grep: support GREP_COLORS 'mt' and warn on deprecated GREP_COLOR

Two color-handling gaps vs GNU: the 'mt' capability in GREP_COLORS (which
sets both the selected- and context-match colors) was ignored, and the
deprecated GREP_COLOR variable produced no warning. Handle 'mt', and emit
GNU's 'GREP_COLOR=... is deprecated; use GREP_COLORS=mt=...' warning when
color output is actually produced. Fixes the GNU testsuite 'color-colors'
test.

.github/workflows/benchmarks.yml

@@ -1,50 +0,0 @@
-name: Benchmarks
-
-# spell-checker:ignore codspeed dtolnay Swatinem sccache
-
-on:
-  push:
-    branches: [ main, master ]
-  pull_request:
-    branches: [ main, master ]
-
-permissions:
-  contents: read # to fetch code (actions/checkout)
-
-# End the current execution if there is a new changeset in the PR.
-concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
-  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-
-jobs:
-  benchmarks:
-    name: Run benchmarks (CodSpeed)
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          persist-credentials: false
-
-      - name: Install Rust
-        uses: dtolnay/rust-toolchain@stable
-      - uses: Swatinem/rust-cache@v2
-
-      - name: Run sccache-cache
-        uses: mozilla-actions/sccache-action@v0.0.10
-
-      - name: Install cargo-codspeed
-        uses: taiki-e/install-action@v2
-        with:
-          tool: cargo-codspeed
-
-      - name: Build benchmarks
-        run: cargo codspeed build -p uu_grep
-
-      - name: Run benchmarks
-        uses: CodSpeedHQ/action@v4
-        env:
-          CODSPEED_LOG: debug
-        with:
-          mode: simulation
-          run: cargo codspeed run -p uu_grep > /dev/null
-          token: ${{ secrets.CODSPEED_TOKEN }}

Cargo.toml

@@ -28,11 +28,4 @@ uucore = "0.8.0"
 walkdir = "2.5"
 
 [dev-dependencies]
-divan = { package = "codspeed-divan-compat", version = "4.0.5" }
-tempfile = "3.10.1"
-uucore = { version = "0.8.0", features = ["benchmark"] }
 uutests = "0.8.0"
-
-[[bench]]
-name = "grep_bench"
-harness = false

README.md

@@ -33,7 +33,6 @@ cargo test
 
 * Does not take `LANG`, etc., into account for handling file encodings (non-UTF8 matches are treated as binary)
 * No localization support yet
-* Performances need to be improved
 
