Re: [PATCH] Fix debug info of nested inline functions

[Jason Merrill <jason at redhat dot com>]

On 09/10/2010 11:54 AM, Eric Botcazou wrote:
> We can test the flag on both DECLs to match the comment:
>
>    /* Emit an abstract instance of nested functions within an abstract instance
>       of their parent.  */
>    int declaration = ((decl != current_function_decl
> 		&&  !(DECL_INITIAL (decl) != NULL_TREE
> 			&&  DECL_ABSTRACT (decl)
> 			&&  current_function_decl
> 			&&  DECL_ABSTRACT (current_function_decl)))
> 		     || class_or_namespace_scope_p (context_die));

OK.  I would expect DECL_ABSTRACT (current_function_decl) to be set 
whenever DECL_ABSTRACT (decl) is set, but I guess it doesn't hurt to 
check both.

> > > +  /* Do not emit concrete instances of abstracted nested functions without
> > > +     actual instances.  */
> > > +  else if (TREE_CODE (decl_or_origin) == FUNCTION_DECL
> > > +       && ((!DECL_ABSTRACT (decl_or_origin)
> > > +        && function_possibly_abstracted_p (decl_or_origin))
> > > +           || (die && get_AT (die, DW_AT_inline))))
> > > +    ;
> >
> > This seems wrong.  We do want to emit concrete instances of abstracted
> > nested functions, just not within the abstract instance of the enclosing
> > function.
>
> This hunk is necessary to prevent abstracted nested functions from being
> emitted twice in the enclosing function if they are inlined there, as a
> concrete inline and as a concrete out-of-line instance.  We wait until after
> actual instances, either inline or out-of-line, are encountered and attach
> the DIEs in dwarf2out_finish.

OK, I see.

> > >   	if (origin)
> > > -	    add_child_die (origin->die_parent, die);
> > > +	    {
> > > +	      /* Find the first non-abstract parent instance.  */
> > > +	      do
> > > +		origin = origin->die_parent;
> > > +	      while (origin
> > > +		&&  (origin->die_tag != DW_TAG_subprogram
> > > +			 || get_AT (origin, DW_AT_inline)));
> > > +	      if (origin)
> > > +		add_child_die (origin, die);
> > > +	      else
> > > +		add_child_die (comp_unit_die, die);
> > > +	    }
> >
> > Here, shouldn't the parent of the concrete instance of the nested
> > function be the concrete instance of the enclosing function?
>
> Yes, the not always fulfilled 4. I mentioned in the message.  How would you do
> that?  If you look up the DIE of the enclosing function, you get the abstract
> instance.  The parent of concrete instances doesn't seem to really matter to
> GDB as long as it isn't abstract.

It seems like the later block that checks DECL_CONTEXT 
(node->created_for) should handle the case of a concrete non-inlined 
instance.

Jason