Call mkdocs using -m

[stsewd]

There are some cases were rtd uses the old installed version of mkdocs
instead of the one installed by the user, like in
#5532 (comment)

Which gives an error because it's using a yaml file
supported only in new mkdocs versions.

This is similar to #5107

[stsewd]

This was added since 0.15 mkdocs/mkdocs@bcb88ab#diff-872c8dbf5130934d0765867735f24482 (2015). So I think we are safe to use it.

[humitos]

What's the problem behind this PR and what's the solution? How does -m workaround this to make it work?

[stsewd]

Is the description.

#5532 (comment) when a user uses system packages option, the call to mkdocs is incorrect/non-deterministic. Sometimes it calls to the mkdocs installled outside the virtualenv and sometimes to the mkdocs installed inside the virtualenv.

I'm guessing this also fixes the issue in the corporate site.

Calling using -m we make sure to always use the mkdocs version from the virtualenv

You can replicate this locally by setting the system packages option on a mkdocs project pinning to mkdocs >=1, some times it gives errors in the mkdocs.yml file and others not, trey building with cache and wipiing.

[stsewd]

More details are in the other linked PR #5107

[humitos]

when a user uses system packages option, the call to mkdocs is incorrect/non-deterministic. Sometimes it calls to the mkdocs installled outside the virtualenv and sometimes to the mkdocs installed inside the virtualenv.

Calling using -m we make sure to always use the mkdocs version from the virtualenv

👍

Please, add this explanation as a comment in the code so it's clear next time we read that part of the code.

[stsewd]

Well, we already use this form for all commands, except sphinx, it doesn't support -m

[humitos]

None of the other commands have a comment about the reason. Please, add the comment. It doesn't bother us at all and it will help all the contributors in the future.