 ## Contributing
 

benches/grep_bench.rs

@@ -1,255 +0,0 @@
-// Benchmarks for the grep utility
-//
-// SPDX-License-Identifier: MIT
-//
-// This file is part of the uutils grep package.
-// It is licensed under the MIT License.
-// For the full copyright and license information, please view the LICENSE
-// file that was distributed with this source code.
-
-use divan::{Bencher, black_box};
-use uu_grep::uumain;
-use uucore::benchmark::{create_test_file, run_util_function};
-
-/// Build an access-log-like data set with `n` lines.
-///
-/// Roughly a quarter of the lines use a non-default HTTP method / status /
-/// user-agent so that selective patterns match a realistic subset rather than
-/// every line or no line at all.
-fn access_log(n: usize) -> Vec<u8> {
-    let mut data = Vec::new();
-    for i in 0..n {
-        let method = if i % 4 == 0 { "POST" } else { "GET" };
-        let status = if i % 7 == 0 { 404 } else { 200 };
-        let agent = if i % 3 == 0 {
-            "Mozilla/5.0 (X11; Linux x86_64) Chrome/120.0"
-        } else {
-            "curl/8.5.0"
-        };
-        let line = format!(
-            "192.168.{}.{} - - [01/Jan/2024:00:00:00 +0000] \"{} /index.html HTTP/1.1\" {} 1234 \"-\" \"{}\"\n",
-            (i / 256) % 256,
-            i % 256,
-            method,
-            status,
-            agent,
-        );
-        data.extend_from_slice(line.as_bytes());
-    }
-    data
-}
-
-/// Benchmark a literal search that matches nothing.
-///
-/// This is the purest measure of raw scan throughput: the whole file is read
-/// and searched but no output is produced.
-#[divan::bench]
-fn literal_no_match(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(
-            uumain,
-            &["ZZZ_NONEXISTENT_PATTERN_ZZZ", file_path_str],
-        ));
-    });
-}
-
-/// Benchmark a literal search that matches a subset of lines.
-#[divan::bench]
-fn literal_match_some(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(uumain, &["POST", file_path_str]));
-    });
-}
-
-/// Benchmark a literal search that matches every line (counting only).
-///
-/// `-c` keeps the output bounded so the benchmark measures matching rather than
-/// terminal I/O.
-#[divan::bench]
-fn literal_match_all_count(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(uumain, &["-c", "HTTP", file_path_str]));
-    });
-}
-
-/// Benchmark a fixed-string search (`-F`).
-#[divan::bench]
-fn fixed_string(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(
-            uumain,
-            &["-F", "Chrome/120.0", file_path_str],
-        ));
-    });
-}
-
-/// Benchmark a case-insensitive search (`-i`).
-#[divan::bench]
-fn case_insensitive(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(uumain, &["-i", "mozilla", file_path_str]));
-    });
-}
-
-/// Benchmark counting matches (`-c`).
-#[divan::bench]
-fn count(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(uumain, &["-c", "POST", file_path_str]));
-    });
-}
-
-/// Benchmark an inverted match (`-v`).
-///
-/// Most lines do not contain "POST", so this selects the majority of lines;
-/// `-c` bounds the output.
-#[divan::bench]
-fn invert_match_count(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(uumain, &["-vc", "POST", file_path_str]));
-    });
-}
-
-/// Benchmark printing line numbers (`-n`).
-#[divan::bench]
-fn line_number(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(1_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(uumain, &["-nc", "POST", file_path_str]));
-    });
-}
-
-/// Benchmark word-boundary matching (`-w`).
-#[divan::bench]
-fn word_match(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(uumain, &["-wc", "GET", file_path_str]));
-    });
-}
-
-/// Benchmark an extended regular expression with alternation (`-E`).
-#[divan::bench]
-fn extended_regex(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(
-            uumain,
-            &["-Ec", "(POST|DELETE|PUT)", file_path_str],
-        ));
-    });
-}
-
-/// Benchmark a basic regular expression with an anchor and character class.
-#[divan::bench]
-fn basic_regex(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(
-            uumain,
-            &["-c", "^192\\.168\\.[0-9]*\\.0 ", file_path_str],
-        ));
-    });
-}
-
-/// Benchmark a Perl-compatible regular expression (`-P`).
-#[divan::bench]
-fn perl_regex(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(
-            uumain,
-            &["-Pc", "\"\\d{3}\" \\d+", file_path_str],
-        ));
-    });
-}
-
-/// Benchmark `--only-matching` (`-o`) extracting a substring from each line.
-#[divan::bench]
-fn only_matching(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(1_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(
-            uumain,
-            &["-Eoc", "[0-9]+\\.[0-9]+\\.[0-9]+\\.[0-9]+", file_path_str],
-        ));
-    });
-}
-
-/// Benchmark quiet mode (`-q`), which can stop at the first match.
-#[divan::bench]
-fn quiet_first_match(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let file_path = create_test_file(&access_log(2_000_000), temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(uumain, &["-q", "POST", file_path_str]));
-    });
-}
-
-/// Benchmark searching short numeric lines (many small lines).
-#[divan::bench]
-fn short_lines(bencher: Bencher) {
-    let temp_dir = tempfile::tempdir().unwrap();
-    let mut data = Vec::new();
-    for i in 0..10_000_000 {
-        data.extend_from_slice(format!("{i}\n").as_bytes());
-    }
-    let file_path = create_test_file(&data, temp_dir.path());
-    let file_path_str = file_path.to_str().unwrap();
-
-    bencher.bench(|| {
-        black_box(run_util_function(uumain, &["-c", "999", file_path_str]));
-    });
-}
-
-fn main() {
-    divan::main();
-}

src/lib.rs

@@ -342,6 +342,13 @@ pub fn uumain(args: impl uucore::Args) -> UResult<()> {
         ColorMode::Never => false,
         ColorMode::Auto => std::io::stdout().is_terminal(),
     };
+    // GREP_COLOR is deprecated in favour of GREP_COLORS' `mt` capability;
+    // GNU warns about it, but only when color output is actually produced.
+    if use_color && !grep_color.is_empty() {
+        eprintln!(
+            "grep: warning: GREP_COLOR='{grep_color}' is deprecated; use GREP_COLORS='mt={grep_color}'"
+        );
+    }
     let color_config = ColorConfig::from_env(&grep_color, &grep_colors);
 
