On Wed, 5 Feb 2014, Jan Hubicka wrote: > > > > > > Seems OK to me. I always had an impression that this machinery exists to > > > make decls that are revmoed fro symbol table (optimized out) to land the > > > debug info, but I think this is mostly broken by partitioning anyway, > > > right? > > > Do we stream them at all? > > > > No, we don't stream those. We'd need to stream what the FEs think > > are the globals (lang_hooks.getdecls()) and somehow partition them > > and ship them somewhere. > > My old idea was to simply output debug for dead stuff from WPA and add > it asone of object files to linking later. But indeed, this is for later > time.
Note that we don't even output (some of) the stuff at compile-time due to the way we walk trees we output (we don't walk lang_hooks.getdecls ()). But yes, that's the way early debug should work, too. Richard.
