Teach trio.socket.fromfd and friends to autodetect socket configuration on Linux

[2easy]

Hi, I'd like to propose this solution to the issue #251 .

[2easy]

There is problem with SO_PROTOCOL on mac, will fix that in a moment.

[codecov]

Codecov Report

Merging #577 into master will decrease coverage by 2.16%.
The diff coverage is 93.93%.

@@            Coverage Diff            @@
##           master    #577      +/-   ##
=========================================
- Coverage   99.27%   97.1%   -2.17%     
=========================================
  Files          89      89              
  Lines       10628   10656      +28     
  Branches      747     750       +3     
=========================================
- Hits        10551   10348     -203     
- Misses         59     274     +215     
- Partials       18      34      +16

Impacted Files	Coverage Δ
trio/tests/test_ssl.py	100% <ø> (ø)	⬆️
trio/tests/test_socket.py	97.83% <ø> (-2.17%)	⬇️
trio/_socket.py	94.94% <93.93%> (-5.06%)	⬇️
trio/_core/_windows_cffi.py	0% <0%> (-91.31%)	⬇️
trio/_core/_io_windows.py	0% <0%> (-77.1%)	⬇️
trio/_core/tests/test_windows.py	23.33% <0%> (-76.67%)	⬇️
trio/_highlevel_open_unix_stream.py	66.66% <0%> (-11.12%)	⬇️
trio/_util.py	92.68% <0%> (-7.32%)	⬇️
trio/tests/test_highlevel_open_unix_stream.py	93.54% <0%> (-6.46%)	⬇️
trio/_highlevel_open_tcp_listeners.py	94.73% <0%> (-5.27%)	⬇️
... and 8 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fd0d877...0a86243. Read the comment docs.

[njsmith]

Hello!

I'm afraid I have misled you and made this much more complicated than it has to be :-(. The code you used for a reference jumps through all kinds of hoops using ctypes, because it's trying to call getsockopt on a socket that it only has a fileno for, not a socket object. But on all the versions of Python we care about, we can create a socket object (maybe with incorrect values), and then we can directly use Python's regular getsockopt method and skip all that ctypes stuff. Also, the versions of Python we care about have the relevant SO_WHATEVER magic constants on the socket module on all the systems where they work at all, so we don't need to have the magic numbers like "39" for SO_DOMAIN. (And besides, those magic numbers are only correct for Linux.)

So my suggestion would be to delete all the ctypes stuff, and instead have a function like this, that takes whatever the user gave us, fixes it up as best we can (which might be more or less depending on the OS), and returns the fixed up values.

def _fix_attrs_for_fileno(family, type, proto, fileno):
    # Wrap the raw fileno into a socket object
    sockobj = _stdlib_socket.socket(fileno=fileno)
    try:
        if hasattr(socket, "SO_DOMAIN"):
            family = sockobj.getsockopt(SOL_SOCKET, SO_DOMAIN)
        if hasattr(socket, "SO_TYPE"):
            type = sockobj.getsockopt(SOL_SOCKET, SO_TYPE)
        if hasattr(socket, "SO_PROTOCOL"):
            proto = sockobj.getsockopt(SOL_SOCKET, SO_PROTOCOL)
    finally:
        # Unwrap it again, so that sockobj.__del__ doesn't try to close our socket
        sockobj.detach()
    return family, type, proto

And then in fromfd and socket we can do something like:

family, type, proto = _fix_attrs_for_fileno(family, type, proto, fileno)

Does that make sense?

[njsmith]

@2easy Hey, just wanted to check in to see whether you're still interested in working on this :-)

[2easy]

Sure thing, sorry for the delay. I'll fix it in the evening.

[njsmith]

Haven't heard anything for a week, so I'm going to close this for now to take it off the list of active PRs. Feel free to reopen it, or open a new PR, if you do decide to come back to it later! We've all been there :-)