32 {
-                    a = fast_field_reader.get_val(i as u64) as u32;
+                    a = fast_field_reader.get(i) as u32;
                 }
                 a
             });

src/fastfield/multivalued/mod.rs

@@ -36,17 +36,19 @@ mod tests {
 
         let searcher = index.reader()?.searcher();
         let segment_reader = searcher.segment_reader(0);
+        let mut vals = Vec::new();
         let multi_value_reader = segment_reader.fast_fields().u64s(field)?;
         {
-            let vals = multi_value_reader.get_vals(2u32).collect::<Vec<_>>();
+            multi_value_reader.get_vals(2, &mut vals);
             assert_eq!(&vals, &[4u64]);
         }
         {
-            let vals = multi_value_reader.get_vals(0u32).collect::<Vec<_>>();
+            multi_value_reader.get_vals(0, &mut vals);
             assert_eq!(&vals, &[1u64, 3u64]);
         }
         {
-            assert!(multi_value_reader.get_vals(1u32).next().is_none());
+            multi_value_reader.get_vals(1, &mut vals);
+            assert!(vals.is_empty());
         }
         Ok(())
     }
@@ -211,13 +213,15 @@ mod tests {
 
         let searcher = index.reader()?.searcher();
         let segment_reader = searcher.segment_reader(0);
+        let mut vals = Vec::new();
         let multi_value_reader = segment_reader.fast_fields().i64s(field).unwrap();
-        let vals = multi_value_reader.get_vals(2u32).collect::<Vec<_>>();
+        multi_value_reader.get_vals(2, &mut vals);
         assert_eq!(&vals, &[-4i64]);
-        let vals = multi_value_reader.get_vals(0u32).collect::<Vec<_>>();
+        multi_value_reader.get_vals(0, &mut vals);
         assert_eq!(&vals, &[1i64, 3i64]);
-        assert!(multi_value_reader.get_vals(1u32).next().is_none());
-        let vals = multi_value_reader.get_vals(3u32).collect::<Vec<_>>();
+        multi_value_reader.get_vals(1, &mut vals);
+        assert!(vals.is_empty());
+        multi_value_reader.get_vals(3, &mut vals);
         assert_eq!(&vals, &[-5i64, -20i64, 1i64]);
         Ok(())
     }
@@ -241,13 +245,15 @@ mod tests {
 
         let searcher = index.reader()?.searcher();
         let segment_reader = searcher.segment_reader(0);
+        let mut vals = Vec::new();
         let multi_value_reader = segment_reader.fast_fields().bools(bool_field).unwrap();
-        let vals = multi_value_reader.get_vals(2u32).collect::<Vec<_>>();
+        multi_value_reader.get_vals(2, &mut vals);
         assert_eq!(&vals, &[false]);
-        let vals = multi_value_reader.get_vals(0u32).collect::<Vec<_>>();
+        multi_value_reader.get_vals(0, &mut vals);
         assert_eq!(&vals, &[true, false]);
-        assert!(multi_value_reader.get_vals(1u32).next().is_none());
-        let vals = multi_value_reader.get_vals(3u32).collect::<Vec<_>>();
+        multi_value_reader.get_vals(1, &mut vals);
+        assert!(vals.is_empty());
+        multi_value_reader.get_vals(3, &mut vals);
         assert_eq!(&vals, &[true, true, false]);
         Ok(())
     }

src/fastfield/multivalued/reader.rs

@@ -18,9 +18,7 @@ pub struct MultiValuedFastFieldReader<Item: FastValue> {
     vals_reader: DynamicFastFieldReader<Item>,
 }
 
-impl<Item: FastValue> MultiValuedFastFieldReader<Item>
-where DynamicFastFieldReader<Item>: Column<Item>
-{
+impl<Item: FastValue> MultiValuedFastFieldReader<Item> {
     pub(crate) fn open(
         idx_reader: DynamicFastFieldReader<u64>,
         vals_reader: DynamicFastFieldReader<Item>,
@@ -43,9 +41,17 @@ where DynamicFastFieldReader<Item>: Column<Item>
 
