PATCH Re: hitting add_abstract_origin_attribute abort()

[Jason Merrill]

>>>>> Mark Mitchell <mark@codesourcery.com> writes:

 > I'm going to look at the DWARF1 problem that Robert Lipe reported
 > yesterday; it may be caused by something similar,

Probably.

 > and perhaps something similar to your patch will help there as well.

Maybe.

 > I found the following to be somewhat odd.  In output_inline_function,
 > we call set_decl_origin_self, which makes the function its own
 > abstract origin.  We have the following comment:

 >   /* We must have already output DWARF debugging information for the
 >      original (abstract) inline function declaration/definition, so

 > I don't see why we must have already output debugging information for
 > the original function at this point; in looking at the DWARF1 problem
 > this seemed to me to be the issue.

We should be emitting the info from rest_of_compilation.  Of course, if we
never generated RTL for the function in question, we'd never get to
rest_of_compilation...I've changed dwarf2out.c to generate the abstract
instance as needed, but that sort of thing is harder with dwarfout.c,
because it emits everything immediately, while dwarf2out saves it up and
writes it all out at the end.

Jason