On 6/23/22 00:37, Jim Meyering wrote:
On Mon, Jun 6, 2022 at 3:06 PM Paul Eggert <egg...@cs.ucla.edu> wrote:
On 6/6/22 12:37, Jim Meyering wrote:
Once you push that (and assuming you have nothing else pending), I'll
prepare another pre-release snapshot.
Thanks, I pushed it into grep master, after fixing the commentary issue
Bruno noted.
I was going to make a snapshot, but figured I should first run the
usual *SAN check:
i.e., configure && make as usual, but then "make clean" followed by
this on a linux system:
san='-fsanitize-address-use-after-scope -fsanitize=address
-static-libasan';
ASAN_OPTIONS=detect_leaks=1,allocator_may_return_null=1 make
CFLAGS='-O0 -ggdb3' AM_CFLAGS="$san" AM_LDFLAGS="$san" check
and it failed nearly every test, all due to leaks.
You can see a few leaks with an even simpler test using valgrind and a
no-ASAN binary:
(this was with the patch attached below, which fixed the first one I found)
Thanks for catching that. These are minor memory leaks in grep, not in
Gnulib, and I fixed it by installing the attached into Grep. With it,
all the Grep tests pass with your recipe. There are a few Gnulib
failures in the memory-management modules but I expect these are merely
the usual suspects.
From e2aec8c91e9d6ed3fc76f9f145dec8a456ce623a Mon Sep 17 00:00:00 2001
From: Paul Eggert <egg...@cs.ucla.edu>
Date: Fri, 24 Jun 2022 17:53:34 -0500
Subject: [PATCH] grep: fix regex compilation memory leaks
Problem reported by Jim Meyering in:
https://lists.gnu.org/r/grep-devel/2022-06/msg00012.html
* src/dfasearch.c (regex_compile): Fix memory leaks when SYNTAX_ONLY.
---
src/dfasearch.c | 24 ++++++++++++++++--------
1 file changed, 16 insertions(+), 8 deletions(-)
diff --git a/src/dfasearch.c b/src/dfasearch.c
index 8d832f0..2720b3a 100644
--- a/src/dfasearch.c
+++ b/src/dfasearch.c
@@ -144,26 +144,34 @@ regex_compile (struct dfa_comp *dc, char const *p, idx_t len,
idx_t pcount, idx_t lineno, reg_syntax_t syntax_bits,
bool syntax_only)
{
- struct re_pattern_buffer pat0;
- struct re_pattern_buffer *pat = syntax_only ? &pat0 : &dc->patterns[pcount];
- pat->buffer = NULL;
- pat->allocated = 0;
+ struct re_pattern_buffer pat;
+ pat.buffer = NULL;
+ pat.allocated = 0;
/* Do not use a fastmap with -i, to work around glibc Bug#20381. */
verify (UCHAR_MAX < IDX_MAX);
idx_t uchar_max = UCHAR_MAX;
- pat->fastmap = (syntax_only | match_icase) ? NULL : ximalloc (uchar_max + 1);
+ pat.fastmap = syntax_only | match_icase ? NULL : ximalloc (uchar_max + 1);
- pat->translate = NULL;
+ pat.translate = NULL;
if (syntax_only)
re_set_syntax (syntax_bits | RE_NO_SUB);
else
re_set_syntax (syntax_bits);
- char const *err = re_compile_pattern (p, len, pat);
+ char const *err = re_compile_pattern (p, len, &pat);
if (!err)
- return true;
+ {
+ if (syntax_only)
+ regfree (&pat);
+ else
+ dc->patterns[pcount] = pat;
+
+ return true;
+ }
+
+ free (pat.fastmap);
/* Emit a filename:lineno: prefix for patterns taken from files. */
idx_t pat_lineno;
--
2.25.1