     /// Returns the array of values associated to the given `doc`.
     #[inline]
-    pub fn get_vals(&self, doc: DocId) -> impl Iterator<Item = Item> + '_ {
+    fn get_vals_for_range(&self, range: Range<u64>, vals: &mut Vec<Item>) {
+        let len = (range.end - range.start) as usize;
+        vals.resize(len, Item::make_zero());
+        self.vals_reader.get_range(range.start, &mut vals[..]);
+    }
+
+    /// Returns the array of values associated to the given `doc`.
+    #[inline]
+    pub fn get_vals(&self, doc: DocId, vals: &mut Vec<Item>) {
         let range = self.range(doc);
-        self.vals_reader.get_range(range)
+        self.get_vals_for_range(range, vals);
     }
 
     /// Returns the minimum value for this fast field.

src/fastfield/reader.rs

@@ -40,39 +40,40 @@ impl<Item: FastValue> DynamicFastFieldReader<Item> {
         mut bytes: OwnedBytes,
         codec_type: FastFieldCodecType,
     ) -> crate::Result<DynamicFastFieldReader<Item>> {
-        let reader =
-            match codec_type {
-                FastFieldCodecType::Bitpacked => DynamicFastFieldReader::Bitpacked(
-                    BitpackedCodec::open_from_bytes(bytes)?.into(),
-                ),
-                FastFieldCodecType::Linear => {
-                    DynamicFastFieldReader::Linear(LinearCodec::open_from_bytes(bytes)?.into())
-                }
-                FastFieldCodecType::BlockwiseLinear => DynamicFastFieldReader::BlockwiseLinear(
-                    BlockwiseLinearCodec::open_from_bytes(bytes)?.into(),
-                ),
-                FastFieldCodecType::Gcd => {
-                    let codec_type = FastFieldCodecType::deserialize(&mut bytes)?;
-                    match codec_type {
-                        FastFieldCodecType::Bitpacked => DynamicFastFieldReader::BitpackedGCD(
-                            open_gcd_from_bytes::<BitpackedCodec>(bytes)?.into(),
-                        ),
-                        FastFieldCodecType::Linear => DynamicFastFieldReader::LinearGCD(
-                            open_gcd_from_bytes::<LinearCodec>(bytes)?.into(),
-                        ),
-                        FastFieldCodecType::BlockwiseLinear => {
-                            DynamicFastFieldReader::BlockwiseLinearGCD(
-                                open_gcd_from_bytes::<BlockwiseLinearCodec>(bytes)?.into(),
-                            )
-                        }
-                        FastFieldCodecType::Gcd => return Err(DataCorruption::comment_only(
+        let reader = match codec_type {
+            FastFieldCodecType::Bitpacked => {
+                DynamicFastFieldReader::Bitpacked(BitpackedCodec::open_from_bytes(bytes)?.into())
+            }
+            FastFieldCodecType::Linear => {
+                DynamicFastFieldReader::Linear(LinearCodec::open_from_bytes(bytes)?.into())
+            }
+            FastFieldCodecType::BlockwiseLinear => DynamicFastFieldReader::BlockwiseLinear(
+                BlockwiseLinearCodec::open_from_bytes(bytes)?.into(),
+            ),
+            FastFieldCodecType::Gcd => {
+                let codec_type = FastFieldCodecType::deserialize(&mut bytes)?;
+                match codec_type {
+                    FastFieldCodecType::Bitpacked => DynamicFastFieldReader::BitpackedGCD(
+                        open_gcd_from_bytes::<BitpackedCodec>(bytes)?.into(),
+                    ),
+                    FastFieldCodecType::Linear => DynamicFastFieldReader::LinearGCD(
+                        open_gcd_from_bytes::<LinearCodec>(bytes)?.into(),
+                    ),
+                    FastFieldCodecType::BlockwiseLinear => {
+                        DynamicFastFieldReader::BlockwiseLinearGCD(
+                            open_gcd_from_bytes::<BlockwiseLinearCodec>(bytes)?.into(),
+                        )
+                    }
+                    FastFieldCodecType::Gcd => {
+                        return Err(DataCorruption::comment_only(
                             "Gcd codec wrapped into another gcd codec. This combination is not \
                              allowed.",
                         )
-                        .into()),
+                        .into())
                     }
                 }
-            };
+            }
+        };
         Ok(reader)
     }
 
