fix(cache): preserve locked probes under directory aliases

[mldangelo-oai]

Summary

• Normalize cache-directory aliases before retaining a locked Windows identity probe during cache clearing.
• Keep stale results removable after a probe-close failure, then remove the retained probe when a later close succeeds.
• Add a deterministic failing-before regression and document the user-visible cache fix.

Validation

• Complete cache suite: 222 passed, 6 skipped.
• Release-workflow regressions: 132 passed.
• Failing-before aliased-probe regression plus existing close-failure/retry regressions: passed.
• Repository-wide Ruff lint and format checks: passed.
• Changed-file mypy and strict modified-test mypy: passed.
• Changelog Prettier and git diff whitespace checks: passed.

Full repository mypy on macOS still reports pre-existing platform/baseline findings; the exact-main GitHub Type Check is green.

[mldangelo-oai]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

Reviewed commit: cbf3b27f09

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[github-actions]

Workflow run and artifacts

Performance Benchmarks

Compared 13 shared benchmarks with a regression threshold of 15%.
Status: 0 regressions, 0 improved, 13 stable, 0 new, 0 missing.
Aggregate shared-benchmark median: 4.283s -> 4.359s (+1.8%).

Workload	Benchmark	Target	Size	Files	Baseline	Current	Change	Status
padded-multi-stream-upload	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_padded_multi_stream_upload	multi_stream_padded	4.1 KiB	1	363.8us	344.9us	-5.2%	stable
nested-payload-review	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_nested_payload_review[nested_base64]	nested_base64	98 B	1	302.8us	288.7us	-4.7%	stable
rejected-basic-auth-candidates	tests/benchmarks/test_scan_benchmarks.py::test_rejected_basic_auth_candidates_scan_linearly	-	371.1 KiB	1	2.433s	2.519s	+3.6%	stable
mixed-model-repository	tests/benchmarks/test_scan_benchmarks.py::test_scan_release_candidate_repository	release-candidate	547.3 KiB	32	656.67ms	633.45ms	-3.5%	stable
clean-training-checkpoint	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_clean_training_checkpoint	safe_large	278.2 KiB	1	109.62ms	113.15ms	+3.2%	stable
direct-malicious-upload	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_direct_malicious_upload	malicious_reduce	52 B	1	231.3us	226.3us	-2.2%	stable
nested-payload-review	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_nested_payload_review[nested_raw]	nested_raw	78 B	1	283.2us	277.1us	-2.1%	stable
nested-payload-review	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_nested_payload_review[nested_hex]	nested_hex	130 B	1	316.2us	310.7us	-1.7%	stable
suspicious-pickle-intake	tests/benchmarks/test_scan_benchmarks.py::test_scan_suspicious_pickle_intake	suspicious-intake	183.8 KiB	4	147.86ms	150.20ms	+1.6%	stable
chunked-upload-stream	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_chunked_upload_stream	chunked_stream	278.2 KiB	1	112.69ms	113.71ms	+0.9%	stable
duplicate-heavy-registry	tests/benchmarks/test_scan_benchmarks.py::test_scan_duplicate_registry_snapshot	registry-snapshot	915.2 KiB	13	567.64ms	572.49ms	+0.9%	stable
single-checkpoint-preflight	tests/benchmarks/test_scan_benchmarks.py::test_scan_single_checkpoint_before_load	single_checkpoint.pkl	183.0 KiB	1	104.01ms	104.44ms	+0.4%	stable
warm-cache-rescan	tests/benchmarks/test_scan_benchmarks.py::test_scan_warm_cached_repository_rescan	release-candidate	547.3 KiB	32	150.16ms	150.65ms	+0.3%	stable

[Copilot]

Pull request overview

This PR fixes a Windows-specific cache-clearing edge case where an open (locked) cache identity probe could be reached via a directory alias (e.g., symlink/junction), causing clear_cache() to mistakenly treat the probe as unrelated and attempt to delete it while still locked. The change normalizes alias paths so stale cache entries can still be cleared safely, and the probe is removed once a later close succeeds.

Changes:

• Normalize/cache-resolve probe and cache directory paths during clear_cache() to correctly retain locked probes even when referenced through directory aliases.
• Add a regression test that simulates a Windows “open file cannot be unlinked” behavior when the probe name points through an aliased cache directory, validating close-failure then close-success cleanup.
• Document the user-visible fix in the changelog.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
modelaudit/cache/scan_results_cache.py	Resolves cache/probe directory aliases before deciding whether to retain a still-locked probe during cache clearing.
tests/cache/test_cache_correctness.py	Adds a deterministic regression test covering aliased cache-directory probes across close-failure and retry success.
CHANGELOG.md	Notes the Windows alias + locked probe preservation behavior as a bug fix.

---

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