     let config = Config {
@@ -903,6 +910,11 @@ impl<'a> ColorConfig<'a> {
         for item in grep_colors.split(':') {
             if let Some((key, value)) = item.split_once('=') {
                 match key {
+                    // `mt` sets both the selected- and context-match colors.
+                    "mt" => {
+                        config.matched_selected = value;
+                        config.matched_context = value;
+                    }
                     "ms" => config.matched_selected = value,
                     "mc" => config.matched_context = value,
                     "fn" => config.filename = value,

src/matcher.rs

@@ -5,10 +5,14 @@
 
 use crate::{Config, RegexMode};
 use onig::{
-    EncodedBytes, Regex, RegexOptions, Region, SearchOptions, Syntax, SyntaxBehavior,
-    SyntaxOperator,
+    EncodedBytes, Error, MatchParam, Regex, RegexOptions, Region, SearchOptions, Syntax,
+    SyntaxBehavior, SyntaxOperator,
 };
-use onig_sys::{OnigEncCtype_ONIGENC_CTYPE_WORD, OnigEncodingUTF8};
+use onig_sys::{
+    ONIGERR_EMPTY_RANGE_IN_CHAR_CLASS, ONIGERR_INVALID_BACKREF, ONIGERR_RETRY_LIMIT_IN_MATCH_OVER,
+    ONIGERR_RETRY_LIMIT_IN_SEARCH_OVER, OnigEncCtype_ONIGENC_CTYPE_WORD, OnigEncodingUTF8,
+};
+use std::io;
 use uucore::error::{UResult, USimpleError};
 
 pub struct Matcher<'a> {
@@ -26,26 +30,30 @@ impl<'a> Matcher<'a> {
     }
 
     /// Decide whether `line` matches and return the positions to highlight.
-    pub fn match_line(&self, line: &[u8]) -> Option<Vec<(usize, usize)>> {
+    ///
+    /// Returns an error if the regex engine bails out (e.g. it exceeds its
+    /// backtracking retry limit on a pathological pattern); the caller turns
+    /// that into a GNU-style diagnostic and exit code 2 rather than aborting.
+    pub fn match_line(&self, line: &[u8]) -> io::Result<Option<Vec<(usize, usize)>>> {
         let mut any_seen = false;
-        let positions: Vec<_> = MatchIter::new(&self.patterns, line)
-            .filter(|&(start, end)| {
-                any_seen = true;
-                // Drop zero-length matches from the output.
-                if start == end {
-                    return false;
-                }
-                // Drop matches that don't span the whole line if `-x` was requested.
-                if self.config.line_regexp && !(start == 0 && end == line.len()) {
-                    return false;
-                }
-                // Drop matches that aren't word matches if `-w` was requested.
-                if self.config.word_regexp && !Self::is_word_match(line, start, end) {
-                    return false;
-                }
-                true
-            })
-            .collect();
+        let mut positions = Vec::new();
+        let mut iter = MatchIter::new(&self.patterns, line).map_err(match_error)?;
+        while let Some((start, end)) = iter.next_match().map_err(match_error)? {
+            any_seen = true;
+            // Drop zero-length matches from the output.
+            if start == end {
+                continue;
+            }
+            // Drop matches that don't span the whole line if `-x` was requested.
+            if self.config.line_regexp && !(start == 0 && end == line.len()) {
+                continue;
+            }
+            // Drop matches that aren't word matches if `-w` was requested.
+            if self.config.word_regexp && !Self::is_word_match(line, start, end) {
+                continue;
+            }
+            positions.push((start, end));
+        }
 
         let raw_matched = if self.config.line_regexp || self.config.word_regexp {
             // -w / -x are authoritative once positions are filtered.
@@ -54,23 +62,25 @@ impl<'a> Matcher<'a> {
             any_seen
         };
 
-        if raw_matched != self.config.invert_match {
-            Some(positions)
-        } else {
-            None
-        }
+        Ok((raw_matched != self.config.invert_match).then_some(positions))
     }
 
