On Mon, Aug 24, 2015 at 4:34 AM, Marek Polacek <pola...@redhat.com> wrote: > > This is hopefully the last attempt to fix the libgo breakage. This time > around I really think I got to the bottom of the problem. The issue is > that the Go FE defines the __builtin_trap call, but doesn't set the > TREE_THIS_VOLATILE flag on it. That's bad because then this call isn't > gimple_call_noreturn_p and the cgraph cleanup code can't do its job properly. > > Bootstrapped/regtested on x86_64-linux. Now, I don't know the Go patch > process here. Should I just wait for you Ian to commit the patch? > (Yeah, I think I don't need the ChangeLog entry.)
For files in gcc/go/gofrontend, you should wait for me to commit the path. This file is in gcc/go, not gcc/go/gofrontend, so you can commit the patch directly, with a ChangeLog entry. (I know it's confusing.) > 2015-08-24 Marek Polacek <pola...@redhat.com> > > PR tree-optimization/67284 > * go-gcc.cc (Gcc_backend::define_builtin): Add NORETURN_P default > argument. Set TREE_THIS_VOLATILE. > (Gcc_backend::Gcc_backend): Mark __builtin_trap as a noreturn call. > > diff --git gcc/go/go-gcc.cc gcc/go/go-gcc.cc > index 6f274fc..aabcd98 100644 > --- gcc/go/go-gcc.cc > +++ gcc/go/go-gcc.cc > @@ -498,7 +498,7 @@ class Gcc_backend : public Backend > private: > void > define_builtin(built_in_function bcode, const char* name, const char* > libname, > - tree fntype, bool const_p); > + tree fntype, bool const_p, bool noreturn_p = false); I would rather not see a default value for this parameter. I would rather see all the calls change to pass false. As the GCC coding conventions say, one should normally avoid default arguments (https://gcc.gnu.org/codingconventions.html#Default). This patch is OK with that change. Thanks. Ian
