Details
Blocker Description
Latest PCRE2 10.45 introduced a behavior change as a bugfix, which lead to behavior change in MariaDB with test failing:
main.func_regexp_pcre w4 [ fail ]
|
Test ended at 2025-02-06 07:00:25
|
CURRENT_TEST: main.func_regexp_pcre
|
@@ -169,13 +169,13 @@
|
\p{Latin} 㐗 0
|
\p{Latin} 갷 0
|
\p{Ll} 1 0
|
-\p{Ll} A 0
|
+\p{Ll} A 1
|
\p{Ll} a 1
|
-\p{Ll} À 0
|
+\p{Ll} À 1
|
\p{Ll} à 1
|
-\p{Ll} Σ 0
|
+\p{Ll} Σ 1
|
\p{Ll} σ 1
|
-\p{Ll} Я 0
|
+\p{Ll} Я 1
|
\p{Ll} я 1
|
\p{Ll} ௨ 0
|
\p{Ll} ප 0
|
This was detected in Debian unstable that recently got PCRE2 updated 10.44 -> 10.45. See details in https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1095286
MariaDB could perhaps update PCRE2 to 10.45 starting from 11.8 series to avoid behavior change in a stable release?
cvicentiu has details as per discussion in https://salsa.debian.org/mariadb-team/mariadb-server/-/merge_requests/102#note_582043
Attachments
Activity
Another idea to help our userbase, should we decide to go the keep updating the PCRE library route:
parse the regex and identify the
\p{Ll} and \p{Lu}
|
characters. Issue warnings when they are encountered.
cvicentiu, my opinion always was (since a week ago, when I first saw https://salsa.debian.org/mariadb-team/mariadb-server/-/merge_requests/102) that it's a PRCE2 bug, old result was incorrect, now it is correct.
Switching to PCRE-10.45 in 11.8 sounds like a good idea.
I propose we don't spend time on parsing the regex to identify
\p{Ll} and \p{Lu}
|
As discussed with monty, we clasify this change in behaviour as a bug in PCRE implementation in the past, thus we consider the new behaviour to be the correct one. Tests will be updated accordingly. We will have a more prominent note of this on the Release notes. This will be true for all versions 10.5 -> 11.8.
bar as the original author of the test case, can you perhaps offer your input on the matter, please?