fix(watcher): skip hidden dirs and exclude_dirs in fs events

[Shengyu Fu (shengyfu)]

The walker uses hidden(true) by default and respects --exclude-dir, but handle_fs_event only filtered the .tgrep index dir. As a result, writes to .git/index.lock, .git/HEAD, etc. produced spurious [trace] reindex: modified .git/... lines (and pointless file I/O + trigram extraction) during normal git operations.

Mirror the walker's filtering in the watcher:

• skip any path with a dot-prefixed component
• skip any path whose components match the configured exclude_dirs

Tested locally: cargo fmt, cargo clippy --workspace --all-targets -- -D warnings, and cargo test --workspace all pass.

[Copilot]

Pull request overview

Aligns the server file watcher’s event handling with the index builder’s directory filtering to avoid reindexing noise and wasted work from paths that should not be indexed (notably .git/ activity).

Changes:

• Add watcher-side filtering to skip dot-prefixed path components (hidden paths) before reindexing.
• Add watcher-side filtering to skip paths whose components match the configured excluded directory names.

Show a summary per file

File	Description
tgrep-cli/src/serve.rs	Adds hidden/exclude directory filtering to handle_fs_event to reduce spurious watcher reindexing work.

Copilot's findings

Tip

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

Comments suppressed due to low confidence (2)

tgrep-cli/src/serve.rs:769

• The exclude_dirs check currently matches any path segment, including the final filename segment. That means a file named the same as an excluded directory (e.g., src/vendor) would be skipped by the watcher even though the walker only skips when the excluded name is a directory entry (tgrep-core/src/walker.rs). Consider applying the match only to directory components (e.g., ancestors / parent segments) so watcher and walker semantics stay consistent.

        if !state.exclude_dirs.is_empty()
            && rel_path
                .split('/')
                .any(|seg| state.exclude_dirs.iter().any(|d| d == seg))
        {
            continue;

tgrep-cli/src/serve.rs:757

• The walker also applies gitignore rules by default (WalkBuilder::git_ignore/git_global/git_exclude), but the watcher-side filter here doesn’t. That means gitignored paths can still be indexed if they change after startup, diverging from the initial walk. If parity with the walker is the goal, consider applying the same ignore rules (or at least clarify in the comment that only hidden + --exclude are mirrored).

        // Mirror the walker's filtering so the watcher does not reindex
        // files the initial walk would have skipped — most notably
        // hidden directories like `.git/`, which fire frequent
        // `index.lock`/HEAD/refs writes during normal git operations.
        // The walker uses `hidden(true)` by default; replicate that by
        // skipping any path component starting with `.`. Also honor the
        // user's --exclude-dir list so the same names are skipped here.

• Files reviewed: 1/1 changed files
• Comments generated: 3

[Copilot]

Copilot's findings

Tip

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

• Files reviewed: 1/1 changed files
• Comments generated: 2

[Copilot]

Copilot's findings

Tip

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

Comments suppressed due to low confidence (3)

tgrep-cli/src/serve.rs:1707

• Tests cover directory exclusion (e.g. target/debug/foo) but not the edge case where the excluded name appears as the file name (e.g. target as a file in the repo root). Adding a regression test for that case will help ensure the watcher’s filtering stays aligned with the walker’s directory-only exclusion behavior.

    #[test]
    fn skip_watcher_path_honors_exclude_dirs() {
        let exclude = vec!["target".to_string(), "node_modules".to_string()];
        assert!(should_skip_watcher_path("target/debug/foo", &exclude));
        assert!(should_skip_watcher_path(
            "node_modules/react/index.js",
            &exclude
        ));
        // Excluded name appearing as an inner component still excludes.
        assert!(should_skip_watcher_path("a/target/b", &exclude));
        // Substring match should NOT trigger — "targets" != "target".
        assert!(!should_skip_watcher_path("targets/foo", &exclude));
        // Unrelated paths are not skipped.
        assert!(!should_skip_watcher_path("src/main.rs", &exclude));

tgrep-cli/src/serve.rs:715

• PR description mentions --exclude-dir, but the CLI flag in this repo (and this comment) is --exclude. To avoid user confusion, consider updating the PR description (or any related docs/release notes) to use the correct flag name.

