pip 19.0.2: IndexError when building wheel on osx

[gonzalocasas]

Environment

• pip version: 19.0.2
• Python version: 3.6
• OS: macos 10.13
• Platform: Travis-ci

The error was observed on travis-ci build's using Miniconda with an environment that contains both conda and pip packages.

Description
It seems release 19.0.2 breaks something related to wheel building under certain circumstances.
The problem started happening yesterday and it popped up exactly after 19.0.2 was released. The build was succeeding on 19.0.1, and as soon as 19.0.2 was available, the build started failing. See successful build #10 and failed build #11. After pinning pip back to 19.0.1, the build started working again.

I don't have a clear idea of what exactly fails, but it seems to be related to cvxpy and its dependency scs, but it is not clear to me if this is in any way connected to the root cause, or is only the visible side-effect.

How to Reproduce

1. Build using this travis file
2. If pip version is not pinned, and it uses 19.0.2, an error occurs (see below for tracebacks and full output).

Output
The full build output is available directly on travis: https://travis-ci.com/gramaziokohler/robotic_assembly_workshop/builds/100338039

  Stored in directory: /Users/travis/Library/Caches/pip/wheels/1d/2f/99/1594715f229f93f235d8b370a3b6f16e2b00fbac1087a99902
  Building wheel for scs (setup.py) ... done
Exception:
Traceback (most recent call last):
  File "/Users/travis/miniconda3/envs/workshop/lib/python3.6/site-packages/pip/_internal/cli/base_command.py", line 179, in main
    status = self.run(options, args)
  File "/Users/travis/miniconda3/envs/workshop/lib/python3.6/site-packages/pip/_internal/commands/install.py", line 355, in run
    session=session, autobuilding=True
  File "/Users/travis/miniconda3/envs/workshop/lib/python3.6/site-packages/pip/_internal/wheel.py", line 980, in build
    python_tag=python_tag,
  File "/Users/travis/miniconda3/envs/workshop/lib/python3.6/site-packages/pip/_internal/wheel.py", line 813, in _build_one
    python_tag=python_tag)
  File "/Users/travis/miniconda3/envs/workshop/lib/python3.6/site-packages/pip/_internal/wheel.py", line 821, in _build_one_inside_env
    wheel_path = builder(req, temp_dir.path, python_tag=python_tag)
  File "/Users/travis/miniconda3/envs/workshop/lib/python3.6/site-packages/pip/_internal/wheel.py", line 898, in _build_one_legacy
    return os.path.join(tempd, sorted(os.listdir(tempd))[0])
IndexError: list index out of range

[cjerdonek]

The line causing this exception was added in PR #6236.

[cjerdonek]

It looks like before, the build in question (of scs) was hitting this line in _build_one_inside_env():
https://github.com/pypa/pip/pull/6236/files#diff-79304b3f74b6bf37043da4d96ccd3755L831
and then causing _build_one_inside_env() to return None. The line in question looked like this, so you can see the exception was being swallowed--

except Exception:
    pass

Now, however, the exception is bubbling up without being handled because the code causing the IndexError was refactored out of that try-except block.

If you look at the Travis log for the successful case (the older one), you can see here: https://travis-ci.com/gramaziokohler/robotic_assembly_workshop/builds/100329545#L284
that it says scs indeed failed to build:

Building wheel for scs (setup.py) ... done
Running setup.py clean for scs
Building wheel for future (setup.py) ... done
Stored in directory: /Users/travis/Library/Caches/pip/wheels/0c/61/d2/d6b7317325828fbb39ee6ad559dbe4664d0896da4721bf379e
Successfully built compas-assembly compas-rbe cvxpy future
Failed to build scs

But this wasn't causing pip to error out before because of the exception swallowing I mentioned above.

[cjerdonek]

I posted a fix for this here: PR #6253.

[cjerdonek]

@gonzalocasas I don’t think it’s absolutely necessary, but would you be able to test with the version from the PR?

[gonzalocasas]

@cjerdonek thanks for the fix! it seems i was too slow to test and someone else already approved :)

thx!

[cjerdonek]

@gonzalocasas Thanks! You might still want to try as it should give you the reason for the failing build in the console output (which even 19.0.1 didn't do). I would be curious to see what the output looks like for you.

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.