On Mon, Oct 17, 2016 at 10:45 AM, Jakub Jelinek <ja...@redhat.com> wrote: > On Mon, Oct 17, 2016 at 10:36:48AM +0200, Richard Biener wrote: >> GCC would have to cease supporting the stabs variants or output them >> from the DWARF IL we have. I believe this is what LLVM does. >> (I specifically discouraged the AdaCore folks from writing a windows >> debug format within the current framework). >> >> Your langhook could have been sth like lang_hooks.annotate_die >> (dw_die_ref *, tree) >> with exposing the dwarf API and thus letting the FE amend a DIE as it likes >> (instead of just exposing three attributes). > > I think that would risk FEs would then repeat too much stuff that dwarf2out > could do for them, many attributes are used by multiple languages for the > same or similar purpose. E.g. the current langhooks for DW_AT_inline is used > multiple times, not just to add the attribute but also for other purposes > before it happens. Plus allowing other files to use the dwarf2out > infrastructure would be quite a lot of work.
Sure. But the current way of trying to put everything into trees somewhere and then make dwarf2out reverse-engineer stuff doesn't really scale. Yes, adding more langhooks and more debug hooks would be the current canonical way of addressing things in a better way. But then debug hooks would just become high-level entry into the DWARF machinery. > As for emitting other debug formats from DWARF, I guess that is reasonable, > but the question is whether anybody still cares about stabs enough to > convert it. I'd be happy to remove stabs support but IIRC we still have targets that default to stabs. Richard. > Jakub
