On Tue, Feb 08, 2022 at 02:20:28PM -0500, Jason Merrill wrote:
> > > + return add_const_value_attribute (die, TYPE_MODE (type), rtl);
> > > + }
> > > + else
> > > + {
> > > + /* For early_dwarf force mangling of all referenced symbols. */
> > > + tree initializer = init;
> > > + STRIP_NOPS (initializer);
> > > + /* rtl_for_decl_init punts on other aggregates, and complex
> > > values. */
> > > + if (AGGREGATE_TYPE_P (type)
> > > + || (TREE_CODE (initializer) == VIEW_CONVERT_EXPR
> > > + && AGGREGATE_TYPE_P (TREE_TYPE (TREE_OPERAND (initializer, 0))))
> > > + || TREE_CODE (type) == COMPLEX_TYPE)
> > > + ;
>
> Hmm, I don't think we need to mirror the limitations of rtl_for_decl_init
> here; IMO we might as well go ahead mangle everything so we don't need to
> change this place as well when the FIXME is resolved. OK either way, but if
> you decide to keep this condition please update the FIXME in
> rtl_for_decl_init to point here.
I did that because the aggregate case can mean very large initializers that
would need mangling a lot of symbols even when we actually don't need that
because we'd punt on it, and because I'm a little bit afraid how safe the
mangling is for -fcompare-debug if done only for the -g case (I vaguely
remember it can try to instantiate some templates etc.).
Can certainly adjust the FIXME comment.
Jakub
