Re: Patch to dwarf inline handling

[Jakub Jelinek <jakub at redhat dot com>]

On Tue, Apr 25, 2000 at 05:34:58PM -0700, Jason Merrill wrote:
> The functions-as-trees changes in the C++ frontend have broken the old
> invariant that the out-of-line copy of an inline function (if any)
> would be emitted from output_inline_function, so set_decl_origin_self
> wasn't being called.  This patch moves that call into the dwarf
> backends, improving modularity.  It also limits the division of the
> debug info into abstract and concrete instances to actual inlines;
> previously we also did this for normal functions that weren't being
> emitted immediately for whatever reason.

This patch breaks inlined nested functions, at least on
i386-redhat-linux and sparc64-redhat-linux.
A simple test case can be e.g.:

int
foo (const char *response)
{
  inline int try (int match)
    {
      return match;
    }
  return (try (1) ?: try (0));
}

with -gdwarf-2 -O2 (taken from glibc).
The only abort() in dwarf2out_finish is triggered.
This is with current CVS.

> 2000-04-26  Jason Merrill  <jason@casey.cygnus.com>
> 
> 	* dwarf2out.c (add_bound_info): Don't crash on an unexpanded SAVE_EXPR.
> 
> 	* dwarfout.c (output_decl): Ignore NAMESPACE_DECLs.
> 
> 	* dwarf2out.c (gen_subprogram_die): The class-scope declaration DIE
> 	is the primary DIE for a member function.
> 	(gen_decl_die): Call set_decl_origin_self here.
> 	* dwarfout.c (output_decl): And here.
> 	* integrate.c (output_inline_function): Not here.
> 	Don't clear DECL_INLINE until after calling rest_of_compilation.
> 	(set_decl_origin_self): No longer static.
> 	* tree.h: Add prototype.
> 	* toplev.c (note_deferral_of_defined_inline_function): Only write
> 	out abstract instance for actual inlines.

	Jakub