import setproctitle after forking on macOS

[Otto-AA]

Fixes #446

On macOS this likely costs a few miliseconds per mutation (the import takes +-1ms on my linux system).

[nkakouros]

What does this need to move forward?

[Otto-AA]

I think the #476 PR makes sense (not using setproctitle on MacOS), but with an opt-in config to use setproctitle.

The reasons I haven't fixed it yet are:

• minimal repros don't really work (i.e. making a small python script with setproctitle and afterwards os.fork() does not crash). This makes it hard to find workarounds
• I work on Linux, hard to reproduce again :)
• there's documentation that importing setproctitle after forking (as done in this PR) should work (see this comment or this PR), but apparently it does not work in mutmut (see this comment)?? 🤯
• even when we remove setproctitle from mutmut, running mutmut on an app with setproctitle or similar will likely also crash

So I don't really understand the bug. Sorry it's taking long, I guess it's time to fix it even without really understanding it.

[nicklafleur]

From what I can tell this is an issue with the MacOS foundations API and setproctitle's C bindings/implementation being not fork safe with python's fork library. There is a way to avoid this entirely with subprocess, but that comes at the cost of the performance improvements that come with fork which are pretty significant when killing a mutants can be <0.5s with forking but need an additional 2s of import time (if you're lucky in large codebases) with subprocesses.

I do have a future PR (commits can be found at https://github.com/lyft/mutmut/tree/nicklafleur/oss-complete) which implements a few different forking strategies (mainly due to the fact that we also use libraries like gunicorn/gevent/etc which pollute the internal state of the main process). We could absolutely add a "subprocess" mode to this if the need is there. It was actually the initial implementation but I moved away due to poor performance.