This will exclude any other formats -- text, or any custom/future one.
Do you know why something like not gettext doesn't work?

I tried not gettext and it did not work unfortunately. (Previous commit)

Is it because not isn't supported, or because gettext isn't the right tag?

Is it because not isn't supported, or because gettext isn't the right tag?

.. only:: gettext works as expected, so it must be the first.

The only directive doesn't seem to support not logical operator, according to the docs and the source code.

not is supported. This doesn't feel like the right solution, though.

That's odd, it must be bugged then, not gettext did not work.

This doesn't feel like the right solution, though.

I second to that. The grammar contains many comments, which definitely would be helpful if translated. In my opinion we should split the grammar include for gettext builder by every double newline character. This could be controlled as an option in the literalinclude directive (or a default)? This would make the translatable units sections smaller.

IMO even if we do it should not be prioritised. If someone is reading the grammar, they should probably know English.

There are a few hundred lines, we won’t translate error messages or var names anyway, is there really a point?

I see it being similar to code snippets – we should translate only comments here. As we don't have a way to parse what's comment and what's not, to allow the comments translation we should allow translation of the whole grammar content. I agree it's not a priority from whole translation perspective.

Note that many comments here mention features that this version of the Grammar omits.

IMO, we should copy any relevant comments to the preceding note, or preceding chapters (many of them are already there!), and then remove the comments altogether.