adjusting socketserver.BaseServer

[tungol]

This one might be a bad idea. fileno, get_request, and server_bind are documented to exist on BaseServer, but they don't. Of these, only get_request is actually called from the existing implementation of BaseServer (as part of handle_request and serve_forever).

In the ideal case, they'd probably all be abstract methods in CPython.

[tungol]

No mypy primer hits, but I'm not convinced that means this is the right thing to do.

[tungol]

I opened a CPython issue about it: python/cpython#127209

[tungol]

I could also try making them abstract in the stubs, even if CPython doesn't.

[JelleZijlstra]

Any specific reason to close this? I haven't been very on top of reviewing typeshed PRs (lots of other things keeping me busy) but I do intend to get back to them.

[tungol]

Yeah, this one has been sitting in draft because I kind of intend to pursue python/cpython#127209 first and see where that goes. This had merge conflicts again, and since I didn't intend to take it out of draft anytime soon I decided to close it instead of fix the merge conflict. I'll get back to it at some point, but I didn't feel great about this particular solution.

[tungol]

Okay, I'm ready to re-open this now. I have an MR out at CPython which this aligns with: python/cpython#127976

The one thing we might want to do differently here in typeshed is that the socket attribute is used by BaseServer.handle_request right now. I think that's a mistake, and my MR would change that, but right now a subclass of BaseServer which doesn't create a socket attribute for itself is not going to work. I moved it to TCPServer here anyway, but it'd also be reasonable to keep socket as an attribute of BaseServer until that fix is available.

I was thinking it wasn't useful to lie about the existence of BaseServer.get_request, but then I realized it probably allows mypy to flag if a child class implements it with the wrong signature. That's enough of a benefit to keep it around. Assuming my other MR is accepted, it'll stop being a lie eventually.

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[srittau]

Thanks, looks to be much closer to the implementation and intent of upstream. Hope your upstream PR gets merged.