Sway 1.12: Possible race condition (headless + pixman)

[a10cat]

Trying to run headless and with the pixman renderer

Sway crashes on startup, but doesn't when running in gdb, with the same options

❯ env WLR_RENDERER="pixman" WLR_BACKENDS="headless" sway --verbose --debug

vs

❯ gdb --args env WLR_RENDERER="pixman" WLR_BACKENDS="headless" sway --verbose --debug

version:

❯ sway --version
sway version 1.12

logs and configs: https://gist.github.com/a10cat/19a58191fd1ec358fd7f4c006cd083fc

Logs are exactly the same until the "gdb-less" one dies

[emersion]

Do you have a stack trace? (bt full in GDB)

[a10cat]

hi @emersion ,

As mentioned, it runs perfectly when executed from within gdb. So no chance to get a backtrace
Also, the normal case doesn't generate any coredump :(

[emersion]

Can you try ASan maybe?

[a10cat]

ASAN reported nothing.
With TSAN there were some warnings, but looked fine to me (E.g. 1st access is malloc from one thread, and later accessed from another thread)
Maybe I'll give UBSAN a go afterwards

After tinkering around a bit more, I found out a SIGKILL is getting received. I'm trying to trace it to its source/reason now