On Thu, Mar 05, 2026 at 06:18:13PM -0500, Jason Merrill wrote: > On 3/5/26 5:35 PM, Marek Polacek wrote: > > On Thu, Mar 05, 2026 at 05:13:32PM -0500, Jason Merrill wrote: > > > On 3/5/26 2:07 PM, Marek Polacek wrote: > > > > On Wed, Mar 04, 2026 at 07:13:33PM +0000, Boris Staletic wrote: > > > > > Bootstrapped and tested on x86_64-pc-linux-gnu. > > > > > I have only ran g++.dg/reflect/* tests. > > > > > If needed/wanted, I can run the entire `make check`, though > > > > > that will take me the rest of the night. > > > > > -- >8 -- > > > > > With an annotation like [[=func]] gcc ends up calling > > > > > handle_annotation_attribute with TREE_VALUE (args) being a > > > > > VIEW_CONVERT_EXPR, whose type is a FUNCTION_TYPE. > > > > > That kind of annotation value should be decayed before checking > > > > > if the type is structural. > > > > > > > > > > Signed-off-by: Boris Staletic <[email protected]> > > > > > > > > > > PR c++/124177 > > > > > > > > > > gcc/cp/ChangeLog: > > > > > > > > > > * tree.cc (handle_annotation_attribute): Decay annotation > > > > > values. > > > > > > > > This can fit on one line (80 is the limit). > > > > > > I prefer to wrap at 76 in commit messages because git log adds 4 spaces at > > > the beginning. > > > > Fair. > > > > > > > gcc/testsuite/ChangeLog: > > > > > > > > > > * g++.dg/reflect/annotations11.C: New test. > > > > > --- > > > > > gcc/cp/tree.cc | 1 + > > > > > gcc/testsuite/g++.dg/reflect/annotations11.C | 17 +++++++++++++++++ > > > > > 2 files changed, 18 insertions(+) > > > > > create mode 100644 gcc/testsuite/g++.dg/reflect/annotations11.C > > > > > > > > > > diff --git a/gcc/cp/tree.cc b/gcc/cp/tree.cc > > > > > index 20288ed04e..5c7df0a1d9 100644 > > > > > --- a/gcc/cp/tree.cc > > > > > +++ b/gcc/cp/tree.cc > > > > > @@ -5912,6 +5912,7 @@ handle_annotation_attribute (tree *node, tree > > > > > ARG_UNUSED (name), > > > > > *no_add_attrs = true; > > > > > return NULL_TREE; > > > > > } > > > > > + TREE_VALUE (args) = decay_conversion (TREE_VALUE (args), > > > > > tf_warning_or_error); > > > > > > > > I'm not sure if it's reasonable to call decay_conversion on something > > > > type-dependent, so perhaps this should be moved inside of the following > > > > block. > > > > > > Good point. > > > > > > > But annotations are treated as late attributes so I'm not sure > > > > how we'd get here with anything type-dependent, and I couldn't construct > > > > such a test. Maybe the !t_d_e_p check could be dropped? > > > > > > Or turned into an assert. > > > > Sounds good. I can do it later. > > > > > > We're supposed to treat the expression as reflect_constant(arg) but > > > > doing the whole convert_reflect_constant_arg thing here seems > > > > unpleasant. > > > > > > Hmm, why unpleasant? It seems straightforward and might be needed to get > > > some messy cases right. > > > > I think it might need > > > > /* Figure out the type for reflect_constant. */ > > type = TREE_TYPE (convert_from_reference (r)); > > type = type_decays_to (type); > > type = cv_unqualified (type); > > convert_for_reference shouldn't be needed here.
Without it we don't compile this: constexpr int a = 3; constexpr const int &ref = a; static_assert (type_of (constant_of (^^ref)) == ^^int); due to error: 'const int&' must be a cv-unqualified structural type that is not a reference type > type_decays_to/cv_unqualified to correspond to the [type.deduct.call] > changes in maybe_adjust_types_for_deduction, sure. And then we wouldn't > need the decay_conversion after all. > > > for the type, then check !structural_type_p and copy constructibility, > > then call convert_reflect_constant_arg, and check for error_mark_node. > > Agreed. > > > But I think convert_reflect_constant_arg isn't entirely correct yet > > (FIXME in reflect_constant6.C). > > Perhaps, but that doesn't seem like an argument against using it here. I agree, it will probably work for all our annotation uses. Boris, let me know if you want me to take this over. Marek
