The Linux gcc-11 test leg is intermittently failing with a heap-corruption crash at DF shutdown, after the test suite reports all tests passed. The subsequent DF launch in the same attempt then aborts at startup.
Observed on two unrelated branches:
Tail of stderr.log from both runs is identical:
Tests completed in 10920 ms:
27/27 tests passed
185/185 checks passed
Invoking: lua scr.breakdown_level=df.interface_breakdown_types.QUIT
free(): invalid pointer
and on add-tests:
Tests completed in 17338 ms:
374/377 tests passed
5581/5583 checks passed
Invoking: lua scr.breakdown_level=df.interface_breakdown_types.QUIT
free(): invalid pointer
So the heap is already corrupted (or the teardown path frees a bad pointer) before DF exits; the test suite itself is green. Retries in the same attempt then fail earlier with Aborted (core dumped) right after dfhack: redirecting stderr to stderr.log, suggesting the first crash leaves the DF directory or process state in a way that breaks the follow-up launch.
This is separate from the Windows 60s-retry timeout that #5943 addresses, and it is currently masking otherwise-green runs.
The Linux gcc-11 test leg is intermittently failing with a heap-corruption crash at DF shutdown, after the test suite reports all tests passed. The subsequent DF launch in the same attempt then aborts at startup.
Observed on two unrelated branches:
Tail of
stderr.logfrom both runs is identical:and on add-tests:
So the heap is already corrupted (or the teardown path frees a bad pointer) before DF exits; the test suite itself is green. Retries in the same attempt then fail earlier with
Aborted (core dumped)right afterdfhack: redirecting stderr to stderr.log, suggesting the first crash leaves the DF directory or process state in a way that breaks the follow-up launch.This is separate from the Windows 60s-retry timeout that #5943 addresses, and it is currently masking otherwise-green runs.