Ah, you implemented that change to the PyInstaller bootloader after it had already influenced the staticx bootloader (167a585). No wonder it's not there! 😄

I agree, all signals should be forwarded, not just termination signals, with the exception of SIGCHLD as you indicate, and SIGKILL and SIGSTOP, since according to signal(7):

The signals SIGKILL and SIGSTOP cannot be caught, blocked, or ignored.

I also think there's a mistake in pyinstaller/pyinstaller@2ddb80d in the bounds of the for() loop: It should start at 1, not 0. And the upper bound is technically correct, but it is a bit misleading: I would use num_signals = 64, and use i <= num_signals.

I'm experimenting with an implementation now, but I'm seeing some strange differences between signal() and sigaction() on musl vs glibc... stay tuned.

I've posted a draft PR in #242.

The "strange differences" I was referring to were the fact that SIGRTMIN starts at 35 on musl-libc (32-34 are reserved for pthreads implementation I believe), and thus sigaction() was failing.

I'm still not sure we want to forward all signals 1-31 with SIGCHLD, SIGKILL, and SIGSTOP in a denylist. If you read all of the signals listed on signal(7), I think there are plenty of others that we probably don't want to forward, too. Specifically ones that originate from hardware exceptions like SIGSEGV, SIGILL, SIGBUS, SIGFPE, etc.

I think we might consider instead an allowlist approach, including e.g.

• SIGABRT (maybe)
• SIGINT
• SIGQUIT
• SIGTERM
• SIGUSR1
• SIGUSR2

So long as the allow list is configurable when running staticx (with some reasonable defaults as you listed), that sounds just fine to me. Thanks for looking into it!

In my original case, I was wrapping a legacy application which arbitrarily used signal 42. (A lot has changed since 2018, though, and I could change this to something else now if needed.)