# ./test-lock Starting test_lock ... OK Starting test_rwlock ... OK Starting test_recursive_lock ... OK Starting test_once ... test_once hangs on ia64 most of the time (I have seen it pass after a few seconds) s/#define ENABLE_DEBUGGING 0/#define ENABLE_DEBUGGING 1/ in test-lock.c gives some insight: Contender 0x2000000005d331f0 waiting for signal for round 29451 Contender 0x2000000005d331f0 got the signal for round 29451 Contender 0x20000000025331f0 got the signal for round 29451 Contender 0x2000000002d331f0 got the signal for round 29451 Contender 0x2000000001d331f0 got the signal for round 29451 Contender 0x2000000001d331f0 waiting for signal for round 29452 Contender 0x20000000045331f0 got the signal for round 29451 Contender 0x2000000002d331f0 waiting for signal for round 29452 Contender 0x20000000025331f0 waiting for signal for round 29452 Contender 0x20000000055331 Definitely looks like a dead lock. Cc'ing toolchain@ since I expect this will be a glibc bug. I am using glibc-2.23-r3.
Tried glibc-2.24-r1 with the same result.
Poked at it a bit today. test-lock is a gnulib's test (not a libunistring-specific test). A few facts: - libunistring from git passes that test (seemingly consistently) - gnulib has quite a few fixes for test-lock since unistring release cut: http://git.savannah.gnu.org/gitweb/?p=gnulib.git;a=history;f=tests/test-lock.c;h=fb18dee9aa02a55c867118e1203c1e08eed79554;hb=HEAD
(In reply to Sergei Trofimovich from comment #2) > Poked at it a bit today. test-lock is a gnulib's test (not a > libunistring-specific test). > > A few facts: > - libunistring from git passes that test (seemingly consistently) > - gnulib has quite a few fixes for test-lock since unistring release cut: > http://git.savannah.gnu.org/gitweb/?p=gnulib.git;a=history;f=tests/test-lock. > c;h=fb18dee9aa02a55c867118e1203c1e08eed79554;hb=HEAD Managed to deadlock gnulib's test from git as well. It's not fixed.
Looks like after v0.9.7 test-lock passes: Starting test_lock ... OK Starting test_rwlock ... OK Starting test_recursive_lock ... OK Starting test_once ... OK glibc is 2.27-r6.
Looks like this was fixed by https://git.savannah.gnu.org/gitweb/?p=gnulib.git;a=commit;h=480d374e596a0ee3fed168ab42cd84c313ad3c89 and then new gnulib was imported into libunistring. I'll file a stabilization request for a newer version.
Ah, nice! I had a reproducer that was able to trigger on guppy reliably until I updated kernel there.
Newer version was marked stable on ia64.
