add socket.all by tungol · Pull Request #13044 · python/typeshed · GitHub

I didn't realize there were so many errors being hidden for linux in here when I started. I only have access to MacOS and FreeBSD locally, so I just took the linux pipeline results at mostly face value. In working through all the things that don't exist in the Linux pipeline, I tested against my BSD machine to sort out what was for example, platform = "win32" versus platform != "darwin" and platform != "linux" or platform = "darwin" versus "platform != "win32" and platform != "linux", etc.

When it didn't seem like any system that had a thing, I commented it out in _socket and removed it from socket. I expect many of these exist on some linux system somewhere - there's a lot of variance in that space. It seemed like leaving them as a comment in _socket was a good placeholder for now.

…what should happen here

If I've done this correctly, this irons out the errors that are different depending on python version, leaving just the errors that show up for every version of linux.

Let's not comment out any of these constants. They're all defined with IFDEF in CPython and presumably exist somewhere, though I suspect some may exist only on versions of FreeBSD from the mid-1990s compiled while Mars is in retrograde. If you want a rabbit hole to dig into, you could try to figure out exactly where those constants come from (e.g., by digging through CPython git blame or Googling for manpages from the right OS).

Let's not comment these out; they definitely exist on some platforms.

I'll do some more digging on the mystery constants.

Thanks! To be clear, I don't think it's too important to figure out exactly when these constants exist, and I'd be fine with defining them under if sys.platform != "linux" and sys.platform != "win32" and sys.platform != "darwin" if we know they don't exist on any of the three major platforms.

Okay, that makes sense. In the case of the ones I commented out, I also confirmed that they didn't exist on (my local version of) FreeBSD either.

Updated, with 7 constants still commented out. 6 of these are linux-only NETLINK_* constants that haven't been in linux for almost 20 years now. One is the linux-only RDS_CMSG_RDMA_UPDATE which I couldn't find in any version of linux and I think is a typo for the actually existing RDS_CMSG_CONG_UPDATE.

Most of the rest I confirmed are in linux header files to this day, so I restored them to both _socket and socket without adding them to __all__. Those that I couldn't find in a linux header file went to platform != the big three branches.

Thanks!

Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>

Suggested change

      
                # This is supported by cpython but doesn't seem to be a real thing.
          
                # This is supported by CPython but doesn't seem to be a real thing.

If you want extra brownie points, you could consider opening an issue at CPython about this ;)

CPython issue for this constant: python/cpython#127069

Suggested change

      
                # Technically still supported by cpython
          
                # Technically still supported by CPython

And a CPython issue for these outdated Netlink constants as well: python/cpython#127072

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

This should not have disabled all of the Bluetooth-related socket constants on Linux.

add socket.__all__

f687707

tungol marked this pull request as draft November 19, 2024 00:05