Add local_address= kwarg to open_tcp_stream

[njsmith]

Fixesgh-275

[codecov]

Codecov Report

Merging #1644 into master will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@ Coverage Diff @@## master #1644 +/- ##
=======================================
Coverage 99.69% 99.69% =======================================
Files 110 110 Lines 13858 13955 +97 Branches 1062 1074 +12 =======================================
+ Hits 13816 13913 +97 
Misses 27 27 Partials 15 15 

Impacted Files	Coverage Δ
trio/_highlevel_open_tcp_stream.py	97.33% <100.00%> (+0.36%)	⬆️
trio/socket.py	100.00% <100.00%> (ø)
trio/tests/test_highlevel_open_tcp_stream.py	100.00% <100.00%> (ø)
trio/_core/_thread_cache.py	100.00% <0.00%> (ø)
trio/_core/tests/test_run.py	100.00% <0.00%> (ø)
trio/_core/tests/test_thread_cache.py	100.00% <0.00%> (ø)
trio/_core/_run.py	99.77% <0.00%> (+<0.01%)	⬆️

[oremanj]

Thanks for the detailed comment on bind() -- I learned something!

[ntninja]

That was way too fast for me…! 😮

Do I see it correctly that with this API limiting a connection to socket.AF_INET/socket.AF_INET6 address families during name resolution is done by passing "0.0.0.0"/"::" respectively for the local_address=… arg?

[njsmith]

@ntninja Currently, open_tcp_stream always does full name resolution, but if you pass local_address="0.0.0.0" then it will only skip trying to connect to the AF_INET6 addresses and if you pass local_address="::" then it will skip trying to connect to the AF_INET addresses. Technically this is very slightly inefficient (getaddrinfo still has to look up both sets of addresses), but this is very minor, and in return, we get better error messages: e.g., if you accidentally try to connect to an IPv6-only host with local_address="0.0.0.0", then it will fail, but the error message will tell you that there were IPv6 addresses it considered connecting to and if you changed your local_address= you might succeed.

[ntninja]

Ok, thanks! I don't have any strong feelings on this, just wanted to know for sure!