@@ -96,6 +97,17 @@ impl<Item: FastValue> Column<Item> for DynamicFastFieldReader<Item> {
             Self::BlockwiseLinearGCD(reader) => reader.get_val(idx),
         }
     }
+    #[inline]
+    fn get_range(&self, start: u64, output: &mut [Item]) {
+        match self {
+            Self::Bitpacked(reader) => reader.get_range(start, output),
+            Self::Linear(reader) => reader.get_range(start, output),
+            Self::BlockwiseLinear(reader) => reader.get_range(start, output),
+            Self::BitpackedGCD(reader) => reader.get_range(start, output),
+            Self::LinearGCD(reader) => reader.get_range(start, output),
+            Self::BlockwiseLinearGCD(reader) => reader.get_range(start, output),
+        }
+    }
     fn min_value(&self) -> Item {
         match self {
             Self::Bitpacked(reader) => reader.min_value(),
@@ -155,6 +167,24 @@ impl<Item: FastValue, D: Column> FastFieldReaderCodecWrapper<Item, D> {
         let data = self.reader.get_val(idx);
         Item::from_u64(data)
     }
+
+    /// Internally `multivalued` also use SingleValue Fast fields.
+    /// It works as follows... A first column contains the list of start index
+    /// for each document, a second column contains the actual values.
+    ///
+    /// The values associated to a given doc, are then
+    ///  `second_column[first_column.get(doc)..first_column.get(doc+1)]`.
+    ///
+    /// Which means single value fast field reader can be indexed internally with
+    /// something different from a `DocId`. For this use case, we want to use `u64`
+    /// values.
+    ///
+    /// See `get_range` for an actual documentation about this method.
+    pub(crate) fn get_range_u64(&self, start: u64, output: &mut [Item]) {
+        for (i, out) in output.iter_mut().enumerate() {
+            *out = self.get_u64(start + (i as u64));
+        }
+    }
 }
 
 impl<Item: FastValue, C: Column + Clone> Column<Item> for FastFieldReaderCodecWrapper<Item, C> {
@@ -170,6 +200,23 @@ impl<Item: FastValue, C: Column + Clone> Column<Item> for FastFieldReaderCodecWr
         self.get_u64(idx)
     }
 
+    /// Fills an output buffer with the fast field values
+    /// associated with the `DocId` going from
+    /// `start` to `start + output.len()`.
+    ///
+    /// Regardless of the type of `Item`, this method works
+    /// - transmuting the output array
+    /// - extracting the `Item`s as if they were `u64`
+    /// - possibly converting the `u64` value to the right type.
+    ///
+    /// # Panics
+    ///
+    /// May panic if `start + output.len()` is greater than
+    /// the segment's `maxdoc`.
+    fn get_range(&self, start: u64, output: &mut [Item]) {
+        self.get_range_u64(start, output);
+    }
+
     /// Returns the minimum value for this fast field.
     ///
     /// The max value does not take in account of possible

src/fastfield/serializer/mod.rs

@@ -64,8 +64,8 @@ impl From<FastFieldCodecType> for FastFieldCodecEnableCheck {
 
 // use this, when this is merged and stabilized explicit_generic_args_with_impl_trait
 // https://github.com/rust-lang/rust/pull/86176
-fn codec_estimation<C: FastFieldCodec, D: Column>(
-    fastfield_accessor: &D,
+fn codec_estimation<C: FastFieldCodec>(
+    fastfield_accessor: &impl Column,
     estimations: &mut Vec<(f32, FastFieldCodecType)>,
 ) {
     if let Some(ratio) = C::estimate(fastfield_accessor) {
@@ -202,13 +202,13 @@ impl CompositeFastFieldSerializer {
         let mut estimations = vec![];
 
         if codec_enable_checker.is_enabled(FastFieldCodecType::Bitpacked) {
-            codec_estimation::<BitpackedCodec, _>(&fastfield_accessor, &mut estimations);
+            codec_estimation::<BitpackedCodec>(&fastfield_accessor, &mut estimations);
         }
         if codec_enable_checker.is_enabled(FastFieldCodecType::Linear) {
-            codec_estimation::<LinearCodec, _>(&fastfield_accessor, &mut estimations);
+            codec_estimation::<LinearCodec>(&fastfield_accessor, &mut estimations);
         }
         if codec_enable_checker.is_enabled(FastFieldCodecType::BlockwiseLinear) {
-            codec_estimation::<BlockwiseLinearCodec, _>(&fastfield_accessor, &mut estimations);
+            codec_estimation::<BlockwiseLinearCodec>(&fastfield_accessor, &mut estimations);
         }
         if let Some(broken_estimation) = estimations.iter().find(|estimation| estimation.0.is_nan())
         {

