Are we sure this will not break pyiodide?

Not positive, but it looks like they've removed their dummy_threading patches?
pyodide/pyodide#796
pyodide/pyodide#144 (comment)

Ping@rth for comment on whether this will break Pyodide because it looks like you did some of the threading work over in Pyodide.

I also side-banded@mdboom , we are good to go here!

Merging over the OSX failures which are due to tk installation issues.

For Pyodide it should be OK, since indeed lately importingthreading no longer raised anImportError there, even though threads could still not be created.

I think the only thing that could be improved is to add

try:timer.start()exceptRuntimeError:pass

in the FontManager init, as otherwise starting that thread would failin Pyodide,

>>>timer=threading.Timer(1,lambdax:print('a'))>>>timer.start()Traceback (mostrecentcalllast):File"<console>",line1,in<module>File"/lib/python3.10/threading.py",line928,instart_start_new_thread(self._bootstrap, ())RuntimeError:can'tstartnewthread

(timer.cancel() appears to not fail even if there was an exception for starting the thread)
This would allow us to remove the following patchhttps://github.com/pyodide/pyodide/blob/main/packages/matplotlib/patches/fix-threading.patch. Also see related discussion inhttps://discuss.python.org/t/expected-behavior-for-unsupported-stdlib-modules-in-the-browser/15689

Would changing the warning timer tosignal.alarm work with pyodide here?

I don't see a major issue with adding the try/except either, the only minor issue is that someone else will come along again next time and say it isn't necessary because the CPython docs say it should be available :)

I would rather try...except than go down the signal path.

I think of this is a secondary feature (it is warning that building the font cache is going slow which if you have oodles of fonts can take a long time and look like a process hang) so if it is degraded on pyodide that is OK and we should un-break them as simply as possible on our side.