runtests.py still uses typing.py from lib-typing/3.2

[gvanrossum]

@ambv: I think some of your recent work on runtests.py may have stopped running some tests.

In particular, PR #2634 introduced something that doesn't work in Python 3.5.1 (from typing import Type -- the 3.5.1 stdlib version of typing.py doesn't define Type). There's a line in .travis.yml to run the tests with Python 3.5.1 specifically to catch this issue (the reason is our internal CI uses 3.5.1 and can't easily be updated). But the tests didn't fail (so our internal CI failed). I've fixed this in #2669 but the issue remains -- why did the test not catch it?

My theory is that somehow runtests.py doesn't run mypy any more.

Here's confirmation: with Python 3.5.1 installed, if I add the line from typing import Type to mypy/main.py, then running python3 -m mypy -c pass instantly fails with an ImportError, but python3.5 runtest.py -x lint passes without errors.

I'm also suspicious that the running time has gone down so much -- from IIRC ~4 minutes to 1:30.

[JukkaL]

This needs to be resolved before the 0.4.7 release.

[gvanrossum]

OK, I debugged this some more and the issue is more prosaic. The tests are run with PYTHONPATH set to include lib-typing/3.2 which means that it always uses the latest typing.py. I'll see what I can do about this.

[gvanrossum]

FWIW I think this can wait until after the 0.4.7 release. It's a little hairy. Related: #2674

[ambv]

@gvanrossum The time difference is legit. It's an effect of:

• sharding the longest running task which took the entire length of the test run before so it's actually parallel (which on Travis is massive: 32 tests at the same time instead of 1)
• only running flake8 on Python 3.6
• poking lxml to release a wheel for 3.6 (https://bugs.launchpad.net/lxml/+bug/1654670) and @ddfisher to release a wheel for typed-ast; cutting down the environment setup time 8X

Let me know if there's anything else for me to fix here.

[gvanrossum]

Amazing work! I still wish we could use pytest for everything, or add a better syntax for running just one test case to runtests...

…

--Guido (mobile)

On Jan 11, 2017 6:39 PM, "Łukasz Langa" ***@***.***> wrote: @gvanrossum <https://github.com/gvanrossum> The time difference is legit. It's an effect of: - sharding the longest running task which took the entire length of the test run before so it's actually parallel (which on Travis is massive: 32 tests at the same time instead of 1) - only running flake8 on Python 3.6 - poking lxml to release a wheel for 3.6 (https://bugs.launchpad.net/ lxml/+bug/1654670) and @ddfisher <https://github.com/ddfisher> to release a wheel for typed-ast; cutting down the environment setup time 8X Let me know if there's anything else for me to fix here. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2674 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACwrMtG4j3Av-prHOpZGhWxUVtS8cmZHks5rRZJEgaJpZM4LgBsT> .

[ilevkivskyi]

This is outdated now, we don't have lib-typing anymore.