While PBFs makes it very efficient to store repeated identical values (each unique string is stored only once), I was also thinking that we should remove those names that we had for historical reasons just for the sake of clarity and compactness. I don't even think we should have name_en -- we could have name_int (for latin-based international name), but all language-specific ones should be name:xx.

OK, I didn't know about this PBF optimization.
Yes, you're right, name_int is a better candidate for this. I can submit a PR if wanted.

This is a breaking change for schema 3.x - if merged and released - people who upgrade need to update their styles.

One of the things we wanted to do on 4.x schema change (in a semver style).

If to do that - then removing all "name_[language]" perhaps, not just "name_de" - as they are replaceable with "name:[language]" and name_int + name_latin.

I agree - we should remove name_en at the same time. What's more, I think we have some weird name:latin and name:nonlatin hack, which I think we should get rid of too -- name:xx should only be used by the real languages, and everything else like latinizated name should be using the underscore (i.e. name_latin).

Currently, the OMT-Tools add name:latin and name:nonlatin - see tileset.py and language.sql.

Can we add this to the milestone list for 4.0?