fix(ignore): allow overriding the built-in **/vendor/** exclusion (#1664) by mustafaadel · Pull Request #1673 · git-ai-project/git-ai

mustafaadel · 2026-06-29T12:39:03Z

Summary

The built-in **/vendor/** pattern in DEFAULT_IGNORE_PATTERNS is unanchored, so
it matches a vendor/ path segment anywhere in the tree. A first-party
package named vendor (e.g. a Java com.<org>.vendor package living under
src/main/java/com/<org>/vendor/…) was therefore treated as vendored and
silently excluded from attribution (git-ai stats → 0% AI), with no way to
override it.

This implements the first two of @Siddhant-K-code's suggested fixes:

1. Negation support in `.git-ai-ignore`

IgnoreMatcher now evaluates patterns in order with last-match-wins
semantics (mirroring .gitignore). A !-prefixed pattern re-includes a path
that an earlier pattern excluded:

# .git-ai-ignore
!src/main/java/com/acme/vendor/**

A literal leading ! can be escaped as \!, just like in .gitignore.

2. Honor `linguist-vendored` in `.gitattributes`

Parsed the same way linguist-generated already is:

linguist-vendored / linguist-vendored=true → exclude the path
-linguist-vendored / !linguist-vendored / linguist-vendored=false →
re-include the path (emits a negation pattern)

# .gitattributes
src/main/java/com/acme/vendor/** -linguist-vendored

Both overrides flow through every consumer of effective_ignore_patterns
(stats / diff / status) and the bash-checkpoint snapshot path
(build_gitignore).

Behavior preserved

Positive-only pattern sets behave exactly as before — with no negation,
last-match-wins reduces to "ignored if any pattern matches". A genuine
third-party vendor/ directory elsewhere in the tree stays excluded.

Tests

Unit tests (src/authorship/ignore.rs): negation re-include, last-match-wins
ordering, escaped \!, a positive-only regression guard, and linguist-vendored
true/false/macro parsing.
Integration tests (tests/integration/ignore_unit.rs): end-to-end
.git-ai-ignore negation and .gitattributes linguist-vendored override
through effective_ignore_patterns, including the worktree variants.

Not included (follow-up)

@Siddhant-K-code's third suggestion — surfacing "excluded as vendored" files in
status/stats so the exclusion isn't silent — is intentionally left for a
separate PR to keep this change focused on restoring the ability to override.

🤖 Generated with Claude Code

CLAassistant · 2026-06-29T12:39:12Z

All committers have signed the CLA.

…roject#1664) The built-in `**/vendor/**` default in `DEFAULT_IGNORE_PATTERNS` is unanchored, so it matches a `vendor/` path segment anywhere in the tree — including first-party packages named `vendor` (e.g. a Java `com.<org>.vendor` package under `src/main/java/com/<org>/vendor/…`). Those files were silently excluded from attribution (0% AI) with no way to override the default. Add the two override mechanisms suggested in the issue: - Negation in `IgnoreMatcher`: patterns are now evaluated in order with last-match-wins semantics, and a `!`-prefixed pattern re-includes a path excluded by an earlier pattern. `.git-ai-ignore` lines such as `!src/main/java/com/acme/vendor/**` now take effect. A literal leading `!` can be escaped as `\!`, mirroring `.gitignore`. - `linguist-vendored` in `.gitattributes`: `linguist-vendored[=true]` adds a positive ignore pattern, while `-linguist-vendored` / `!linguist-vendored` / `linguist-vendored=false` adds a negation pattern — mirroring how `linguist-generated` is already honored. Positive-only pattern sets are unaffected: with no negation, last-match-wins reduces to "ignored if any pattern matches". Both the `IgnoreMatcher` path (stats/diff/status) and the bash checkpoint snapshot path (`build_gitignore`) honor the new sources. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

devin-ai-integration

Devin Review found 2 potential issues.

devin-ai-integration · 2026-06-29T12:45:23Z

🚩 dedupe_patterns keep-last change could reorder patterns across sources if duplicates exist

The dedupe_patterns function (src/authorship/ignore.rs:372-381) now keeps the last occurrence of each pattern string. In effective_ignore_patterns (src/authorship/ignore.rs:345-363), patterns from multiple sources are concatenated: defaults → linguist-generated → linguist-vendored → .git-ai-ignore → extra → user. If a default pattern like **/vendor/** also appears in .git-ai-ignore, the default's copy is removed and the .git-ai-ignore copy is kept — shifting the positive pattern to a later position. This matters if a negation pattern (e.g., from linguist-vendored) appeared between them: the negation would now precede the re-asserted positive pattern, making the negation ineffective. This is actually the correct last-match-wins behavior: the user explicitly re-asserted the pattern after the negation, so it should win. But teams with existing .git-ai-ignore files that redundantly list default patterns may find that negation patterns from .gitattributes stop working as expected.

Was this helpful? React with 👍 or 👎 to provide feedback.

Addressed in dda2527. dedupe_patterns now keeps the last occurrence of a duplicate instead of the first, which is the correct dedupe under last-match-wins: a pattern re-asserted after an intervening negation keeps its later position rather than collapsing onto an earlier duplicate. Added dedupe_keeps_last_occurrence and reasserted_positive_after_negation_wins unit tests to lock this in.

