Since UTF-8 support is "in the workings", and number of Serbian man-pages low and not likely to surge anytime soon, I think it's reasonable to presently ignore the issue of possibility of sr_RS@latin; when UTF-8 is ready, a single "sr" entry will anyway cover all variants.

Off now to hunt for and warn about non-ISO-8859-5/2 characters in present and future man-pages... Thanks for the thorough resolution of proposed changes!

(As for intl/ and gnulib/ parts of the patch, the maintainer explained to me that the changes are only for obsolete systems, so there is no point in committing untested code in most likely never again used sections.)

I've applied the remainder of this patch; thanks! However, what happens if somebody creates /usr/share/man/sr_RS@latin/, which as far as I can see ought to be perfectly valid? Since the functions that iterate over directory_table only match prefixes, this won't work. Perhaps you could expand them to do some more intelligent matching, maybe with the aid of unpack_locale_bits?

There are few translated man pages in Serbian as of now, so some issues with our parallel use of Cyrillic and Latin scripts went unnoticed so far. Patch #6658 was about some technical corrections needed for dual-script languages (except Serbian, also e.g. Uzbek and Belarusian), and this one tries to disentangle Serbian-specific locale issues.

Some geopolitical notes. Current Serbian locales in GNU libc are sr_RS and sr_RS@latin, and these are likely to be stable. We went through few short-lived ones, like sr_CS, sr_CS@Latn (used by some distros but never in libc), sr_YU, etc. In general, no-modifer locale in Serbian should always be Cyrillic, and Latin should have @latin modifer.

There is one issue that the patch doesn't cover, as I couldn't figure it out. In recent times, especially in localization projects from which translated man pages normally originate, we use some characters which no 8-bit code page can handle (interpunction mostly: quotes, dashes...) So, if such a character is found in source man page (UTF-8 encoded), I get garbage in output. If I put UTF-8 in directory_table[] in encodings.cpp, still garbage; if I also add sr_RS to override roff_encoding to UTF-8 (like for CJK languages), still garbage.