///   * any path component starting with `.` (matches `WalkBuilder::hidden(true)`)
///   * any path component matching one of the configured `--exclude` names

tgrep-cli/src/serve.rs:729

• should_skip_watcher_path treats any path segment equal to an exclude_dirs entry as excluded, including the final segment (the file name). The walker only applies exclude_dirs to directory entries, so a file named target (or similar) at the repo root would be indexed by the initial walk but then ignored by the watcher, leaving stale index content. Consider applying exclude_dirs checks only to directory components (e.g., all segments except the last for file events), or otherwise matching the walker's is_dir semantics.

        if seg.starts_with('.') {
            return true;
        }
        if exclude_dirs.iter().any(|d| d == seg) {
            return true;
        }

• Files reviewed: 2/3 changed files
• Comments generated: 1

[Copilot]

Copilot's findings

Tip

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

Comments suppressed due to low confidence (1)

tgrep-cli/src/serve.rs:1709

• Test coverage for should_skip_watcher_path doesn't include the edge case where the file name (final segment) equals an excluded directory name (e.g. "src/target" where target is a file). Adding a regression test for that case would help ensure the watcher matches the walker's semantics (exclude directory names only).

    #[test]
    fn skip_watcher_path_honors_exclude_dirs() {
        let exclude = vec!["target".to_string(), "node_modules".to_string()];
        assert!(should_skip_watcher_path("target/debug/foo", &exclude));
        assert!(should_skip_watcher_path(
            "node_modules/react/index.js",
            &exclude
        ));
        // Excluded name appearing as an inner component still excludes.
        assert!(should_skip_watcher_path("a/target/b", &exclude));
        // Substring match should NOT trigger — "targets" != "target".
        assert!(!should_skip_watcher_path("targets/foo", &exclude));
        // Unrelated paths are not skipped.
        assert!(!should_skip_watcher_path("src/main.rs", &exclude));
    }

• Files reviewed: 2/3 changed files
• Comments generated: 2

[Copilot]

Copilot's findings

Tip

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

• Files reviewed: 3/4 changed files
• Comments generated: 3

[Copilot]

Copilot's findings

Tip

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

Comments suppressed due to low confidence (1)

tgrep-core/src/walker.rs:240

• build_gitignore_matcher uses .hidden(false) and only filters out .git, so it will still recurse into other dot-directories (including .tgrep/ if present) when searching for .gitignore files. To better match the indexer’s hidden(true) behavior and avoid extra filesystem churn, consider a filter_entry that skips dot-prefixed directories (except allowing .gitignore files to be visited).

        .hidden(false)
        .git_ignore(true)
        .git_global(true)
        .git_exclude(true)
        .filter_entry(|entry| entry.file_name() != ".git")

• Files reviewed: 3/4 changed files
• Comments generated: 1

[Copilot]

Copilot's findings

Tip

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

• Files reviewed: 4/5 changed files
• Comments generated: 2

[Copilot]

build_matcher walks the entire tree to find .gitignore files but doesn't apply exclude_dirs-style subtree filtering. For repos that rely on --exclude to skip large directories that are not gitignored (or before gitignore rules are present), this can add a lot of unnecessary I/O at server startup. Consider adding an exclude_dirs parameter (or a more general filter) and skipping matching directory entries in the filter_entry closure, similar to walker.rs.

[Copilot]

PR description mentions --exclude-dir, but the CLI flag is --exclude (repeatable) for both index and serve (see tgrep-cli/src/main.rs). Updating the PR description to match the actual flag name would avoid confusion for reviewers/users.