     /// Cheap match check that doesn't enumerate positions.
-    pub fn is_match(&self, line: &[u8]) -> Option<Vec<(usize, usize)>> {
+    pub fn is_match(&self, line: &[u8]) -> io::Result<Option<Vec<(usize, usize)>>> {
         // `-w` / `-x` need positions to filter, so we fall back to `match_line`.
         let matched = if self.config.line_regexp || self.config.word_regexp {
-            self.match_line(line).is_some()
+            self.match_line(line)?.is_some()
         } else {
-            let raw_matched = self.patterns.iter().any(|p| p.is_match(line));
+            let mut raw_matched = false;
+            for p in &self.patterns {
+                if p.is_match(line).map_err(match_error)? {
+                    raw_matched = true;
+                    break;
+                }
+            }
             raw_matched != self.config.invert_match
         };
-        matched.then(Vec::new)
+        Ok(matched.then(Vec::new))
     }
 
     /// Word-boundary check `-w`.
@@ -112,35 +122,31 @@ struct MatchIter<'a> {
 }
 
 impl<'a> MatchIter<'a> {
-    fn new(patterns: &'a [CompiledPattern], line: &'a [u8]) -> Self {
-        Self {
-            cursors: patterns
-                .iter()
-                .map(|pattern| {
-                    let mut c = Cursor {
-                        pattern,
-                        line,
-                        offset: 0,
-                        pending: None,
-                    };
-                    c.refill();
-                    c
-                })
-                .collect(),
-            last_end: 0,
+    fn new(patterns: &'a [CompiledPattern], line: &'a [u8]) -> Result<Self, Error> {
+        let mut cursors = Vec::with_capacity(patterns.len());
+        for pattern in patterns {
+            let mut c = Cursor {
+                pattern,
+                line,
+                offset: 0,
+                pending: None,
+            };
+            c.refill()?;
+            cursors.push(c);
         }
+        Ok(Self {
+            cursors,
+            last_end: 0,
+        })
     }
-}
 
-impl<'a> Iterator for MatchIter<'a> {
-    type Item = (usize, usize);
-
-    fn next(&mut self) -> Option<Self::Item> {
+    /// Yield the next match across all patterns, or `None` when exhausted.
+    fn next_match(&mut self) -> Result<Option<(usize, usize)>, Error> {
         // Discard stale pendings that fall before the last emit.
         for cursor in &mut self.cursors {
             if matches!(cursor.pending, Some((s, _)) if s < self.last_end) {
                 cursor.offset = self.last_end;
-                cursor.refill();
+                cursor.refill()?;
             }
         }
 
@@ -153,12 +159,15 @@ impl<'a> Iterator for MatchIter<'a> {
             .enumerate()
             .filter_map(|(i, c)| c.pending.map(|p| (i, p)))
             .min_by_key(|&(_, (s, e))| (s, std::cmp::Reverse(e)))
-            .map(|(i, _)| i)?;
+            .map(|(i, _)| i);
+        let Some(best_idx) = best_idx else {
+            return Ok(None);
+        };
 
         let (start, end) = self.cursors[best_idx].pending.unwrap();
-        self.cursors[best_idx].refill();
+        self.cursors[best_idx].refill()?;
         self.last_end = end;
-        Some((start, end))
+        Ok(Some((start, end)))
     }
 }
 
@@ -173,24 +182,25 @@ struct Cursor<'a> {
 }
 
 impl Cursor<'_> {
-    fn refill(&mut self) {
+    fn refill(&mut self) -> Result<(), Error> {
         if self.offset >= self.line.len() {
             self.pending = None;
-            return;
+            return Ok(());
         }
-        let Some((start, leftmost_end)) = self.pattern.search_leftmost(self.line, self.offset)
+        let Some((start, leftmost_end)) = self.pattern.search_leftmost(self.line, self.offset)?
         else {
             self.pending = None;
-            return;
+            return Ok(());
         };
         let end = self
             .pattern
-            .longest_end_at(self.line, start)
+            .longest_end_at(self.line, start)?
             .unwrap_or(leftmost_end);
         // Advance the next search past the match we just found.
         // Zero-length matches need a +1 nudge to avoid spinning forever.
         self.offset = end.max(start + 1);
         self.pending = Some((start, end));
+        Ok(())
     }
 }
 
