bpo-29704: Fix asyncio.SubprocessStreamProtocol closing

[sethmlarson]

Was told by @1st1 to open this PR here from python/asyncio. See the original Pull Request for discussion and review: python/asyncio#485

[mention-bot]

@SethMichaelLarson, thanks for your PR! By analyzing the history of the files in this pull request, we identified @Haypo and @1st1 to be potential reviewers.

[1st1]

LGTM. @asvetlov@fafhrd91 want to take a look too?

[1st1]

@SethMichaelLarson Do we have an open issue for this on bugs.python.org? If not can you create one?

[sethmlarson]

We do not, I will create one now.

[sethmlarson]

I created an issue with @BotoX comments and links to issue on python/asyncio http://bugs.python.org/issue29704

[1st1]

Please add a Misc/NEWS entry (to the top of Library section for 3.7.0 alpha)

[sethmlarson]

@1st1 Done, is that entry acceptable?

[1st1]

Yep. Looking good now.

[1st1]

This PR is merged, however, I didn't see that it had some trailing whitespace. Please make sure that your editor strips trailing whitespace.

[1st1]

Note: tested this PR with all uvloop functional subprocess tests, everything looks alright.

[sethmlarson]

@1st1 Sorry about that! My fault for not noticing. GitHub's web editor doesn't strip whitespace like PyCharm does. :) I'm glad it also works in uvloop, I should probably test asyncio changes with uvloop as well first before submitting.

[1st1]

I'm glad it also works in uvloop, I should probably test asyncio changes with uvloop as well first before submitting.

It's not so much about uvloop as it is about some unittests there that test both uvloop & vanilla asyncio. I plan to port some of those tests to CPython to increase coverage.

[1st1]

Anyways, thanks for the contribution!

[sethmlarson]

@1st1 Would be a good idea to port them, I've got some tests for selectors2 that don't occur in Python's stdlib selectors that might be considered good as well for unique situations.

[1st1]

I've got some tests for selectors2 that don't occur in Python's stdlib selectors that might be considered good as well for unique situations.

Please open a PR with those tests, selectors module is very important. For an extra benefit, you'll see your tests running on the full CPython buildbots matrix.

[sethmlarson]

Will do, might have to experiment with different Python versions, PEP 475 really split the behavior for anything syscall-related in Python.

[sethmlarson]

What is the protocol for tests that behave differently (skip?) on different Python versions? Would they only be put in the version where they should be or is it normal to have tests that only run on Python 3.5+, for example.

[vstinner]

Will do, might have to experiment with different Python versions, PEP 475

really split the behavior for anything syscall-related in Python. I lost track of asyncio. PEP 475 was implemented in Python 3.5. Python 3.4 doesn't accept bugfixes anymore. Why do you want to test different Python versions? Is asyncio on Python 3.3 or 3.4 still a thing?

[sethmlarson]

@Haypo Ah 3.4 is off bug-fix. Makes that a lot easier! :) Disregard my above comment then.
The tests would have to do with selectors.

[vstinner]

We try to keep this table up to date: https://docs.python.org/devguide/#status-of-python-branches

[fafhrd91]

@Haypo some rasberypi dev still on 3.4.2
https://home-assistant.io this is still on 3.4.2
I had to ship backported cookies module with aiohttp.

[vstinner]

@Haypo some rasberypi dev still on 3.4.2 https://home-assistant.io this is still on 3.4.2 I had to ship backported cookies module with aiohttp.

You are on the CPython project. There is also a http://github.com/python/asyncio project because there is no more release on http://pypi.python.org/pypi/asyncio/ since 2 years. I understand that asyncio doesn't evolve anymore on Python 3.4. Upgrade or backport yourself, no?