src/indexer/doc_id_mapping.rs

@@ -471,13 +471,15 @@ mod tests_indexsorting {
 
         let multi_numbers = index.schema().get_field("multi_numbers").unwrap();
         let multifield = fast_fields.u64s(multi_numbers).unwrap();
-        let vals = multifield.get_vals(0u32).collect::<Vec<_>>();
+        let mut vals = vec![];
+        multifield.get_vals(0u32, &mut vals);
         assert_eq!(vals, &[] as &[u64]);
-
-        let vals = multifield.get_vals(1u32).collect::<Vec<_>>();
+        let mut vals = vec![];
+        multifield.get_vals(1u32, &mut vals);
         assert_eq!(vals, &[5, 6]);
 
-        let vals = multifield.get_vals(2u32).collect::<Vec<_>>();
+        let mut vals = vec![];
+        multifield.get_vals(2u32, &mut vals);
         assert_eq!(vals, &[3]);
         Ok(())
     }

src/indexer/index_writer.rs

@@ -1535,11 +1535,13 @@ mod tests {
             let ff_reader = segment_reader.fast_fields().u64s(multi_numbers).unwrap();
             let bool_ff_reader = segment_reader.fast_fields().bools(multi_bools).unwrap();
             for doc in segment_reader.doc_ids_alive() {
-                let vals = ff_reader.get_vals(doc).collect::<Vec<_>>();
+                let mut vals = vec![];
+                ff_reader.get_vals(doc, &mut vals);
                 assert_eq!(vals.len(), 2);
                 assert_eq!(vals[0], vals[1]);
 
-                let bool_vals = bool_ff_reader.get_vals(doc).collect::<Vec<_>>();
+                let mut bool_vals = vec![];
+                bool_ff_reader.get_vals(doc, &mut bool_vals);
                 assert_eq!(bool_vals.len(), 2);
                 assert_ne!(bool_vals[0], bool_vals[1]);
 

src/indexer/merger.rs

@@ -578,7 +578,6 @@ impl IndexMerger {
             stats: FastFieldStats,
         }
         impl<'a> Column for FieldIndexAccessProvider<'a> {
-            #[inline]
             fn get_val(&self, doc: u64) -> u64 {
                 self.offsets[doc as usize]
             }
@@ -669,13 +668,15 @@ impl IndexMerger {
         {
             let mut serialize_vals =
                 fast_field_serializer.new_u64_fast_field_with_idx(field, 0u64, max_term_ord, 1)?;
+            let mut vals = Vec::with_capacity(100);
 
             for old_doc_addr in doc_id_mapping.iter_old_doc_addrs() {
                 let term_ordinal_mapping: &[TermOrdinal] =
                     term_ordinal_mappings.get_segment(old_doc_addr.segment_ord as usize);
 
                 let ff_reader = &fast_field_reader[old_doc_addr.segment_ord as usize];
-                for prev_term_ord in ff_reader.get_vals(old_doc_addr.doc_id) {
+                ff_reader.get_vals(old_doc_addr.doc_id, &mut vals);
+                for &prev_term_ord in &vals {
                     let new_term_ord = term_ordinal_mapping[prev_term_ord as usize];
                     serialize_vals.add_val(new_term_ord)?;
                 }
@@ -728,6 +729,8 @@ impl IndexMerger {
         let mut max_value = u64::MIN;
         let mut num_vals = 0;
 
+        let mut vals = Vec::with_capacity(100);
+
         let mut ff_readers = Vec::new();
 
         // Our values are bitpacked and we need to know what should be
@@ -745,11 +748,12 @@ impl IndexMerger {
                      Please report.",
                 );
             for doc in reader.doc_ids_alive() {
-                for val in ff_reader.get_vals(doc) {
+                ff_reader.get_vals(doc, &mut vals);
+                for &val in &vals {
                     min_value = cmp::min(val, min_value);
                     max_value = cmp::max(val, max_value);
-                    num_vals += 1;
                 }
+                num_vals += vals.len();
             }
             ff_readers.push(ff_reader);
             // TODO optimize when no deletes
@@ -792,10 +796,11 @@ impl IndexMerger {
                 let num_vals = self.fast_field_readers[old_doc_addr.segment_ord as usize]
                     .get_len(old_doc_addr.doc_id);
                 assert!(num_vals >= pos_in_values);
+                let mut vals = Vec::new();
                 self.fast_field_readers[old_doc_addr.segment_ord as usize]
-                    .get_vals(old_doc_addr.doc_id)
-                    .nth(pos_in_values as usize)
-                    .expect("computation error in SortedDocIdMultiValueAccessProvider")
+                    .get_vals(old_doc_addr.doc_id, &mut vals);
+
+                vals[pos_in_values as usize]
             }
 
             fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
@@ -805,7 +810,9 @@ impl IndexMerger {
                         .flat_map(|old_doc_addr| {
                             let ff_reader =
                                 &self.fast_field_readers[old_doc_addr.segment_ord as usize];
-                            ff_reader.get_vals(old_doc_addr.doc_id)
+                            let mut vals = Vec::new();
+                            ff_reader.get_vals(old_doc_addr.doc_id, &mut vals);
+                            vals.into_iter()
                         }),
                 )
             }
@@ -1968,32 +1975,49 @@ mod tests {
         }
         let reader = index.reader()?;
         let searcher = reader.searcher();
+        let mut vals: Vec<u64> = Vec::new();
 
         {
             let segment = searcher.segment_reader(0u32);
             let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
 
-            assert_eq!(&ff_reader.get_vals(0).collect::<Vec<_>>(), &[1, 2]);
-            assert_eq!(&ff_reader.get_vals(1).collect::<Vec<_>>(), &[1, 2, 3]);
-            assert_eq!(&ff_reader.get_vals(2).collect::<Vec<_>>(), &[4, 5]);
-            assert_eq!(&ff_reader.get_vals(3).collect::<Vec<_>>(), &[1, 2]);
-            assert_eq!(&ff_reader.get_vals(4).collect::<Vec<_>>(), &[1, 5]);
-            assert_eq!(&ff_reader.get_vals(5).collect::<Vec<_>>(), &[3]);
-            assert_eq!(&ff_reader.get_vals(6).collect::<Vec<_>>(), &[17]);
+            ff_reader.get_vals(0, &mut vals);
+            assert_eq!(&vals, &[1, 2]);
+
+            ff_reader.get_vals(1, &mut vals);
+            assert_eq!(&vals, &[1, 2, 3]);
+
+            ff_reader.get_vals(2, &mut vals);
+            assert_eq!(&vals, &[4, 5]);
+
+            ff_reader.get_vals(3, &mut vals);
+            assert_eq!(&vals, &[1, 2]);
+
+            ff_reader.get_vals(4, &mut vals);
+            assert_eq!(&vals, &[1, 5]);
+
+            ff_reader.get_vals(5, &mut vals);
+            assert_eq!(&vals, &[3]);
+
+            ff_reader.get_vals(6, &mut vals);
+            assert_eq!(&vals, &[17]);
         }
 
         {
             let segment = searcher.segment_reader(1u32);
             let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
-            assert_eq!(&ff_reader.get_vals(0).collect::<Vec<_>>(), &[28, 27]);
+            ff_reader.get_vals(0, &mut vals);
+            assert_eq!(&vals, &[28, 27]);
 
-            assert_eq!(&ff_reader.get_vals(1).collect::<Vec<_>>(), &[1000]);
+            ff_reader.get_vals(1, &mut vals);
+            assert_eq!(&vals, &[1_000]);
         }
 
         {
             let segment = searcher.segment_reader(2u32);
             let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
-            assert_eq!(&ff_reader.get_vals(0).collect::<Vec<_>>(), &[20]);
+            ff_reader.get_vals(0, &mut vals);
+            assert_eq!(&vals, &[20]);
         }
 
         // Merging the segments
@@ -2010,16 +2034,35 @@ mod tests {
             let segment = searcher.segment_reader(0u32);
             let ff_reader = segment.fast_fields().u64s(int_field).unwrap();
 
