Detect address family when using --fd option #479
Conversation
InvalidInterrupt
force-pushed
the
issue-287
branch
2 times, most recently
from
1fdc1e9 to
56f87f9
Compare
Additionally, allow the caller to disable the default listener on 127.0.0.1.
By detecting the address family rather than assuming AF_INET, we allow unix domain sockets to be inherited properly.
Setuptools handled this fine but tox skips installing modules within the top-level package
InvalidInterrupt
force-pushed
the
issue-287
branch
from
56f87f9 to
50829d9
Compare
|
Thanks for this @InvalidInterrupt. Just FYI I'll be a cycle or two to get to this, but have a couple of issues to pull together for a release. |
|
A new utility package I added to the test dependencies seems not to advertise Python 3.7 support. |
Eliminate dependency on httpunixsocketconnection; it does not declare compatibility with Python 3.7. Patching an HTTPConnection object like this is hackish, but may be good enough for a test suite. I think it's pretty unlikely to lead to a false negative.
|
@InvalidInterrupt If you'd like, better would be to drop Python 3.7, which is EOL now. If you'd like to take that on (and related cleanups) in a separate PR that would be great, and speed me up 😅 |
|
I can make a PR to remove 3.7 support, but I'm not exactly sure what other changes I should (or should not) make. I'm inclined to keep the latest change regardless; I think this is less likely to result in trouble in the future. You can decide yourself once you have time to review it. |
|
OK, super thanks @InvalidInterrupt. I will get to this as soon as I can (but targeting a fall release for a new version) If you wanted to glance over the other open PRs and comment, that's handy too, but no stress. |
2 participants
I believe this fixes #287.
The changes I've made to tests and setuptools automatic package discovery may be a little messy still.