migrating multi-hop tests from diskann-providers to diskann

[JordanMaples]

This is a first step in the effort to migrate unit tests from diskann_async to diskann #927. If you don't think tests that have been brought over should be moved, they probably shouldn't have been. Please point them out and I'll do my best to put them back where I found them.

• Remove multihop and filter-related tests from diskann_async.rs
• Add multihop.rs test module in diskann with equivalent coverage using test_provider::Provider::grid() instead
  of the inmem provider
• Add baseline JSON files for callback filtering, even filtering, and two-hop-through-non-matching scenarios
• Make multihop_search module visible to tests (pub(crate) export in search/mod.rs)

Key differences from the original tests:

• Tests exercise the graph algorithm directly via Provider::grid() rather than going through the inmem provider
  with PQ quantization — appropriate since these test traversal behavior, not quantization
• Start point filtering: the test provider doesn't strip start points from results (unlike the inmem
  post-processor), so reject_all_returns_zero_results asserts zero non-start-point results rather than zero total
  results

[codecov-commenter]

Codecov Report

❌ Patch coverage is 95.37445% with 21 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.60%. Comparing base (3d09483) to head (0c77da1).

Files with missing lines	Patch %	Lines
diskann/src/graph/test/cases/multihop.rs	95.11%	21 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #928      +/-   ##
==========================================
+ Coverage   89.46%   90.60%   +1.13%     
==========================================
  Files         458      459       +1     
  Lines       85530    85452      -78     
==========================================
+ Hits        76522    77425     +903     
+ Misses       9008     8027     -981     

Flag	Coverage Δ
miri	90.60% <95.37%> (+1.13%)	⬆️
unittests	90.56% <95.37%> (+1.47%)	⬆️

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

Files with missing lines	Coverage Δ
diskann-providers/src/index/diskann_async.rs	96.37% <ø> (+0.37%)	⬆️
diskann/src/graph/search/multihop_search.rs	99.44% <100.00%> (+1.36%)	⬆️
diskann/src/graph/test/cases/multihop.rs	95.11% <95.11%> (ø)

... and 42 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

Note

Copilot was unable to run its full agentic suite in this review.

Migrates a first set of multi-hop search traversal unit tests from diskann_async into the diskann crate’s graph test suite, and introduces shared search ground-truth utilities to support those tests.

Changes:

• Added multihop test cases under diskann/src/graph/test/cases/ and wired them into the test module.
• Introduced diskann::graph::test::search_utils with ground-truth + assertion helpers for search verification.
• Removed the migrated multi-hop test helpers/cases from diskann-providers/src/index/diskann_async.rs and adjusted diskann-providers test utils module visibility/docs.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
diskann/src/graph/test/search_utils.rs	Adds ground-truth computation and assertion helpers for graph search tests.
diskann/src/graph/test/mod.rs	Exposes the new search_utils module in the graph test module.
diskann/src/graph/test/cases/multihop.rs	Adds migrated multi-hop traversal/filtering/termination/callback tests.
diskann/src/graph/test/cases/mod.rs	Registers the new multihop test module.
diskann-providers/src/test_utils/search_utils.rs	Updates/clarifies docs around duplicated ground-truth helpers for provider-side tests.
diskann-providers/src/test_utils/mod.rs	Makes search_utils publicly accessible from diskann-providers::test_utils.
diskann-providers/src/index/diskann_async.rs	Removes migrated multi-hop-related tests and supporting helpers.

---

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

[hildebrandmw]

Thanks Jordan! This is moving the tests in the right direction, but we need to be careful about just moving the test infrastructure from diskann_async.rs as-is.
The utilities in search_utils.rs are extremely awkward for actually running tests, are very sensitive, and don't provide much useful information when they fire.
They've been used in diskann_async.rs because it was kind of the best thing we had back then.

My hope is that the new diskann can take a higher signal approach using baselines and VerboseEq.
Not only does this provide a really good way of viewing the expected results as a whole, it's also great for storing additional metrics.
For example, the stats, ids, and distances from multi-hop search can all be checked in as part of the baseline and get protected for free.

My ask is to not migrate the search_utils.rs as is - especially if it means including test methods that aren't actually used by diskann.
Also, use the baseline capturing mechanism to capture everything about both test setups and results.
We cannot rely solely on the baseline to protect against regression (someone could check-in a broken baseline in the future), but a baseline in combination with some invariant checks (returned items should be filtered/adjusted) will go a long way toward good tests.

[JordanMaples]

converting back to draft as it needs some more human refinement

[JordanMaples]

@copilot would you mind taking another review over this pr as it has changed significantly from your previous review.

[Copilot]

Thanks Jordan! This is moving the tests in the right direction, but we need to be careful about just moving the test infrastructure from diskann_async.rs as-is.
The utilities in search_utils.rs ar...

Addressed in 9c85d26: diskann no longer carries the migrated search_utils.rs, multihop tests now use checked-in baselines with VerboseEq, and invariants remain in-test for filtered/adjusted behavior.

@copilot would you mind taking another review over this pr as it has changed significantly from your previous review.

I re-reviewed the updated PR state and the major requested shifts are present (baseline-driven multihop tests + removal of migrated diskann search utils) in 9c85d26 and follow-up cleanup in a77f01a.