@@ -205,6 +215,12 @@ struct CompiledPattern {
 
 impl CompiledPattern {
     fn compile(pattern: &str, config: &Config) -> UResult<Self> {
+        // GNU grep rejects the confusing `[:name:]` bracket form (a misspelled
+        // `[[:name:]]`) in basic/extended modes; oniguruma accepts it silently.
+        if matches!(config.regex_mode, RegexMode::Basic | RegexMode::Extended) {
+            check_confusing_bracket(pattern)?;
+        }
+
         let mut syntax = *match config.regex_mode {
             RegexMode::Fixed => Syntax::asis(),
             RegexMode::Basic => Syntax::grep(),
@@ -232,17 +248,29 @@ impl CompiledPattern {
             options |= RegexOptions::REGEX_OPTION_IGNORECASE;
         }
 
-        fn compile_with(pattern: &str, syntax: &Syntax, options: RegexOptions) -> UResult<Regex> {
+        let mode = config.regex_mode;
+        fn compile_with(
+            pattern: &str,
+            syntax: &Syntax,
+            options: RegexOptions,
+            mode: RegexMode,
+        ) -> UResult<Regex> {
             Regex::with_options_and_encoding(pattern, options, syntax).map_err(|err| {
-                USimpleError::new(2, format!("invalid pattern \"{pattern}\": {err}"))
+                // Prefer GNU grep's wording for the errors it has a dedicated
+                // message for; fall back to oniguruma's text otherwise.
+                match gnu_error_message(err.code(), mode) {
+                    Some(msg) => USimpleError::new(2, msg.to_string()),
+                    None => USimpleError::new(2, format!("invalid pattern \"{pattern}\": {err}")),
+                }
             })
         }
 
-        let leftmost = compile_with(pattern, &syntax, options)?;
+        let leftmost = compile_with(pattern, &syntax, options, mode)?;
         let longest_anchored = compile_with(
             pattern,
             &syntax,
             options | RegexOptions::REGEX_OPTION_FIND_LONGEST,
+            mode,
         )?;
         Ok(Self {
             leftmost,
@@ -251,41 +279,216 @@ impl CompiledPattern {
     }
 
     /// Find the leftmost match starting at or after `offset`.
-    fn search_leftmost(&self, line: &[u8], offset: usize) -> Option<(usize, usize)> {
+    fn search_leftmost(&self, line: &[u8], offset: usize) -> Result<Option<(usize, usize)>, Error> {
         let mut region = Region::new();
-        self.leftmost.search_with_encoding(
+        let found = self.leftmost.search_with_param(
             EncodedBytes::from_parts(line, &raw mut OnigEncodingUTF8),
             offset,
             line.len(),
             SearchOptions::SEARCH_OPTION_NONE,
             Some(&mut region),
+            MatchParam::default(),
         )?;
-        region.pos(0)
+        Ok(found.and_then(|_| region.pos(0)))
     }
 
     /// Given a known leftmost start `start`, return the longest extent
     /// of a match anchored exactly there = POSIX leftmost-longest end.
-    fn longest_end_at(&self, line: &[u8], start: usize) -> Option<usize> {
+    fn longest_end_at(&self, line: &[u8], start: usize) -> Result<Option<usize>, Error> {
         let mut region = Region::new();
-        self.longest_anchored.match_with_encoding(
+        self.longest_anchored.match_with_param(
             EncodedBytes::from_parts(line, &raw mut OnigEncodingUTF8),
             start,
             SearchOptions::SEARCH_OPTION_NONE,
             Some(&mut region),
-        );
-        region.pos(0).map(|(_, end)| end)
+            MatchParam::default(),
+        )?;
+        Ok(region.pos(0).map(|(_, end)| end))
     }
 
