Who doesn't like multiprocessing?

[goodboy]

I've been hearing grumblings about distaste for multiprocessing. I can't discount these opinions because my inability to get a reliable system with multiprocessing is one of the main reasons I started building this project.

I do think it would be interesting to start some work on alternatives for process spawning/forking.

A couple ideas I have offhand:

• use trio's new built-in process spawning as an alternative spawn method
• implement our own "forkserver" on nix systems so that we don't have to deal with the stdlib quiffs I have
• consider alternative fork implementations such as @catern's sfork which is wrapped for python as the supervise package
• investigate how other projects such as distex and pulsar do it (I guess by looking inside asyncio.create_subprocess_exec().

Very open to opinions and ideas here!

[catern]

sfork is fairly experimental (and I never released a version of supervise that uses it - all the released versions use os.fork). rsyscall is the more robust form of it.

[goodboy]

@catern thanks for the clarification 👍 and my bad for the misread.

[goodboy]

Thanks to @ryanhiebert for finding trio_run_in_process which already implements the

use trio's new built-in process spawning as an alternative spawn method

approach!