Subprocess crashes not reported within blocks

[mqudsi]

~> foo
fish: “foo…” terminated by signal SIGABRT (Abort)
~> begin; foo; end;
~>

Gave me a hell of a time trying to understand why my script was failing without any visible errors.

[faho]

This seems to have been originally added on purpose. See parse_execution.cpp:run_1_job():

    job->set_flag(JOB_SKIP_NOTIFICATION,
                  is_subshell || is_block || is_event || !shell_is_interactive());

It seems safe to remove blocks from this, though I'd probably even report crashes (SIGABRT) for events and "subshell"s (this is talking about command substitutions).

[faho]

Yeah, this was added way back in 2005, in 838ba08.

With the full commit message of "Event handling and job reaping updates".

[faho]

Bash's policy here seems to be to notify about crashes except in command substitutions - even if the shell isn't interactive!

So that would be

    job->set_flag(JOB_SKIP_NOTIFICATION,
                  is_subshell);

Which I don't quite get.

Anyone got any ideas for a better policy than what we have now?

[faho]

So.... how about we just always report crashes?

I'm not entirely sure how to define "crash", though. My best attempt is "everything that was killed via one of the following signals: SIGABRT, SIGBUS, SIGFPE, SIGILL, SIGSEGV, SIGSYS".

[mqudsi]

I would be on board with that.

[faho]

Gracias!