…wins Address review feedback on the negation/last-match-wins change: - build_gitignore (bash checkpoint path) chained `.git-ai-ignore` before the linguist sources, so under GitignoreBuilder's last-match-wins a user's `!` negation in `.git-ai-ignore` could be overridden by a linguist pattern — inconsistent with `effective_ignore_patterns`. Reorder to defaults → linguist-generated → linguist-vendored → `.git-ai-ignore` so the two paths agree and the user's negation wins. - dedupe_patterns kept the first occurrence of a duplicate pattern. Under last-match-wins the final occurrence is what decides a path, so keep the last occurrence instead; a pattern re-asserted after an intervening negation is no longer collapsed onto an earlier duplicate. Add unit tests for keep-last dedupe and re-asserted-positive-after-negation. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

devin-ai-integration

Devin Review found 1 new potential issue.

devin-ai-integration · 2026-06-30T10:31:31Z

 fn parse_linguist_generated_patterns(contents: &str) -> Vec<String> {
    let mut patterns = Vec::new();

    for raw_line in contents.lines() {
-        let line = raw_line.trim();
-        if line.is_empty() || line.starts_with('#') {
+        let Some((path_pattern, attrs)) = parse_gitattributes_line(raw_line) else {
            continue;
-        }
+        };

-        let tokens = split_gitattributes_tokens(line);
-        if tokens.len() < 2 {
-            continue;
+        if attribute_state(&attrs, "linguist-generated") == Some(true) {
+            patterns.push(path_pattern);
        }
+    }
+
+    dedupe_patterns(patterns)
+}


🚩 linguist-generated=false does not produce negation patterns unlike linguist-vendored=false

The parse_linguist_generated_patterns function (src/authorship/ignore.rs:166-180) only emits patterns for attribute_state == Some(true), ignoring Some(false). Meanwhile, parse_linguist_vendored_patterns (src/authorship/ignore.rs:190-206) emits negation patterns for Some(false). This asymmetry is intentional: vendored negation is needed to override the built-in **/vendor/** default, while there's no broad default for generated files that would need overriding. However, the default *.generated.* pattern could match files explicitly marked linguist-generated=false, and those won't get un-ignored. This is pre-existing behavior unchanged by this PR, but may be worth a follow-up.

Was this helpful? React with 👍 or 👎 to provide feedback.

Agreed, and thanks for the precise framing. This asymmetry is intentional and I'd like to keep it out of this PR:

Scope: Default ignore **/vendor/** silently excludes first-party packages named vendor from attribution (0% AI, no override) #1664 is specifically about the unanchored **/vendor/** default having no override. linguist-vendored=false emits a negation precisely to override that built-in default. There's no equivalently broad built-in for generated files, so linguist-generated keeps its existing Some(true)-only behavior.

Behavior change: making linguist-generated=false emit negations would change semantics for existing users who already set it (today it's a no-op), and would flip the existing loads_positive_linguist_generated_only test, which deliberately asserts that linguist-generated=false / -linguist-generated produce no patterns.

The *.generated.* + linguist-generated=false edge you describe is real but pre-existing and unchanged here. Now that the negation machinery exists, a focused follow-up could make linguist-generated symmetric (emit !path on false) if maintainers want that — happy to open one. Keeping this PR scoped to the vendor fix.

mustafaadel force-pushed the fix/vendor-ignore-override branch from 1cb47bc to f7ef2db Compare June 29, 2026 12:41

devin-ai-integration Bot reviewed Jun 29, 2026

View reviewed changes

mustafa-fawry and others added 2 commits June 29, 2026 15:52

Merge branch 'main' into fix/vendor-ignore-override

06955a7

devin-ai-integration Bot reviewed Jun 30, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(ignore): allow overriding the built-in /vendor/ exclusion (#1664)#1673

fix(ignore): allow overriding the built-in /vendor/ exclusion (#1664)#1673
mustafaadel wants to merge 3 commits into
git-ai-project:mainfrom
mustafaadel:fix/vendor-ignore-override

mustafaadel commented Jun 29, 2026 •

edited by devin-ai-integration Bot

Loading

Uh oh!

CLAassistant commented Jun 29, 2026 •

edited

Loading

Uh oh!

devin-ai-integration Bot left a comment

Uh oh!

devin-ai-integration Bot Jun 29, 2026 •

edited

Loading

Uh oh!

mustafaadel Jun 29, 2026

Uh oh!

Uh oh!

devin-ai-integration Bot left a comment

Uh oh!

devin-ai-integration Bot Jun 30, 2026

Uh oh!

mustafaadel Jun 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

mustafaadel commented Jun 29, 2026 • edited by devin-ai-integration Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

1. Negation support in .git-ai-ignore

2. Honor linguist-vendored in .gitattributes

Behavior preserved

Tests

Not included (follow-up)

Uh oh!

CLAassistant commented Jun 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

devin-ai-integration Bot left a comment

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Jun 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mustafaadel Jun 29, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

devin-ai-integration Bot left a comment

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Jun 30, 2026

Choose a reason for hiding this comment

Uh oh!

mustafaadel Jun 30, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

mustafaadel commented Jun 29, 2026 •

edited by devin-ai-integration Bot

Loading

1. Negation support in `.git-ai-ignore`

2. Honor `linguist-vendored` in `.gitattributes`

CLAassistant commented Jun 29, 2026 •

edited

Loading

devin-ai-integration Bot Jun 29, 2026 •

edited

Loading