fix: handle broken symbolic links that were listing as missing files

[Psycojoker]

Hello,

This MR is to fix a hard to debug situation.

Consider this situation:

• file "a" exists
• symbolic links "points-to-a" exists and points to "a"
• both the file and the symbolic links are added and committed into the code repository

Later on:

• file "a" is removed and a commit is made
• "check-manifest" will complains that "points-to-a" is a missing file
• but "points-to-a" is still present and appears if a "ls" is done on it
• therefor this can be very confusing and time consuming to debug

This MR changes the output from:

some files listed as being under source control are missing:
  points-to-a

To:

some symbolic links listed as being under source control are pointing to missing files:                                                                                                                                                                                                     
  to-qsd -> /home/psycojoker/code/python/check-manifest/qsd (missing)

I'm not sure this is the best output so feel free to modify it.

As a side note, this MR doesn't handle this situation for the "missing from sdist" analysis and it should probably be added as you can tell from this screenshot:

Thanks a lot for this tool ❤️

[mgedmin]

Looks mostly good. I'll want to come up with a regression test for this case, to test keep coverage at 100%.

[mgedmin]

Hmm, is it better to use os.path.realpath(x) here, or is it better to use os.readlink(x) here?

[Psycojoker]

I've switch to os.readlink to try it but the result ended up being the same?

Apparently it can returns relative path sometime according to the documentation but I don't know when. I guess this will be more readable in some situations?

[mgedmin]

I suppose the difference would be if you had two symlinks, like a -> b and b -> c. os.realpath('a') would return 'c', but os.readlink('a') would return 'b'.

But we know that in this case b doesn't exist, so it can't be a symlink to c.

How about symlinks to directories? a -> b/c, where b -> d, and d/c is missing. os.realpath('a') is 'd/c', and os.readlink('a') is 'b/c'. Which would be better to show in the error message?

To me readlink() feels a bit better.

[Psycojoker]

Looks mostly good. I'll want to come up with a regression test for this case, to test keep coverage at 100%.

I don't know how I managed to miss the test file 😅

I've added a test but I'm not fully sure I did it correctly and in the good section :/

[mgedmin]

Thank you very much for the test!

Weirdly I cannot reproduce the GitHub Actions CI failure locally (with SKIP_NO_TESTS=1 FORCE_TEST_VCS=svn tox -e py310).

For the Appveyor failures, see inline comments:

[mgedmin]

Seeing the Appveyor CI failure on Windows, I think

Suggested change

	os.symlink(os.path.join(self.tmpdir, "some-file"),
	os.path.join(self.tmpdir, "some-symbolic-link"))
	os.symlink(os.path.join(self.tmpdir, "some-file"), "some-symbolic-link")

might work better ...

[mgedmin]

... because then you might be able to use

Suggested change

	f" some-symbolic-link -> {os.path.join(self.tmpdir, 'some-file')} (missing)",
	" some-symbolic-link -> some-file (missing)",

here.

Of course, Windows being Windows, there's a chance that even relative symlinks will be converted into those weird absolute UNC paths, in which case plan B is

# On Windows os.readlink() returns weird UNC paths so we cannot hardcode os.path.join(self.tmpdir, "...") here
link_target = os.readlink(os.path.join(self.tmpdir, "some-symbolic-link"))

followed by using

Suggested change

	f" some-symbolic-link -> {os.path.join(self.tmpdir, 'some-file')} (missing)",
	f" some-symbolic-link -> {link_target} (missing)",

here.