[omega_rythm]

Why not reverse the argument? If llvm can already be used for that sort of task, why not simply use it in place of Gcc for that very precise task? I'm pretty sure RMS would not mind it as much as opening gcc's AST, and he would certainly embrace the idea of making Gcc able to match llvm feature wise to make sure that whatever becomes possible in emacs is handled at compilation by Gcc.

[jsnell]

Yes, you'd think that, wouldn't you... But the only reason for this discussion is that RMS had already prevented the inclusion of clang-based tools in the Emacs code base. He thinks it'd undermine GCC, and by extension his cause.

[omega_rythm]

Hmmm, Charybde and Scylla. There's no solution I suppose, or maybe just go by what RMS wants: produce a data structure for the necessary data for the simplest possible improvement, then define a more involved data structure for a more ambitious feature, and reiterate again and again, until the data structure becomes isomorphic to what Gcc deals with. No need to actually code all these steps, just specifying them should help RMS understand that it's inevitable at some point.