     /// True if any match exists in `line` (including zero-length).
-    fn is_match(&self, line: &[u8]) -> bool {
-        self.leftmost
-            .search_with_encoding(
+    fn is_match(&self, line: &[u8]) -> Result<bool, Error> {
+        Ok(self
+            .leftmost
+            .search_with_param(
                 EncodedBytes::from_parts(line, &raw mut OnigEncodingUTF8),
                 0,
                 line.len(),
                 SearchOptions::SEARCH_OPTION_NONE,
                 None,
-            )
-            .is_some()
+                MatchParam::default(),
+            )?
+            .is_some())
     }
 }
+
+/// Convert a regex-engine match-time error into an I/O error carrying GNU
+/// grep's wording. The only error we expect in practice is the backtracking
+/// retry limit being exceeded on a pathological pattern; GNU reports this as
+/// `exceeded PCRE's backtracking limit` and exits 2 instead of aborting.
+fn match_error(err: Error) -> io::Error {
+    let message = if matches!(
+        err.code(),
+        ONIGERR_RETRY_LIMIT_IN_MATCH_OVER | ONIGERR_RETRY_LIMIT_IN_SEARCH_OVER
+    ) {
+        "exceeded PCRE's backtracking limit".to_string()
+    } else {
+        err.description().to_string()
+    };
+    io::Error::other(message)
+}
+
+/// Map an oniguruma compile-error code to GNU grep's wording for the same
+/// condition, when one exists. GNU emits a bare POSIX-style diagnostic (e.g.
+/// `Invalid range end`) rather than oniguruma's phrasing, so translating keeps
+/// us byte-compatible. Returns `None` for errors with no GNU equivalent, where
+/// the caller falls back to oniguruma's own message.
+fn gnu_error_message(code: i32, mode: RegexMode) -> Option<&'static str> {
+    match code {
+        // e.g. `[b-a]`: a range whose end precedes its start.
+        ONIGERR_EMPTY_RANGE_IN_CHAR_CLASS => Some("Invalid range end"),
+        // e.g. `(.)\2`: a back-reference to a group that does not exist. GNU
+        // (via PCRE2) and gnulib's regex word this differently.
+        ONIGERR_INVALID_BACKREF if mode == RegexMode::Perl => {
+            Some("reference to non-existent subpattern")
+        }
+        ONIGERR_INVALID_BACKREF => Some("Invalid back reference"),
+        _ => None,
+    }
+}
+
+/// Reject the confusing `[:name:]` bracket form the way GNU grep does.
+///
+/// A bracket expression like `[:space:]` is almost always a misspelled
+/// `[[:space:]]`; GNU grep flags it with a dedicated diagnostic and exits 2,
+/// whereas oniguruma silently treats it as the set `{':','s','p',…}`. This
+/// scans the pattern for that form and returns the same error.
+fn check_confusing_bracket(pattern: &str) -> UResult<()> {
+    let bytes = pattern.as_bytes();
+    let mut i = 0;
+    while i < bytes.len() {
+        match bytes[i] {
+            // Outside a bracket a backslash escapes the next character, so
+            // `\[` does not open a bracket expression.
+            b'\\' => i += 2,
+            b'[' => {
+                i += 1;
+                if bracket_warns(bytes, &mut i) {
+                    return Err(USimpleError::new(
+                        2,
+                        "character class syntax is [[:space:]], not [:space:]".to_string(),
+                    ));
+                }
+            }
+            _ => i += 1,
+        }
+    }
+    Ok(())
+}
+
+/// Consume a single bracket expression starting just past its opening `[` and
