Better integration with faulthandler

[enkore]

One reason to use xdist is to run multithreaded tests in which case errors in the code under test tend to signal, terminating the slave. xdist allows to continue running tests in these circumstances.

It would be nice if it were possible to attach stack trace(s) (e.g. via faulthandler) or logs to such events from within the application code (immediately before signalling).

Currently the next best thing is to
(a) either have pytest-faulthandler set faulthandler up before any redirection happens, which will spew the stack traces as they happen (so they won't be reported in the test failure details). This doesn't allow to attach custom log text.
(b) run in a TTY and spew to /dev/tty which has the same problems as (a) plus it requires a TTY. This does allow for extra text.

[RonnyPfannschmidt]

na integration of pytest-faulthandler and execnet is potentially necessary - im not familiar withthe tooling

[rggjan]

I think the faulthandler stuff is part of upstream python and pytest now? Might have gotten easier to integrate?

[nicoddemus]

The problem is not really the integration with pytest, but handling worker streams: faulthandler uses that streams to output errors, and execnet (pytest-xdist's backend) uses the same streams for communication.