summaryrefslogtreecommitdiff
path: root/regparse.c
diff options
context:
space:
mode:
authorPeter Zhu <[email protected]>2024-01-08 10:09:13 -0500
committerPeter Zhu <[email protected]>2024-01-08 19:50:34 -0500
commit37ed86fd3c798e298fad9db6e7df1f3f45e1e03b (patch)
tree4f11b417a69e2d476f39d3a46870a5fcb538dfda /regparse.c
parent0a30fc6211230f23a84d63344cba32e879ab4770 (diff)
Fix memory leak in regexp grapheme clusters
[Bug #20161] The cc->mbuf gets overwritten, so we need to free it to not leak memory. For example: str = "hello world".encode(Encoding::UTF_32LE) 10.times do 1_000.times do str.grapheme_clusters end puts `ps -o rss= -p #{$$}` end Before: 15536 15760 15920 16144 16304 16480 16640 16784 17008 17280 After: 15584 15584 15760 15824 15888 15888 15888 15888 16048 16112
Diffstat (limited to 'regparse.c')
-rw-r--r--regparse.c3
1 files changed, 2 insertions, 1 deletions
diff --git a/regparse.c b/regparse.c
index 1ce15da8f5..57ecd78dee 100644
--- a/regparse.c
+++ b/regparse.c
@@ -6105,7 +6105,8 @@ node_extended_grapheme_cluster(Node** np, ScanEnv* env)
R_ERR(add_code_range(&(cc->mbuf), env, 0x000A, 0x000A)); /* CR */
R_ERR(add_code_range(&(cc->mbuf), env, 0x000D, 0x000D)); /* LF */
R_ERR(not_code_range_buf(env->enc, cc->mbuf, &inverted_buf, env));
- cc->mbuf = inverted_buf; /* TODO: check what to do with buffer before inversion */
+ bbuf_free(cc->mbuf);
+ cc->mbuf = inverted_buf;
env->warnings_flag &= dup_not_warned; /* TODO: fix false warning */
}