pantsbuild #general

Channels

welcome

github-notifications

announce

enough-analyst-54434

10/07/2015, 5:32 PM

@ancient-stone-50795: do you have a python_requirement in your codebase for six? - please file an issue with all the details - the full trace would be helpful to see what pants commant leads here, etc

ancient-stone-50795

10/07/2015, 5:33 PM

Full_stacktrace.txt

ancient-stone-50795

10/07/2015, 5:34 PM

-.txt

ancient-stone-50795

10/07/2015, 5:34 PM

https://github.com/pantsbuild/pex/issues/167

ancient-stone-50795

10/07/2015, 5:40 PM

I just tried adding

six==1.9.0

to this requirements file and got some more details (updated the ticket)

enough-analyst-54434

10/07/2015, 5:44 PM

Yeah - I think the proof of the pudding is zapping your pantsbuild.pants.testinfra dep

enough-analyst-54434

10/07/2015, 5:44 PM

Then running all tests that don't need that

enough-analyst-54434

10/07/2015, 5:44 PM

Predict success

ancient-stone-50795

10/07/2015, 5:47 PM

I was able to successfully run a subset of the tests that did not depend on that requirement.

enough-analyst-54434

10/07/2015, 5:49 PM

So that's the quick gross fix, exclude those specs. The next fix is an emergency release today that unpins pants req to be >=1.9,<2.0 - can you test out such a change if I RB it?

enough-analyst-54434

10/07/2015, 5:49 PM

I can look at the actual resolve error after all this

ancient-stone-50795

10/07/2015, 5:49 PM

Yes, I can test it out.

enough-analyst-54434

10/07/2015, 5:50 PM

I want to underscore from the above that the pex does no dynamic resolution, its pants test.pytest creating a chroot for your tests that did the failing resolve

ancient-stone-50795

10/07/2015, 5:51 PM

OK, gotcha.

enough-analyst-54434

10/07/2015, 5:52 PM

Basically pants dogfooding itself for the backend its written in will always be more confusing in backtraces

enough-analyst-54434

10/07/2015, 5:52 PM

for python backend operations

ancient-stone-50795

10/07/2015, 5:55 PM

@enough-analyst-54434: thanks for the quick response.

ancient-stone-50795

10/07/2015, 5:57 PM

@strong-bird-83478: the dependency on six is declared:

Copy code

~/Src/Pants master find . -name req*.txt | xargs grep six
./3rdparty/python/requirements.txt:six==1.9.0

The six>=1.4 dependency is somewhere in unittest2 I think.

enough-analyst-54434

10/07/2015, 6:03 PM

Yes. The pants unittest2 used for the test runner fully floats: https://github.com/pantsbuild/pants/blob/master/src/python/pants/backend/python/tasks/pytest_run.py#L78 So grabbing latest:

Copy code

$ grep six unittest2-1.1.0/setup.py 
# six for compat helpers
REQUIRES = ['argparse', 'six>=1.4', 'traceback2'],

enough-analyst-54434

10/07/2015, 6:04 PM

@ancient-stone-50795: If you can patch in https://rbcommons.com/s/twitter/r/2942/, burn some sdists and verify you're all good I can prep a release

bored-art-40741

10/07/2015, 6:06 PM

Copy code

I want to underscore from the above that the pex does no dynamic resolution, its pants test.pytest creating a chroot for your tests that did the failing resolve

bored-art-40741

10/07/2015, 6:06 PM

But creating chroots is just having pex do a resolve and then unzipping that pex

ancient-stone-50795

10/07/2015, 6:06 PM

(I’m burning a new release now)

bored-art-40741

10/07/2015, 6:07 PM

These types of resolution issues are precisely why we got entirely off of pex

enough-analyst-54434

10/07/2015, 6:10 PM

There are 2 classes of fears, I wanted to dispel the FUD on the 1st

enough-analyst-54434

10/07/2015, 6:10 PM

once you have built a pex, its done resolving

enough-analyst-54434

10/07/2015, 6:10 PM

The resolver bug is real, may be in pants may be in pex, I'll be working that

ancient-stone-50795

10/07/2015, 6:10 PM

understood

bored-art-40741

10/07/2015, 6:11 PM

Ah okay

bored-art-40741

10/07/2015, 6:11 PM

The resolver bug is definitely in pex