+/// report whether GNU grep's colon warning fires for it.
+///
+/// This is a faithful port of the `colon_warning_state` logic in GNU grep's
+/// `parse_bracket_exp` (gnulib `dfa.c`). The state is a bitmask:
+///   bit 0 — first character is a colon
+///   bit 1 — last character is a colon
+///   bit 2 — includes some other (non-colon) character
+///   bit 3 — includes a range, char/equivalence class, or collating element
+/// The warning fires exactly when the state ends equal to `7` (bits 0–2 set,
+/// bit 3 clear). On the way it advances `i` past the closing `]`.
+fn bracket_warns(bytes: &[u8], i: &mut usize) -> bool {
+    fn fetch(bytes: &[u8], i: &mut usize) -> Option<u8> {
+        let b = bytes.get(*i).copied();
+        if b.is_some() {
+            *i += 1;
+        }
+        b
+    }
+
+    let Some(first) = fetch(bytes, i) else {
+        return false;
+    };
+    let mut c = first;
+    if c == b'^' {
+        match fetch(bytes, i) {
+            Some(x) => c = x,
+            None => return false,
+        }
+    }
+    let mut state: u8 = u8::from(c == b':');
+
+    'scan: loop {
+        state &= !2;
+        let mut c1: Option<u8> = None;
+
+        if c == b'[' {
+            let Some(nc1) = fetch(bytes, i) else {
+                return false;
+            };
+            // `[:`, `[.` and `[=` introduce a class / collating / equivalence
+            // element; consume it whole and mark bit 3.
+            if nc1 == b':' || nc1 == b'.' || nc1 == b'=' {
+                loop {
+                    match fetch(bytes, i) {
+                        None => break,
+                        Some(cc) if cc == nc1 && bytes.get(*i).copied() == Some(b']') => break,
+                        Some(_) => {}
+                    }
+                }
+                if fetch(bytes, i).is_none() {
+                    return false; // consumes the `]`
+                }
+                state |= 8;
+                match fetch(bytes, i) {
+                    Some(b']') => break 'scan,
+                    Some(x) => {
+                        c = x;
+                        continue 'scan;
+                    }
+                    None => return false,
+                }
+            }
+            // Otherwise `[` is an ordinary character; `nc1` is the lookahead.
+            c1 = Some(nc1);
+        }
+
+        if c1.is_none() {
+            c1 = fetch(bytes, i);
+        }
+
+        if c1 == Some(b'-') {
+            let Some(mut c2) = fetch(bytes, i) else {
+                return false;
+            };
+            if c2 == b'[' && bytes.get(*i).copied() == Some(b'.') {
+                c2 = b']';
+            }
+            if c2 == b']' {
+                // `[x-]`: the hyphen is a literal; put the `]` back so the
+                // loop terminator sees it next.
+                *i -= 1;
+            } else {
+                state |= 8;
+                match fetch(bytes, i) {
+                    Some(b']') => break 'scan,
+                    Some(x) => {
+                        c = x;
+                        continue 'scan;
+                    }
+                    None => return false,
+                }
+            }
+        }
+
+        state |= if c == b':' { 2 } else { 4 };
+
+        match c1 {
+            Some(b']') => break 'scan,
+            Some(x) => c = x,
+            None => return false,
+        }
+    }
+
+    state == 7
+}

