Switch SearchOutputBuffer to accepting Neighbors instead of tuples

[hildebrandmw]

With #1273, the structural Eq bound on Neighbor has been removed. This PR is a follow-up that changes over SearchOutputBuffer to accept Neighbor instead of raw tuples. Outside of being symmetric, this has the nice property of allowing full-precision reranking methods to use diskann::neighbor::ord::fast_distance as their sorting callback instead of hand-rolling the implementation every time.

This PR does make the distance member of Neighbor generic, but defaults it to f32 to keep syntactic compatibility.

Key files to review:

• diskann/src/neighbors.rs: Additional (defaulted) generic distance parameter to Neighbor.
• diskann/src/graph/search_output_buffer.rs: Updated trait definition.

[hildebrandmw]

Note: The original code was unnecessarily wrapping the intermediate value in a Result but didn't call any fallible methods.

[Copilot]

Pull request overview

This PR follows up on #1273 by completing the migration from “(id, distance) tuples” to the unified Neighbor type for search result emission. It also makes Neighbor’s distance type generic (defaulting to f32) so reranking/postprocessing code can sort with neighbor::ord::fast_distance directly rather than re-implementing distance ordering.

Changes:

• Generalize Neighbor to Neighbor<I, D = f32> and update downstream uses to accommodate distance() returning a reference.
• Update SearchOutputBuffer to accept Neighbor<I, D> in push/extend, and adjust implementations and call sites across the workspace.
• Switch multiple reranking implementations to sort with neighbor::ord::fast_distance instead of hand-rolled partial_cmp logic.

Reviewed changes

Copilot reviewed 19 out of 19 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
diskann/src/neighbor/mod.rs	Makes Neighbor distance generic and updates BackInserter/buffer impls to accept Neighbor directly.
diskann/src/graph/search_output_buffer.rs	Changes SearchOutputBuffer trait to accept Neighbor in push/extend and updates internal buffer implementations/tests.
diskann/src/graph/search/range_search.rs	Updates range search logic and DistanceFiltered buffer wrapper to work with Neighbor and ref-returning distance().
diskann/src/graph/test/cases/range_search.rs	Adjusts range-search test invariants for ref-returning Neighbor::distance().
diskann/src/graph/index.rs	Updates distance extraction to dereference Neighbor::distance().
diskann/src/graph/glue.rs	Removes tuple mapping and passes Neighbor iterators directly into output buffers.
diskann/src/flat/test/harness.rs	Adjusts test harness to collect copied distances from Neighbor::distance().
diskann-tools/src/utils/ground_truth.rs	Updates ground-truth writing to dereference Neighbor::distance().
diskann-providers/src/test_utils/search_utils.rs	Updates distance comparisons for ref-returning distance() (but one arithmetic use still needs updating).
diskann-providers/src/model/graph/provider/async_/postprocess.rs	Removes tuple conversion in postprocess output and forwards Neighbor directly.
diskann-providers/src/model/graph/provider/async_/inmem/full_precision.rs	Builds reranked results as Neighbor and sorts via neighbor::ord::fast_distance.
diskann-providers/src/index/wrapped_async.rs	Updates assertions to compare dereferenced distances.
diskann-providers/src/index/diskann_async.rs	Updates distance comparisons/mutations to dereference Neighbor::distance().
diskann-garnet/src/provider.rs	Updates reranking paths to use Neighbor and sort via fast_distance.
diskann-garnet/src/lib.rs	Updates SearchOutputBuffer implementation and tests to accept/push Neighbor.
diskann-disk/src/search/provider/disk_provider.rs	Migrates reranking to store Neighbor (including cached results) and sort via fast_distance.
diskann-bftree/src/provider.rs	Updates reranking to produce/sort Neighbor instead of (id, f32) tuples.
diskann-benchmark-core/src/search/api.rs	Updates benchmark test plumbing to emit Neighbor into output buffers.
diskann-benchmark-core/src/internal/buffer.rs	Updates internal benchmark output buffer implementation to accept Neighbor items.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (1)

diskann/src/neighbor/mod.rs:150

• Neighbor now has a generic distance parameter (Neighbor<I, D>), but ord::fast_distance is still hard-coded to Neighbor<I> (i.e., D = f32). That prevents using fast_distance as a sorting callback for Neighbor<I, D> when D is not f32 (even though partial_cmp would work for any D: PartialOrd). Consider making this function generic over D (and updating the doc comment that currently mentions f32::NAN).

    pub fn fast_distance<I>(x: &Neighbor<I>, y: &Neighbor<I>) -> std::cmp::Ordering {

[codecov-commenter]

Codecov Report

❌ Patch coverage is 97.64151% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.62%. Comparing base (3e7c1f3) to head (7c938f5).

Files with missing lines	Patch %	Lines
...odel/graph/provider/async_/inmem/full_precision.rs	42.85%	4 Missing ⚠️
diskann-providers/src/index/diskann_async.rs	66.66%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1284   +/-   ##
=======================================
  Coverage   90.62%   90.62%           
=======================================
  Files         512      512           
  Lines       98807    98817   +10     
=======================================
+ Hits        89541    89553   +12     
+ Misses       9266     9264    -2     

Flag	Coverage Δ
miri	90.62% <97.64%> (+<0.01%)	⬆️
unittests	90.30% <97.64%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
diskann-benchmark-core/src/internal/buffer.rs	97.24% <100.00%> (+0.06%)	⬆️
diskann-benchmark-core/src/search/api.rs	97.85% <100.00%> (+<0.01%)	⬆️
diskann-bftree/src/provider.rs	90.42% <100.00%> (-0.01%)	⬇️
diskann-disk/src/search/provider/disk_provider.rs	93.92% <100.00%> (-0.01%)	⬇️
diskann-garnet/src/lib.rs	92.06% <100.00%> (+0.02%)	⬆️
diskann-garnet/src/provider.rs	80.27% <100.00%> (-0.08%)	⬇️
diskann-inmem/src/provider.rs	86.94% <100.00%> (+0.06%)	⬆️
diskann-providers/src/index/wrapped_async.rs	60.12% <100.00%> (ø)
...ers/src/model/graph/provider/async_/postprocess.rs	100.00% <100.00%> (ø)
diskann-providers/src/test_utils/search_utils.rs	88.05% <100.00%> (ø)
... and 10 more

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (1)

diskann/src/neighbor/mod.rs:154

• Neighbor now has a generic distance type parameter (D), but ord::fast_distance is still specialized to Neighbor<I> (i.e., D = f32). This prevents reusing the sorting callback for Neighbor<I, D> with other distance types (e.g., f64 or NotNan<f32>), which is one of the main benefits of making distance generic.

    pub fn fast_distance<I>(x: &Neighbor<I>, y: &Neighbor<I>) -> std::cmp::Ordering {
        x.distance()
            .partial_cmp(y.distance())
            .unwrap_or(std::cmp::Ordering::Equal)
    }