The commit log (http://git.savannah.gnu.org/cgit/groff.git/commit/?id=f486938c5) says:
> Spell string translations using groff special character escape > sequences instead of Latin-1 or Latin-9 code points; this way they > work with a document that uses them no matter what its own encoding. > > I didn't do "ru.tmac"; that one's more of a pickle. The goal identified > above could be achieved by sifting the string translations through > preconv(1) and committing that, but that would come at the cost of > rendering them unintelligible to humans (and therefore prone to error). Spitballing an idea: leave ru.tmac as-is in git. Make the build run it through preconv and prepend a header identifying it as so modified, and pointing users to the original file. The processed file would be installed as ru.tmac, so that groff gets the benefit of its encoding-agnosticism, while the original would be installed as, say, ru.tmac.orig, so humans get the benefit of being able to read it. This would mildly clutter a system's tmac directory, which would be offset by improved usability.
