On Mon, Nov 17, 2025 at 07:39:58PM +0100, Jakub Jelinek wrote:
> On Mon, Nov 17, 2025 at 09:39:13AM -0500, Marek Polacek wrote:
> > > --- gcc/cp/parser.cc.jj 2025-11-15 16:03:45.587512374 +0100
> > > +++ gcc/cp/parser.cc 2025-11-17 10:21:57.667975827 +0100
> > > @@ -18779,9 +18779,14 @@ cp_parser_conversion_type_id (cp_parser*
> > > error ("invalid use of %<auto%> in conversion operator");
> > > return error_mark_node;
> > > }
> > > - else if (template_parm_scope_p ())
> > > - warning (0, "use of %<auto%> in member template "
> > > - "conversion operator can never be deduced");
> > > + else
> > > + {
> >
> > I suppose auto_diagnostic_group doesn't belong here?
>
> I thought about that, but am not sure.
> The two diagnostics are for the same issue, sure, but sometimes it
> will be only one of those, sometimes the other one, sometimes both,
> sometimes none.
>
> Could be also
> if (!pedwarn (...)
> && template_parm_scope_p ())
> warning (0, ...);
> i.e. only emit the warning if pedwarn was silent.
Yeah, then we would definitely not need an auto_diagnostic_group. The
warning() here could be a nice inform() note, though. Anyway, I don't
want to complicate things here, what you have looks good (without
an auto_diagnostic_group), thanks.
> > > + pedwarn (input_location, OPT_Wpedantic,
> > > + "invalid use of %<auto%> in conversion operator");
> > > + if (template_parm_scope_p ())
> > > + warning (0, "use of %<auto%> in member template "
> > > + "conversion operator can never be deduced");
Marek
