I was puzzled why e.g. <https://ci.libreoffice.org//job/lo_tb_master_linux_dbg/45699/consoleFull#17836743462b3c3940-aa94-44b1-926f-2d1e7de0f995>, where JunitTest_starmath_unoapi failed due to a soffice.bin SIGABRT, claimed "No core file identified" and didn't show a gdb-generated backtrace of the crash. It ran on tb99. And lode's bin/tb_slave_wrapper does include `ulimit -c unlimited`.

But tb99 presumably now uses systemd coredumpctl and a /proc/sys/kernel/core_pattern that prevents creation of core.* files in CWD, and instead sends them off to coredumpctl. We have <https://git.libreoffice.org/core/+/d7ba78e9c7be835a1e2ecdacd25995663e96862f%5E!/> "New --with-coredumpctl to obtain core dumps of crashed tests from coredumpctl" to handle that situation. But it requires a few tools to be available for our configure.ac to auto-enable --with-coredumpctl, namely coredumpctl, jq, systemd-escape, and systemd-run.

And it appears that jq is missing on tb99. Can you please make sure that all the relevant Linux machines have all those four tools available? Thanks!

Reply via email to