There's no need for a clang-only solution, you can handle enums just fine with only regular expressions as long as the names are sufficiently distinctive.

I don't want this library to dictate naming of enums all around. The whole idea of using text parser to parse C files is fragile. Clang delivers a proper solution. Hopefully gcc will catch up.

To be clear, GCC by design is never going to emit an IR-like thing, so this is a "break forever for GCC" solution. You could perhaps generate a simple .c file with 'enum <foo> x;' and compile it with -g and parse the resulting DWARF with readelf or the like. (Or possibly use CTF as the output from ctfdump is likely easier to post-process than any of the DWARF dumps). That would not be tied to any specific compiler.

Most enums use names distinctive enough to handle with a regex just like we do for #defines. For the few cases where that doesn't work, we can use awk to limit the search to the lines between enum { and };. Give me a few days to come up with a poc, and please approve D57915 in the meantime.

EDIT: perhaps you forget that mktables doesn't need to know the value of each constant, it just needs the name.

In D57865#1329350, @des wrote:

Most enums use names distinctive enough to handle with a regex just like we do for #defines. For the few cases where that doesn't work, we can use awk to limit the search to the lines between enum { and };. Give me a few days to come up with a poc, and please approve D57915 in the meantime.

EDIT: perhaps you forget that mktables doesn't need to know the value of each constant, it just needs the name.