On Mon, 25 Aug 2025, Jakub Jelinek wrote:
> Hi!
>
> The following testcase ICEs during expansion, because cfun->returns_struct
> wasn't cleared, despite auto being deduced to int.
>
> The problem is that check_return_type -> apply_deduced_return_type
> is called when parsing the expansion stmt body, at that time
> processing_template_decl is non-zero and apply_deduced_return_type
> in that case doesn't do the
> if (function *fun = DECL_STRUCT_FUNCTION (fco))
> {
> bool aggr = aggregate_value_p (result, fco);
> #ifdef PCC_STATIC_STRUCT_RETURN
> fun->returns_pcc_struct = aggr;
> #endif
> fun->returns_struct = aggr;
> }
> My assumption is that !processing_template_decl in that case
> is used in the sense "the fco function is not a function template",
> for function templates no reason to bother with fun->returns*struct,
> nothing will care about that.
> When returning a type dependent expression in the expansion stmt
> body, apply_deduced_return_type just won't be called during parsing,
> but when instantiating the body and all will be fine. But when
> returning a non-type-dependent expression, while check_return_type
> will be called again during instantiation of the body, as the return
> type is no longer auto in that case apply_deduced_return_type will not
> be called again and so nothing will fix up fun->returns*struct.
>
> The following patch fixes that by testing DECL_TEMPLATE_INFO, dunno
> what else would be more appropriate for this function is a template,
> maybe decl_dependent_p?
I was going to suggest just replacing the !processing_template_decl check
with !in_template_context to distinguish a true template context but that
wouldn't work here because we also push a dummy template parameter list
when parsing an expansion statement... so failing that, your current
approach makes sense to me.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> Or some other test?
>
> 2025-08-25 Jakub Jelinek <ja...@redhat.com>
>
> PR c++/121583
> * semantics.cc (apply_deduced_return_type): Adjust
> fun->returns*_struct even if processing_template_decl if
> in_expansion_stmt and fco is not a template.
>
> * g++.dg/cpp26/expansion-stmt23.C: New test.
> * g++.dg/cpp26/expansion-stmt24.C: New test.
>
> --- gcc/cp/semantics.cc.jj 2025-08-15 22:38:28.907872033 +0200
> +++ gcc/cp/semantics.cc 2025-08-23 18:02:09.130079505 +0200
> @@ -14213,7 +14213,8 @@ apply_deduced_return_type (tree fco, tre
> result);
> DECL_RESULT (fco) = result;
>
> - if (!processing_template_decl)
> + if (!processing_template_decl
> + || (in_expansion_stmt && !DECL_TEMPLATE_INFO (fco)))
> if (function *fun = DECL_STRUCT_FUNCTION (fco))
> {
> bool aggr = aggregate_value_p (result, fco);
> --- gcc/testsuite/g++.dg/cpp26/expansion-stmt23.C.jj 2025-08-23
> 17:49:13.816155626 +0200
> +++ gcc/testsuite/g++.dg/cpp26/expansion-stmt23.C 2025-08-23
> 18:08:32.125064559 +0200
> @@ -0,0 +1,17 @@
> +// PR c++/121583
> +// { dg-do compile { target c++14 } }
> +// { dg-options "" }
> +
> +auto
> +foo ()
> +{
> + template for (int i : { 0 }) // { dg-warning "'template for'
> only available with" "" { target c++23_down } }
> + return i;
> +}
> +
> +auto
> +bar ()
> +{
> + template for (auto i : { 0 }) // { dg-warning "'template for'
> only available with" "" { target c++23_down } }
> + return i;
> +}
> --- gcc/testsuite/g++.dg/cpp26/expansion-stmt24.C.jj 2025-08-23
> 18:08:46.988868108 +0200
> +++ gcc/testsuite/g++.dg/cpp26/expansion-stmt24.C 2025-08-23
> 18:12:46.856697841 +0200
> @@ -0,0 +1,10 @@
> +// PR c++/121583
> +// { dg-do compile { target c++14 } }
> +// { dg-options "" }
> +
> +auto
> +foo ()
> +{
> + template for (auto i : { 0, 1, 2LL }) // { dg-warning
> "'template for' only available with" "" { target c++23_down } }
> + return i; // { dg-error
> "inconsistent deduction for auto return type: 'int' and then 'long long int'"
> }
> +}
>
> Jakub
>
>
