Translating the compiler

[Manishearth]

@jan_hudec

I really don’t like that ftl files are key-to-string rather than English-to-translation, because it makes keeping track of which translations need to be updated difficult, but in this case it could be taken advantage of. If the error codes doubled as the Fluent IDs, then they would have to uniquely identify the message, so the conversion would force fixing this

Nah, we wouldn't use error codes as ids here, we'd have a new id scheme, likely derived from the name of the diagnostic struct. There are plenty of messages without error codes. This would still be pretty automatic -- the diagnostic struct conversion is something we've wanted to do anyway, and it's really easy to piggyback robust id-to-translation support on top of that.

@jan_hudec

The harder part is integrating it with Fluent, because it requires IDs. Maybe the tool could simply generate them.

I also can’t imagine writing the documentation by writing an ID in the comment and the text somewhere else (or writing the text somewhere, using the full symbol name as the ID), so the English catalog would have to be generated forever. And to be honest, I think it would be more maintainable if it was the case for the compiler as well.

You're responding to outdated comments about rustdoc here, the plan is not to use Fluent there, but to keep using an id-based format with autogenerated ids. See Translating the stdlib docs. This is still somewhat like English-to-translation, but it uses id-to-translation under the hood to be tracked better.

@jan_hudec

The hard part is initially marking what needs to be translated and/or splitting it up to translation units and then setting up the process and tooling for maintaining the translations, and doing it so that it does not make things harder for the programmers working on the code.

The error index is already generated from a very straightforward format, extracting the strings isn't hard. I already wrote a hacky tool that does this without any integration into the rust build system.

---

None of the plans involve forcing programmers to "add ids and then add the text in another file", they all work via extracting an English .ftl file from source at build time, using the same proc macro.