Feature #20053: M:N Threads, now w/ macOS support (kqueue)

• Status: Closed
• Assignee: ko1 (Koichi Sasada)
• Target version: 3.3

[jpcamara (JP Camara)]

I have a PR here that adds support for macOS to M:N threads using kqueue: https://github.com/ruby/ruby/pull/9178

It adds in kqueue/kevent calls when present in the OS. Technically this would open up support for FreeBSD as well, but I don't have a way of testing that so i'm not sure how well it does or doesn't work there.

I wanted to get support going for macOS so more devs can try out M:N threads and test it. I do think there should be a larger discussion around the potential relationship between this and some of the awesome fiber scheduler work over the last few years (and the potential of utilizing anything from io-event?). But that can be a topic for another day - let's get macOS support in the meantime!

Disclaimer: C isn't my day-to-day language, so I could definitely use feedback there. I'm also more of a kernel event queue (kqueue, epoll, io_uring) enthusiast, but kqueue isn't something I have specific experience writing with - just lots of reading code and small toy code up until this point. So someone like @ioquatix (Samuel Williams) would probably be a good person to get eyes on it.

[byroot (Jean Boussier)]

• Assignee set to ko1 (Koichi Sasada)

[jpcamara (JP Camara)]

@ko1 (Koichi Sasada) you’ll see a huge amount of discussion on the PR between me and @ioquatix (Samuel Williams) - you can ignore all of that and just look at the PR code. I will open up a separate issue to discuss that separately from this kqueue work.

I’ve made some changes since I first opened the PR - some small refactors for clarity, and all the tests pass now. It’s stable now and I won’t make any other changes (unless you request some!).

[jpcamara (JP Camara)]

@ko1 (Koichi Sasada) My updates are in based on your feedback. At this point is it too close to the Ruby 3.3 release to potentially get Mac support in?

I also responded about the scenario in #readpartial where native threads end up getting created in MN threads. I think I may have a solution - but since it was present before this PR, i'll open a separate PR for the fix.

[hsbt (Hiroshi SHIBATA)]

• Target version set to 3.3

[hsbt (Hiroshi SHIBATA)]

• Status changed from Open to Assigned

[hsbt (Hiroshi SHIBATA)]

• Status changed from Assigned to Closed

https://github.com/ruby/ruby/pull/9178 has been merged.