Simplify vcs_support backend git by using GitPython

[agjohnson]

We already started using GitPython for some submodules pieces. We should continue porting some of our hand rolled logic to use GitPython instead.

Pieces of readthedocs.vcs_support.backend.git we should definitely port:

• parse_branches() -- this supersedes a branch I started at Add support for unicode Git tags/branches #2997 to replace parsing this with a csvreader
• parse_tags() -- same silly code using a csvreader
• find_ref()
• ref_exists()
• repo_exists()

Maybe:

• tags()
• branches()

These would be good first targets to port over, as we are executing these commands to get data out of the repository. It's not important that these messages are surfaced to users in build command output.

I'm going to block on getting a feature out that executes these commands in docker first though, as we need to clone and submodule checkout inside the docker container to isolate these calls. I can't quite consider how relying on gitpython for some of these calls, but not all, works with regard to docker vcs checkouts.

[humitos]

Just want to note here that I'm starting to feel uncomfortable with gitpython because it has several issues around unicode that cause our builds to fail and to debug them, write tests, etc.

I mentioned some of these problem at: #4318 (comment)

I'd like to start thinking for better alternatives. Although, I don't have one in mind yet.

[stsewd]

We are only missing git status and git submodule status here

[stsewd]

But not sure if we want to replace those, the code looks more complex with git python p:

-        code, _, _ = self.run('git', 'status', record=False)
-        return code == 0
+        try:
+            git.Repo(self.working_dir)
+        except InvalidGitRepositoryError:
+            return False
+        return True

-        code, out, _ = self.run('git', 'submodule', 'status', record=False)
-        return code == 0 and bool(out)
+        repo = git.Repo(self.working_dir)
+        return bool(repo.submodules)

[stsewd]

If you think that we are ok with git status and git submodule status, we can close this issue.

[humitos]

I like the consistency: all git command or all gitpython. A mixture of both it's hard to maintain and to keep in mind. On the other hand here, there are git commands that we can't replace: those with record=True

The code maybe looks longer, but it's more Pythonic, I'd say and we are relying on an external library instead of a local and custom command.

I'm 👍 on migrating them also.