fix(indexer): correct inverted gitignore polarity in ShouldIndexFile

[oc-engteam]

What & why

ShouldIndexFile (the incremental single-file update path, reached from
POST /api/v1/document/update → handleUpdateDocument) inverted the gitignore filter for
workspaces configured with Exclude.UseGitIgnore=true.

The three path filters have different polarity:

constructor	Match(p)==true means	contract
GitIgnoreFilter (= !ignore.IsIgnored)	not ignored → keep	keep-filter
utils.NewSimpleFilter(patterns)	matches a pattern → exclude	exclude-filter
utils.NewSimpleFilterExclude(patterns)	doesn't match → keep	keep-filter

ShouldIndexFile built the gitignore branch with a keep-filter but consumed it with
NewSimpleFilter's exclude-convention (if exclude.Match(relPath,false) { return false }).
Result for UseGitIgnore=true workspaces on the incremental path:

• a not-ignored source file (Match=true) → return false → skipped (never re-indexed)
• a gitignored file (Match=false) → falls through → indexed (junk like build/, vendor/)

The full-scan path (processWorkspace) already used NewSimpleFilterExclude (keep-filter) + ListFiles
(keep where Match==true) and was correct — so only the incremental path was inverted. Default is
use_git_ignore: false, so only opt-in workspaces were affected.

Fix

Introduce one helper, buildExcludeFilter, as the single source of truth for the exclude
keep-filter, and route both call sites through it so the two can never drift to opposite polarity
again:

// buildExcludeFilter returns a keep-filter: Match==true => keep, Match==false => exclude.
func buildExcludeFilter(baseDir string, exclude types.Exclude) fsutils.ListFileFilter {
    if exclude.UseGitIgnore {
        return &GitIgnoreFilter{ignore: gitutils.NewGitIgnore(baseDir, true)}
    }
    return utils.NewSimpleFilterExclude(exclude.Customized)
}

• ShouldIndexFile now skips when the keep-filter returns false: if !exclude.Match(relPath, false) { return false }.
• processWorkspace routed through the same helper — behavior unchanged (its two arms are the
  filter it already constructed inline).

No on-disk format / API / reindex impact.

Tests

Rewrote the previously assertion-free TestShouldIndexFile_WithGitIgnoreFilter
(_ = ShouldIndexFile(...) — which is why the inversion slipped through) into an assertive test:

• main.go (not ignored) → indexed
• app.log (ignored by *.log) → skipped
• build/out.go (under ignored build/) → skipped

All three fail on the pre-fix code (verified red) and pass after. The customized-exclude branch stays
covered by the existing TestShouldIndexFile_WithCustomExcludeFilters, so both buildExcludeFilter
branches reach 100% coverage.

Verification

• Gates: gofmt/go build/go vet/go test/go test -race all green; buildExcludeFilter 100%,
  ShouldIndexFile 100% coverage.

• End-to-end with the real server binary (use_git_ignore: true, incremental /document/update):

  file	before	after
  keep2.go (not ignored)	File Ignored ❌ / unsearchable	Ok ✓ / searchable
  ignored2.go (gitignored)	Ok ❌ / searchable	File Ignored ✓ / absent

🤖 Generated with Claude Code
via Happy