feat(sindi): integrate reasoning mechanism for search diagnostics

[LHT129]

Summary

Integrate the reasoning mechanism into the SINDI (Sparse Inverted Non-redundant Distance Index) to enable diagnostic analysis of sparse vector search recall.

Changes

• src/algorithm/sindi/sindi.h: Added forward declaration for ReasoningContext, SearchWithRequest override, AttachReasoningReport method, modified search_impl and immutable_search_impl signatures
• src/algorithm/sindi/sindi.cpp: Implemented SearchWithRequest, added reasoning tracking in window loops (bucket selections, candidate visits, filter rejections), implemented AttachReasoningReport

Design

SINDI reasoning tracks:

• Window access: which windows are visited during search
• Candidate visits: which expected targets are encountered with their distances
• Filter rejections: which expected targets are rejected by the filter

Zero overhead when reasoning_ctx is null.

Related

• Closes [Feature] SINDI Reasoning Integration #1831
• Parent: [Feature] Add Reasoning Mechanism for vsag Indexes #1829
• Depends on: [Feature] ReasoningContext Implementation #1836

[mergify]

Merge Protections

🟢 All 3 merge protections satisfied — ready to merge.

Show 3 satisfied protections

🟢 Require kind label

• label~=^kind/

🟢 Require version label

• label~=^version/

🟢 Require linked issue for feature/bug PRs

• body~=(?im)(?:^|[\s\-\*])(?:close[sd]?|fix(?:e[sd])?|resolve[sd]?)\s*:?\s+(?:#\d+|[\w.\-]+/[\w.\-]+#\d+|https?://github\.com/[\w.\-]+/[\w.\-]+/issues/\d+)

[gemini-code-assist]

Code Review

This pull request introduces search reasoning capabilities to the SINDI index by adding SearchWithRequest and integrating a ReasoningContext to record bucket selections, document visits, and filter rejections. The review feedback highlights several critical issues: potential null pointer dereferences for request.query_ and dataset_results, inefficient heap allocations and overwriting behavior in bucket selection recording, incorrect distance scaling when recording visits, and missing diagnostic records during the reordering stage.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[LHT129]

All review feedback from @gemini-code-assist addressed in dd4c65d:

1. Bucket selection overwriting + heap alloc in loop (search_impl & immutable_search_impl): Moved selected_buckets declaration before the window loop, accumulate with push_back, call RecordBucketSelection once after the loop. Eliminates per-iteration heap allocation and correctly preserves all bucket IDs.

2. Incorrect distance scale: RecordVisit now uses 1.0F + dists[i] to match the actual 1 - ip distance metric.

3. Null check for request.query_: Added CHECK_ARGUMENT(request.query_ != nullptr) before dereferencing.

4. Missing reorder diagnostics: Added RecordReorder and RecordReorderEviction (with label-to-inner-id conversion) in both search_impl and immutable_search_impl reorder stages.

5. Null check for dataset_results: AttachReasoningReport now guards against null dataset_results.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

[vsag-bot]

/label S-waiting-on-review
/waiting-on reviewer
/request-review @jiaweizone
/request-review @wxyucs
/request-review @inabao

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 7 comments.