[Bug malloc/33899] [2.43 regression] glib tests fail since 0b9210bd760b5281f2e9f3e6640368ccb5f4a7ae

[renodr at linuxfromscratch dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=33899

--- Comment #5 from Douglas R. Reno <renodr at linuxfromscratch dot org> ---
Hello! I've been doing some digging into this because we're running into this
at Linux From Scratch, and we're going through our test process for our
upcoming release on March 1st which uses glibc-2.43.

I'm able to confirm that the gdatetime test fails over here as well. Folks who
want to reproduce the failure will need the th_TH.UTF-8 locale installed on
their system.

Similar to Sam, I've run the test under Valgrind and can't reproduce it that
way either. I have also modified it to use an alternative malloc implementation
(in this case jemalloc) and the problem does not occur over there either.

Over here, setting GLIBC_TUNABLES=glibc.malloc.tcache_count to 15 or any number
higher causes the failure to occur, but setting the tcache_count value to 14 or
lower does *not* cause a failure. This is also reproducible on glibc-2.42 if
you set GLIBC_TUNABLES=glibc.malloc.tcache_count=15 on that as well, so the
issue may be present long before "malloc: set default tcache fill count to 16",
and setting the cache value higher is just uncovering it...

I'm not quite sure how to diagnose it further from here, but I wanted to report
in with what we're experiencing as well :)

-- 
You are receiving this mail because:
You are on the CC list for the bug.