Nathan Goldbaum

Argh: the test timing is flaky. There's probably a better way to synchronize things but I'm running out of steam. Suggestions to improve the tests are very welcome :)

It also occurred to me while working on this that the ubuntu wheel builds probably benefit from the new aarch64 CI runners on github actions. That would allow skipping the...

ping @cclauss @zmedico - do you want to look this over?

So, this isn't thread-safe: https://github.com/airspeed-velocity/asv/blob/5b01ec1a1bb9a9f43742456a40c91df75f675cac/asv/_rangemedian.cpp#L231-L234 It could be made thread-safe pretty easily using [`std::call_once`](https://en.cppreference.com/w/cpp/thread/call_once.html). It does look like asv uses `threading` internally to track output streams, but it doesn't try...

> I wonder if we should be promoting pymutex as the first choice rather than critical sections in the documentation? This is tough because without conditional compilation, any new locks...

My view is more often people want the same amount of thread safety they used to have on the GIL-enabled build. That is, they don't want to deal with new...

I think that might help a bit, especially since `DEF` and `IF` are discouraged. That said, I think warnings guiding people toward using critical sections safely long with some community...

As a big proponent of meson-python IMO it's a great choice for this.

I hid all the comments about removing scalars completely. There are other issues for that discussion and they distract from the main issue here.

Ping @colesbury - what do you think about using an RWLock based on `PyMutex` instead of `std::shared_lock`?