Repository navigation
test_ask_relief: 'the waiting core's other actors run' races a 400 ms pause (11 of 20 under macOS ASan) #3483
Description
tests/runtime/test_ask_relief.c, "Ask relief: the waiting core's other actors run", failed on the macOS ASan leg of PR #3477 (a docs and examples PR that changes no runtime code; the same job passes on main):
ASSERT_EQ failed at tests/runtime/test_ask_relief.c:154: expected 20, got 11
[FAIL] Ask relief: the waiting core's other actors run (444ms)
The test sleeps 50 ms after sending A, then sends D 20 messages, and expects all 20 handled (and counted in a->watched) while A waits 400 ms for B's reply. Under AddressSanitizer on a busy macOS runner, 11 of the 20 got through before the reply: the assertion is bounded by wall-clock time, not by an event. The check should wait for D to have taken the 20 (or hold B's reply until it has) rather than race a 400 ms pause.
Filed for the next batch; the failed job was re-run.
Activity
nicolas-maman commentedon Oct 30, 2026
Again on #3477 (ba6da17), under Valgrind on Linux this time: the same assertion (
test_ask_relief.c:154), and because a failed ASSERT returns before the test'sfree(a); free(b); free(d), Valgrind also reports its actors definitely lost (1,760 bytes direct, 3,520 indirect, fromrelief_actorat test_ask_relief.c:84). Fixing the wait should end both; the test could also free on its failure path.test: ask relief waits for D's messages, not a 400 ms pause (#3483)
test: ask relief's B on core 2, its hold inside the asker's timeout (#…