src/searcher.rs

@@ -281,7 +281,7 @@ impl<'a> Searcher<'a> {
                 return Ok(false);
             }
 
-            if let Some(positions) = self.session_match_line(line) {
+            if let Some(positions) = self.session_match_line(line)? {
                 // TODO: GNU grep respects LANG. Here, I'm always checking for valid UTF-8.
                 if !self.session_mark_binary_if(|| std::str::from_utf8(line).is_err()) {
                     return Ok(false);
@@ -321,9 +321,9 @@ impl<'a> Searcher<'a> {
         self.config.binary_mode != BinaryMode::WithoutMatch
     }
 
-    fn session_match_line(&self, line: &[u8]) -> Option<Vec<(usize, usize)>> {
+    fn session_match_line(&self, line: &[u8]) -> io::Result<Option<Vec<(usize, usize)>>> {
         if !self.session_can_match() {
-            None
+            Ok(None)
         } else if self.session_needs_match_positions() {
             self.matcher.match_line(line)
         } else {

tests/test_grep.rs

@@ -126,6 +126,142 @@ fn ere_invalid_pattern_is_error() {
         .stderr_contains("invalid pattern");
 }
 
+#[test]
+fn confusing_bracket_class_is_error() {
+    // GNU grep rejects the misspelled `[:name:]` form (meant to be
+    // `[[:name:]]`) with a dedicated diagnostic and exit code 2.
+    // No piped input: the pattern is rejected at compile time, before stdin is
+    // read, so feeding stdin would race with the child exiting (broken pipe).
+    for pattern in ["[:space:]", "[:digit:]", "[^:space:]", "x[:space:]y"] {
+        let (_s, mut c) = ucmd();
+        c.args(&[pattern])
+            .fails_with_code(2)
+            .stderr_is("grep: character class syntax is [[:space:]], not [:space:]\n");
+    }
+
+    // The same diagnostic applies in extended mode.
+    let (_s, mut c) = ucmd();
+    c.args(&["-E", "[:space:]"])
+        .fails_with_code(2)
+        .stderr_is("grep: character class syntax is [[:space:]], not [:space:]\n");
+}
+
+#[test]
+fn lookalike_brackets_are_not_confusing() {
+    // Patterns that are NOT the confusing `[:name:]` form must compile
+    // normally (no diagnostic). A proper class, a colon set, a range, a
+    // trailing colon set, and `-F` literal text all stay valid.
+    for pattern in [
+        "[[:space:]]",
+        "[::]",
+        "[:space]",
+        "[:spac-e:]",
+        "[a:space:]",
+    ] {
+        let (_s, mut c) = ucmd();
+        c.args(&[pattern])
+            .pipe_in("z\n")
+            .fails_with_code(1)
+            .no_output();
+    }
+
+    // `\[` does not open a bracket expression.
+    let (_s, mut c) = ucmd();
+    c.args(&["\\[:space:]"])
+        .pipe_in("z\n")
+        .fails_with_code(1)
+        .no_output();
+
+    // `-F` treats the text literally, so no diagnostic.
+    let (_s, mut c) = ucmd();
+    c.args(&["-F", "[:space:]"])
+        .pipe_in("x\n")
+        .fails_with_code(1)
+        .no_output();
+}
+
+#[test]
+fn reversed_range_uses_gnu_wording() {
+    // A range like `[b-a]` is an error; GNU prints the bare POSIX diagnostic
+    // "Invalid range end" (not oniguruma's phrasing) and exits 2.
+    // No piped input: the pattern is rejected before stdin is read.
+    for args in [&["[b-a]"][..], &["-E", "[b-a]"][..]] {
+        let (_s, mut c) = ucmd();
+        c.args(args)
+            .fails_with_code(2)
+            .stderr_is("grep: Invalid range end\n");
+    }
+}
+
+#[test]
+fn pcre_backtracking_limit_does_not_abort() {
+    // A pathological PCRE pattern can exceed oniguruma's retry limit. GNU
+    // grep reports this and exits 2 (it must not crash); stdout stays empty.
+    let (_s, mut c) = ucmd();
+    c.args(&["-P", "((a+)*)+$"])
+        .pipe_in("aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaab\n")
+        .fails_with_code(2)
+        .stdout_is("")
+        .stderr_contains("backtracking limit");
+}
+
+#[test]
+fn invalid_backreference_uses_gnu_wording() {
+    // A back-reference to a non-existent group is worded differently by GNU
+    // depending on the engine: PCRE (-P) vs gnulib regex (BRE/ERE).
+    // No piped input: these patterns are rejected before stdin is read.
+    let (_s, mut c) = ucmd();
+    c.args(&["-P", r"(.)\2"])
+        .fails_with_code(2)
+        .stderr_is("grep: reference to non-existent subpattern\n");
+
+    for args in [&["-E", r"(.)\2"][..], &[r"\(.\)\2"][..]] {
+        let (_s, mut c) = ucmd();
+        c.args(args)
+            .fails_with_code(2)
+            .stderr_is("grep: Invalid back reference\n");
+    }
+
+    // A valid back-reference with -Pw / -Px must still match.
+    for flag in ["-Pw", "-Px"] {
+        let (_s, mut c) = ucmd();
+        c.args(&[flag, r"(.)\1"])
+            .pipe_in("aa\n")
+            .succeeds()
+            .stdout_is("aa\n");
+    }
+}
+
+#[test]
+fn grep_colors_mt_and_grep_color_deprecation() {
+    // GREP_COLORS `mt` sets the match color (both selected and context).
+    let (_s, mut c) = ucmd();
+    c.args(&["--color=always", "."])
+        .env("GREP_COLORS", "mt=36")
+        .pipe_in("x\n")
+        .succeeds()
+        .stdout_is("\u{1b}[36m\u{1b}[Kx\u{1b}[m\u{1b}[K\n");
+
+    // GREP_COLOR is deprecated: it still sets the match color, but emits a
+    // warning when color is actually produced.
+    let (_s, mut c) = ucmd();
+    c.args(&["--color=always", "."])
+        .env("GREP_COLOR", "36")
+        .pipe_in("x\n")
+        .succeeds()
+        .stdout_is("\u{1b}[36m\u{1b}[Kx\u{1b}[m\u{1b}[K\n")
+        .stderr_is("grep: warning: GREP_COLOR='36' is deprecated; use GREP_COLORS='mt=36'\n");
+
+    // No warning when color output is disabled.
+    let (_s, mut c) = ucmd();
+    c.args(&["--color=never", "."])
+        .env("GREP_COLOR", "36")
+        .pipe_in("x\n")
+        .succeeds()
+        .stdout_is("x\n")
+        .no_stderr();
+}
+
 #[test]
 fn fixed_string_is_literal() {
     // Metacharacters are not interpreted.