-            assert_eq!(&ff_reader.get_vals(0).collect::<Vec<_>>(), &[1, 2]);
-            assert_eq!(&ff_reader.get_vals(1).collect::<Vec<_>>(), &[1, 2, 3]);
-            assert_eq!(&ff_reader.get_vals(2).collect::<Vec<_>>(), &[4, 5]);
-            assert_eq!(&ff_reader.get_vals(3).collect::<Vec<_>>(), &[1, 2]);
-            assert_eq!(&ff_reader.get_vals(4).collect::<Vec<_>>(), &[1, 5]);
-            assert_eq!(&ff_reader.get_vals(5).collect::<Vec<_>>(), &[3]);
-            assert_eq!(&ff_reader.get_vals(6).collect::<Vec<_>>(), &[17]);
-            assert_eq!(&ff_reader.get_vals(7).collect::<Vec<_>>(), &[28, 27]);
-            assert_eq!(&ff_reader.get_vals(8).collect::<Vec<_>>(), &[1_000]);
-            assert_eq!(&ff_reader.get_vals(9).collect::<Vec<_>>(), &[20]);
+            ff_reader.get_vals(0, &mut vals);
+            assert_eq!(&vals, &[1, 2]);
+
+            ff_reader.get_vals(1, &mut vals);
+            assert_eq!(&vals, &[1, 2, 3]);
+
+            ff_reader.get_vals(2, &mut vals);
+            assert_eq!(&vals, &[4, 5]);
+
+            ff_reader.get_vals(3, &mut vals);
+            assert_eq!(&vals, &[1, 2]);
+
+            ff_reader.get_vals(4, &mut vals);
+            assert_eq!(&vals, &[1, 5]);
+
+            ff_reader.get_vals(5, &mut vals);
+            assert_eq!(&vals, &[3]);
+
+            ff_reader.get_vals(6, &mut vals);
+            assert_eq!(&vals, &[17]);
+
+            ff_reader.get_vals(7, &mut vals);
+            assert_eq!(&vals, &[28, 27]);
+
+            ff_reader.get_vals(8, &mut vals);
+            assert_eq!(&vals, &[1_000]);
+
+            ff_reader.get_vals(9, &mut vals);
+            assert_eq!(&vals, &[20]);
         }
         Ok(())
     }

src/indexer/merger_sorted_index_test.rs

@@ -383,7 +383,9 @@ mod tests {
         assert_eq!(fast_field.get_val(5), 1_000u64);
 
         let get_vals = |fast_field: &MultiValuedFastFieldReader<u64>, doc_id: u32| -> Vec<u64> {
-            fast_field.get_vals(doc_id).collect()
+            let mut vals = vec![];
+            fast_field.get_vals(doc_id, &mut vals);
+            vals
         };
         let fast_fields = segment_reader.fast_fields();
         let fast_field = fast_fields.u64s(multi_numbers).unwrap();
@@ -478,11 +480,11 @@ mod tests {
 #[cfg(all(test, feature = "unstable"))]
 mod bench_sorted_index_merge {
 
-    use fastfield_codecs::Column;
     use test::{self, Bencher};
 
     use crate::core::Index;
-    use crate::fastfield::DynamicFastFieldReader;
+    // use cratedoc_id, readerdoc_id_mappinglet vals = reader.fate::schema;
+    use crate::fastfield::{DynamicFastFieldReader, FastFieldReader};
     use crate::indexer::merger::IndexMerger;
     use crate::schema::{Cardinality, NumericOptions, Schema};
     use crate::{IndexSettings, IndexSortByField, IndexWriter, Order};
@@ -544,7 +546,7 @@ mod bench_sorted_index_merge {
             // add values in order of the new doc_ids
             let mut val = 0;
             for (doc_id, _reader, field_reader) in sorted_doc_ids {
-                val = field_reader.get_val(doc_id as u64);
+                val = field_reader.get(doc_id);
             }
 
             val

src/snippet/mod.rs

@@ -639,7 +639,10 @@ Survey in 2016, 2017, and 2018."#;
     #[test]
     fn test_collapse_overlapped_ranges() {
         assert_eq!(&collapse_overlapped_ranges(&[0..1, 2..3,]), &[0..1, 2..3]);
-        assert_eq!(collapse_overlapped_ranges(&[0..1, 1..2,]), &[0..1, 1..2]);
+        assert_eq!(
+            collapse_overlapped_ranges(&vec![0..1, 1..2,]),
+            vec![0..1, 1..2]
+        );
         assert_eq!(collapse_overlapped_ranges(&[0..2, 1..2,]), vec![0..2]);
         assert_eq!(collapse_overlapped_ranges(&[0..2, 1..3,]), vec![0..3]);
         assert_eq!(collapse_overlapped_ranges(&[0..3, 1..2,]), vec![0..3]);