Drop python3.5 #1396
Drop python3.5 #1396
Conversation
Since we no longer support Python 3.5.
Since we no longer support Python 3.5.
Since we no longer support Python 3.5.
|
I was able to reproduce the issues locally with both 3.7-dev and 3.8-dev. It looks like the issues with the tests are all related to this error: I'm not sure how to proceed from here. I found this: Any pointers? Maybe here is a good place to start? |
I think that would be ideal. It could be like, merge the warning, release, and then immediately merge this PR within a 10 minute span, though. |
|
The pypy failure looks real. |
|
The correct bpo link for the new aclose/athrow thing is: https://bugs.python.org/issue39386 |
|
OK and I tracked down the |
|
Phew, it's lucky we dug into that |
|
That fix is merged upstream now, so sometime in the next day or so the Travis builds should update to include it, and the tests here should start passing.
I don't have an example, no, but I guess it should just require adding something like this to the bottom of import sys
if sys.version_info < (3, 6):
_deprecate.warn_deprecated("Support for Python 3.5", "0.14", issue=75, instead="Python 3.6+")
del sysIt might also need some fiddling with our pytest warnings filters to keep the tests passing. |
|
It looks like the pypy nightly branch is failing because of these 3.6 additions not being present: https://docs.python.org/3/whatsnew/3.6.html#socket Related: 4f3030f |
|
I filed a bug with PyPy about the missing socket constants: https://foss.heptapod.net/pypy/pypy/issues/3171 |
|
We should add a comment next to the |
|
I'm a bit puzzled by the latest CI failure.. |
It's a mysterious issue that happens occasionally for reasons unknown: #200 (comment) It doesn't seem to be related to this PR though, so we can just re-run the CI here. |
|
Now that v0.14.0 is out, including the 3.5 deprecation, I guess this is unblocked. (Sorry for the long delay there!) I think we can be relatively aggressive about dropping support here. For most deprecations, if folks haven't updated, then removing the code breaks them. But thanks to the magic of Python-Requires: metadata, in this case when we remove the code then folks who haven't updated just remain on the old version. |
|
Hey @njsmith, is there anything I can do to help here? I've somewhat lost track of where we left off :-) |
|
The idea is to fix the conflict and remove code that's no longer covered. I can take over from here if you want. |
|
Sure, feel free @pquentin!
|
|
Merged as part of #1481 |
…-18475) The fix for [bpo-39386](https://bugs.python.org/issue39386) attempted to make it so you couldn't reuse a agen.aclose() coroutine object. It accidentally also prevented you from calling aclose() at all on an async generator that was already closed or exhausted. This commit fixes it so we're only blocking the actually illegal cases, while allowing the legal cases. The new tests failed before this patch. Also confirmed that this fixes the test failures we were seeing in Trio with Python dev builds: python-trio/trio#1396 https://bugs.python.org/issue39606 (cherry picked from commit 925dc7f) Co-authored-by: Nathaniel J. Smith <njs@pobox.com>
The fix for bpo-39386 attempted to make it so you couldn't reuse a agen.aclose() coroutine object. It accidentally also prevented you from calling aclose() at all on an async generator that was already closed or exhausted. This commit fixes it so we're only blocking the actually illegal cases, while allowing the legal cases. The new tests failed before this patch. Also confirmed that this fixes the test failures we were seeing in Trio with Python dev builds: python-trio/trio#1396
Continues @pquentin's work from #1390
Closes #75