test(file_index): make index/query/snapshot/watcher tests portable to Windows

[dose-dot-dev]

Third slice of fixing the 38 Windows test failures that became visible once #499 made cargo test loadable on Windows (context in #498). Test-only — no product change.

Two distinct Windows-only failures

1. Hardcoded unix paths in fixtures (9 tests: 6 index, 2 query, 1 snapshot).
The index materializes paths with MAIN_SEPARATOR and compares that against requested paths, so on Windows a /tmp/rootA root joined with \ (/tmp/rootA\docs) never equals the unix-spelled lookup key — lookups miss and materialize/.path assertions differ by separator. Route fixture paths and their expected strings through a local np() helper (/ → MAIN_SEPARATOR; no-op on Unix). Slash-token queries are unaffected: tokenize() splits the query on / and matches component names, not the stored separator.

2. The two real-filesystem watcher e2e tests watched the OS temp dir (2 tests).
On Windows that is C:\Users\<u>\AppData\Local\Temp, and AppData/Local is one of DEFAULT_IGNORE_PATTERNS — so build_exclusion_set(&[]) matched the tests' own watched root via **/AppData/Local/** and the coalescer correctly dropped every event (got []). It only passed on Linux because /tmp matches no default pattern. The exclusion logic was working perfectly — the test just put its root inside an excluded directory.

Fix: give those tests an exclusion set scoped to just the pattern under test (node_modules, or none for the rename test), so the temp root isn't collaterally excluded. The default patterns are already covered by the pure-coalescer unit tests.

While there, I also replaced the fixed post-arm sleep with an active arming probe + bounded poll: write a throwaway file until an event lands (proving the backend delivers), then run the real writes and poll up to a budget for the expected events. It's robust to notify's async ReadDirectoryChangesW arming and self-diagnosing — it asserts clearly if a backend ever genuinely fails to deliver, instead of a silent empty list. (That assertion is what pinpointed the AppData/Local collision.)

Why no product change

The index's MAIN_SEPARATOR handling is correct — real OS paths use the native separator on each platform. And excluding AppData/Local is intended; production watches user-configured roots (Documents, projects, …), never the OS temp dir. Both bugs are in the test fixtures.

Verification

• Windows 11: cargo test file_index fully green (126 tests), including the 11 previously-failing and the two real-FS watcher e2e tests (run repeatedly, stable).
• Linux (WSL): full cargo test green (known fs_watcher inotify env failure aside); cargo clippy --all-targets -- -D warnings and cargo fmt clean.

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request improves cross-platform test compatibility (particularly for Windows) by introducing a helper function np to convert Unix-style fixture paths to native platform separators across several test files. It also refactors the file watcher tests to prevent flaky races and avoid default exclusions matching Windows temp directories. The review feedback correctly highlights that using std::path::MAIN_SEPARATOR_STR requires Rust 1.78.0 or later, which may break compilation on older toolchains. It is recommended to use a more compatible character-mapping approach with std::path::MAIN_SEPARATOR instead.

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.