On Mon, Aug 12, 2019 at 10:30 AM Alexander Korotkov <a.korotkov@postgrespro.ru> wrote: > > > This appears to have upset prion when testing on en_US.iso885915. > > > > Also lapwing's "InstallCheck-fr_FR" stage crashed on this commit, when > > running JSON queries, on HEAD and REL_12_STABLE:
> Thank you for pointing! I hope I can investigate this shortly.
Hi Alexander,
I can reproduce this by running LANG="fr_FR.ISO8859-1" initdb, then running installcheck (on some other OSes that might be called just "fr_FR"). See this comment in mbutils.c:
* The functions return a palloc'd, null-terminated string if conversion * is required. However, if no conversion is performed, the given source * string pointer is returned as-is.
You call pfree() on the result of pg_server_to_any() without checking if it just returned in the input pointer (for example, it does that if you give it an empty string). That triggers an assertion failure somewhere inside pfree(). The following fixes that for me, and is based on code I found elsewhere in the tree.
- pfree(utf8str1); - pfree(utf8str2); + if (mbstr1 != utf8str1) + pfree(utf8str1); + if (mbstr2 != utf8str2) + pfree(utf8str2);
With that fixed it no longer crashes, but then the regression test fails due to differences in the output, which look like locale ordering differences.
Thank you for the diagnostics. Should be fixed by 251c8e39.
BTW, test failures appears to be caused not by locale differences, but by using strlen() on non null-terminated original strings.
------ Alexander Korotkov Postgres Professional: http://www.postgrespro.com The Russian Postgres Company