On Wed, Feb 18, 2026 at 11:18:03AM +0100, Jakub Jelinek wrote:
> On Wed, Feb 18, 2026 at 05:50:09PM +0900, Jason Merrill wrote:
> > This seems like the sort of flag that we need to avoid passing down to
> > recursive overload resolution, like tf_decltype in tsubst_expr.
> >
> > I'm not sure that it's actually necessary in this case since you don't check
> > it in build_user_type_conversion_1, but I suspect it will get passed on to
> > places that do general expression substitution.
>
> Ok. In finish_call_expr with many spots passing complain around I've
> just filtered tf_any_viable out of complain and passed the original
> one only to the single specific case (clearly I was wrong when thinking
> perform_koenig_lookup could return a class object, so I've dropped the
> build_op_call case too).
> And in build_new_function_call I've filtered it out from resolve_args,
> the rest seems to be ok. Similarly in perform_overload_resolution
> I've filtered it out from add_candidates call.
>
> So far tested with
> GXX_TESTSUITE_STDS=98,11,14,17,20,23,26 make -j32 -k check-g++
> RUNTESTFLAGS="dg.exp='reflect/* expansion-stmt*'"
> Ok for trunk if it passes full bootstrap/regtest?
Bootstrapped/regtested successfully on both x86_64-linux and i686-linux.
Jakub
