GNU grep flags a bracket expression of the form [:name:] (an almost
certain misspelling of [[:name:]]) with a dedicated diagnostic and exit
code 2, whereas oniguruma silently treats it as the character set
{':','n','a','m','e'}. Port GNU's colon_warning_state logic from
parse_bracket_exp (gnulib dfa.c) so basic/extended patterns produce the
same error. Fixes the GNU testsuite 'warn-char-classes' test.
GNU grep tolerates options given more than once (boolean flags are
idempotent, value options take the last occurrence), but clap rejected
them with "cannot be used multiple times" (exit 2). Set
args_override_self(true) so clap replaces rather than errors; args with
ArgAction::Append (-e/-f/--include/--exclude) still accumulate.
Found by the differential fuzzer (e.g. `grep -e .* -E -n -o -n -x`).
Add a regression test covering repeated booleans, repeated value options
(last wins), and that -e still accumulates patterns.
Add a GnuComment workflow that runs after GnuTests completes on a pull
request, downloads the 'comment' artifact (PR number + comparison text), and
posts it as a PR comment. Mirrors ../sed's GnuComment workflow.
Add a GnuTests workflow that, on push and PR, fetches the GNU grep release
tarball, builds the Rust grep binary, runs util/run-gnu-testsuite.sh, and
uploads the JSON results. An aggregate job compares the run against the
reference summary from the default branch (util/compare_test_results.py,
borrowed from ../sed) and fails only on new, non-intermittent regressions;
known-flaky tests are listed in .github/workflows/ignore-intermittent.txt.
Add util/fetch-gnu.sh (downloads the GNU grep 3.12 release tarball from
ftp.gnu.org) and util/run-gnu-testsuite.sh, which reuses the gnulib test
framework shipped in the tarball (tests/init.sh + init.cfg) and injects the
Rust grep binary via PATH, replicating tests/Makefile.am's TESTS_ENVIRONMENT.
Each test is classified by its gnulib exit code (0=PASS, 77=SKIP, else FAIL)
and results are emitted as JSON.
Modelled on ../sed (lightweight PATH-injection runner) and ../coreutils
(release-tarball fetch). Current baseline: 61 pass / 39 fail / 28 skip of 128
tests -- the failures quantify the remaining GNU-compatibility gap.
The test suite spawns the grep binary inside per-test temporary
directories (via uutests). With a relative LLVM_PROFILE_FILE, each
subprocess wrote its .profraw into that tempdir, which gets deleted
when the test ends, so the binary executing the src/ code left no
coverage data and grcov reported every line as a miss (0%).
Anchoring the path to the workspace root makes the profraw files
survive, raising reported coverage from 0% to ~96%.
Sylvestre Ledru
Leonard Hecker