Inline flags such as `(?i)` in one `-e` pattern leak into the other patterns, and indexed and scanned results disagree
メンテナーはふだん 1 日以内に返信
関連するプルリクエストがすでにマージされています。
- #187 @shengyfu による — マージ済み
評価
調査の方向性
Start with combine_patterns in tgrep-cli/src/matching.rs and compare how it combines multiple patterns with the per-pattern parsing in build_multi_pattern_plan in tgrep-core/src/query.rs. Add a regression test for the two reported inline-flag queries, comparing scan and indexed results. Done means inline flags affect only their own pattern and indexed, scanned, and serve results agree.
索引モデルが issue の本文から書いたものです。
説明
Summary
When several patterns are given with -e (or -f), an inline flag group such as (?i) or (?-i) at the start of one pattern also applies to the patterns after it. ripgrep scopes inline flags to the pattern they appear in.
The index planner parses each pattern on its own, so it and the matcher disagree: the same query can give different results with and without the index.
Reproduce
mkdir repro && cd repro
printf 'bar\n' > bar.txt
tgrep index .
tgrep -i -e '(?-i)foo' -e 'BAR' .
tgrep --no-index -i -e '(?-i)foo' -e 'BAR' .
tgrep -e '(?i)foo' -e 'BAR' .
tgrep --no-index -e '(?i)foo' -e 'BAR' .
| Query | Actual: tgrep (index, serve) |
Actual: tgrep --no-index |
Expected: rg 15.1.0 |
|---|---|---|---|
-i -e '(?-i)foo' -e 'BAR' |
no match (exit 1) | no match (exit 1) | bar.txt:bar |
-e '(?i)foo' -e 'BAR' |
no match (exit 1) | bar.txt:bar |
no match (exit 1) |
- In the first query,
(?-i)fromfooturns-ioff forBARtoo, sobaris missed everywhere. - In the second,
(?i)fromfoomakesBARmatchbarduring a scan. The index plannedBARcase-sensitively and drops the file, so the indexed result differs from the scan.
With -w or -x the queries behave like ripgrep, because each pattern is then wrapped in its own group.
Environment
- tgrep 1.1.0, built from 1120aca (current
main) with rustc 1.95.0 (59807616e 2026-04-14) - Windows 11 Pro (build 28000); the code involved is platform-independent
- The indexed result is the same with a local index and through
tgrep serve
Cause
combine_patterns only wraps a pattern in a group for -w/-x. Otherwise it joins the raw patterns as alternatives of one group:
-e '(?i)foo' -e 'BAR' becomes (?:(?i)foo|BAR). In regex syntax a bare flag group applies until the end of the enclosing group, so it reaches BAR too. build_multi_pattern_plan plans each pattern separately, where the flag does not leak.
Possible fix
Wrap every pattern in its own non-capturing group before joining, for example (?:(?i)foo)|(?:BAR), whether or not -w/-x is set. Flags then stay inside their pattern, which is also how the planner already reads them. The fancy-regex path is built from the same combined string, so it would be covered too.
A regression test could compare both queries above across --no-index, a local index and serve.
- 主要言語
- Rust
- スター
- 3.5k
- フォーク
- 139
- 平均マージ
- 16時間 31分
- マージ済み PR（30日）
- 27
環境構築
- Dockerfile・Docker Compose ファイルなし
- プルリクエストのテンプレートなし
- コントリビューションガイドを読む
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
microsoft/tgrep のほかの issue
microsoft/tgrep の issue をすべて見る
似ている issue
-
難易度 2/5 1〜3時間 初心者へのやさしさ 79/100
メンテナーはふだん 1 日以内に返信
-
type: enhancement
難易度 2/5 1〜3時間 初心者へのやさしさ 62/100
franciscoBSalgueiro/en-croissant#1061 ·
メンテナーはふだん 3 日以内に返信
-
chore good first issue macros spire
難易度 2/5 1〜3時間 初心者へのやさしさ 78/100
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 78/100
メンテナーはふだん 1 日以内に返信
-
groupmod -U: members change in /etc/group but not in /etc/gshadow対応中かも @jayhemnani9910 が今日担当しました。 オープン
難易度 2/5 1〜3時間 初心者へのやさしさ 79/100
メンテナーはふだん 1 日以内に返信