I won't include this in the current patch release, as I don't have enough bandwidth to avoid blocking the Jupyter Book release that needs to follow. I'll circle back around to this PR soon :)

@OriolAbril - Adding a test here would be really useful. I would a translated version of a test notebook should be sufficient.
I would think adding the locale in the tests folder should be fine as long as the test is being picked up and pytest doesn't start complaining about it.

I did my best to modify one of the tests in test_parser to use translations.

It seems different versions of sphinx generate different translation related metadata and the xml comparison fails. I don't know how to fix that

It seems different versions of sphinx generate different translation related metadata and the xml comparison fails. I don't know how to fix that

I have some open PRs that revisit the test requirements, as well as planning to do another cleanup of the versions in GHA. One those are done and this still runs into CI issue, we can revisit with some version dependent conditionals, or just state that translations support will require a newer sphinx than the stated minimum supported one.

The way the PR introduces to handle translations works in all versions. However, the html/xml metadata related to translations differs for the different versions. The xml generated with sphinx 8 which is the one I added to the tests folder is the following:

<document source="basic_run_intl" translation_progress="{'total': 4, 'translated': 2}">
    <section ids="a-title" names="a\ title un\ título">
        <title translated="True">
            un título
        <paragraph translated="True">
            algo de texto
        <container cell_index="1" cell_metadata="{}" classes="cell" exec_count="1" nb_element="cell_code">
            <container classes="cell_input" nb_element="cell_code_source">
                <literal_block language="ipython3" xml:space="preserve">
                    a=1
                    print(a)
            <container classes="cell_output" nb_element="cell_code_output">
                <literal_block classes="output stream" language="myst-ansi" xml:space="preserve">
                    1

In sphinx<8 the translated="True" in title and paragraph elements and translation_progress="{'total': 4, 'translated': 2}" in the document element are not there. All of them have the translated content properly injected and rendered though.

If it were possible to ignore this metadata when comparing against the xml output then tests would pass for all versions, and would still check that the actual translated content is injected together.

@OriolAbril - Experimenting and combing the code a little bit through I'm fairly certain that the logic is the opposite: we need the old sphinx-generated file for comparison and then ignore the extra bit of metadata in conftest. There are already a couple of similar cases.

As I was iterating on this locally, I've pushed the changes directly to your branch.

I think this is good to go now, but I'll leave this open so @agoose77 can double-check it, too.

Thanks @bsipocz!

I don't see any objections, so let's go ahead with the merge here. Thanks @OriolAbril for your patience.