Adds call to os.path.abspath() in pkg_resources.normalize_path() on Cygwin

[wildmichael]

This works around problems that stem from getcwd(3) on Cygwin returning paths containing symlinks. I am not sure at all whether this is a good place to fix it, but that's where I got hit by the issue when doing a python setup.py develop (or pip install -e .).

[pganssle]

What bug does this fix? I'm not a huge fan of adding platform-specific code that actually works around bugs on those platforms (as opposed to platform-specific code that is designed to deal with actual differences between platforms). It's particularly bad because it's kinda tough to test.

If this is going to be included, I think it needs a test, but I don't really want to add a new executor that runs everything on cygwin. I think something that mocks out sys.platform and some of the other components such that the bug would be triggered if the fix were not in place.

[wildmichael]

My problem was with invoking pip install -e . where the working directory path contains a symlink. This fails with a rather obscure message, essentially because the target and egg path are not identical due to the failed symlink resolution.

But I agree, it is an ugly, duct tape type patch. I'll see whether I can find a way to patch Python on Cygwin to behave as expected. The proper fix in Cygwin is probably more problematic since it could break all kinds of things...

[jaraco]

Unfortunately, I don't think even having a test case will help because we don't run any tests on cygwin. I also agree with Paul that we really should have a better understanding of the factors that lead to the failure and a clear understanding of why this fix patches it. At least a minimal example posted here would set that context and could be linked to the workaround.

Can you show output from a cygwin shell using only Python that illustrates the conditions that lead to the unexpected condition?

[wildmichael]

Below the transcript of a session that failed:

~ $ mkdir -p ~/tmp/somedir
~ $ cd ~/tmp/somedir
~/tmp/somedir $ mkdir a
~/tmp/somedir $ # make a proper symbolic NTFS directory link, requiring elevated privileges and stuff
~/tmp/somedir $ cygstart --action=runas cmd /c "mklink /d $(cygpath -w $PWD/b) $(cygpath -w $PWD/a)"
~/tmp/somedir $ # check that symlink
~/tmp/somedir $ ls -la
total 16
drwxr-xr-x+ 1 themiwi       Domain Users  0 May 15 20:57 .
drwx------+ 1 themiwi       Domain Users  0 May 15 20:35 ..
drwxr-xr-x+ 1 themiwi       Domain Users  0 May 15 20:46 a
lrwxrwxrwx  1 Administrators Domain Users 38 May 15 20:57 b -> /drives/c/Users/themiwi/tmp/somedir/a
~/tmp/somedir $ cd b/
~/tmp/somedir/b $ git clone https://github.com/pypa/sampleproject.git
Cloning into 'sampleproject'...
remote: Counting objects: 320, done.
remote: Compressing objects: 100% (12/12), done.
remote: Total 320 (delta 4), reused 11 (delta 3), pack-reused 305
Receiving objects: 100% (320/320), 70.48 KiB | 546.00 KiB/s, done.
Resolving deltas: 100% (156/156), done.
~/tmp/somedir/b $ # make a virtual env for cleanliness reasons...
~/tmp/somedir/b $ python3 -m venv ../.venv
~/tmp/somedir/b $ source ../.venv/bin/activate
(.venv) ~/tmp/somedir/b $ cd sampleproject
(.venv) ~/tmp/somedir/b $ # and this will go boom!
(.venv) ~/tmp/somedir/b/sampleproject $ pip install -e .
Obtaining file:///drives/c/Users/themiwi/tmp/somedir/b/sampleproject
Collecting peppercorn (from sampleproject==1.2.0)
  Using cached https://files.pythonhosted.org/packages/45/ec/a62ec317d1324a01567c5221b420742f094f05ee48097e5157d32be3755c/peppercorn-0.5.tar.gz
Installing collected packages: peppercorn, sampleproject
  Running setup.py install for peppercorn ... done
  Running setup.py develop for sampleproject
    Complete output from command /drives/c/Users/themiwi/tmp/somedir/.venv/bin/python3 -c "import setuptools, tokenize;__file__='/drives/c/Users/themiwi/tmp/somedir/b/sampleproject/setup.py';f=getattr(tokenize, 'open', open)(__file__);code=f.read().replace('\r\n', '\n');f.close();exec(compile(code, __file__, 'exec'))" develop --no-deps:
    running develop
    /usr/lib/python3.6/distutils/dist.py:261: UserWarning: Unknown distribution option: 'long_description_content_type'
      warnings.warn(msg)
    /usr/lib/python3.6/distutils/dist.py:261: UserWarning: Unknown distribution option: 'project_urls'
      warnings.warn(msg)
    error: --egg-path must be a relative path from the install directory to /drives/c/Users/themiwi/tmp/somedir/b/sampleproject

    ----------------------------------------
Command "/drives/c/Users/themiwi/tmp/somedir/.venv/bin/python3 -c "import setuptools, tokenize;__file__='/drives/c/Users/themiwi/tmp/somedir/b/sampleproject/setup.py';f=getattr(tokenize, 'open', open)(__file__);code=f.read().replace('\r\n', '\n');f.close();exec(compile(code, __file__, 'exec'))" develop --no-deps" failed with error code 1 in /drives/c/Users/themiwi/tmp/somedir/b/sampleproject/

The base problem seems to be that getcwd(3) does not resolve intermediate symlinks on Cygwin: https://www.cygwin.com/ml/cygwin/2000-10/msg00906.html. My patch just fixes this by calling os.path.abspath() to force this symlink resolution.

But as I already said, this is probably better fixed by patching the python3 package of Cygwin to properly deal with the getcwd(3) antics.

[jaraco]

I honestly don't understand why the codecov/patch is failing. It says 75% of diff hit, when actually in a line-by-line comparison, it would have to be hit. But more importantly, I've marked it as pragma: nocover, so it shouldn't be including that function in the coverage stats anyway.

So I'm going to merge this so I can get a release cut.

[pganssle]

It may be that we have codecov configured with branch coverage? Or possibly ternary expressions count as multiple lines? Unclear.