Bug: the Sphinx extension translation_graph.py does not handle languages with different .po files sets

[flpm]

Currently the Sphinx extension that calculates the translation stats (./_ext/translation.graph.py) expects that all languages will have the exact same set of .po files. That is often the case, but not always.

Right now, adding a new language causes an error because the English guide no longer has a continuous-integration section: the previous languages all have the continuous-integration.po, but the new languages don't. This causes a np.Array error because the dimensions of the different rows of the table are not the same

Since deleting the .po could cause some translation work to be lost (if strings migrated to another section, they need to be migrated in the .po file), we are using a workaround to create a new language: copy an empty continuous-integration.po file from another language (e.g. ES) into the new language locale.

[sneakers-the-rat]

oops. ya should be easy enough to just pad with nans or 0s or something

[flpm]

I will be working on this one!

[flpm]

Thinking about this a bit more, I realize that is a little more complicated than we initially thought.

The current calculation counts the translated strings against the total number of strings in the .po file. This only produces the current translation percentage when the .po file is up to date with the English source.

The best way to ensure that the translation percentage is accurate even when the .po file is not up to date would be to compare the entries against the updated template files (.pot files) generated from the current English source. The .pot file looks exactly like a .po file but without any translation and they are used to update the .po files during nox -s update-language.

Right now, we create the .pot files inside _build/gettext, so we treat them as ephemeral files. This does not happen on every build so the extension cannot assume it will be there when computing the translation percentage.

One option would be to make sure they are always there by generating them on every build. This means running the 'gettext' Sphinx builder on every build before the Html builder runs. This would couple both builders, which is not ideal.

Another option would be to store the .pot files in locales/pot and check them into the repository. This would require to check them into the repository in CI, which is also not ideal either.

@sneakers-the-rat what are your thoughts?