https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118791
Bug ID: 118791
Summary: declare variant messes up template instantiation with
rvalue arguments
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: sandra at gcc dot gnu.org
CC: burnus at gcc dot gnu.org, jakub at gcc dot gnu.org,
waffl3x at protonmail dot com
Target Milestone: ---
Created attachment 60417
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60417&action=edit
test case quux.C
Like PR118530, this was discovered by waffl3x and seems to be a problem with
the template instantiation done by omp_declare_variant_finalize_one.
$ x86_64-linux-gnu-g++ -fopenmp quux.C -S
quux.C: In instantiation of 'void rvalue_int_variant(T&&) [with T = int]':
quux.C:50:30: required by substitution of 'template<class T> void
rvalue_int(T&&) [with T = int&&]'
50 | #pragma omp declare variant (rvalue_int_variant) match
(implementation={vendor("gnu")})
| ^~~~~~~~~~~~~~~~~~
quux.C:78:20: required from here
78 | rvalue_int<int&&>(0);
| ~~~~~~~~~~~~~~~~~^~~
quux.C:46:37: error: static assertion failed
46 | static_assert(is_same<T, int &&>::value);
| ^~~~~
quux.C:46:37: note: 'is_same<int, int&&>::value' evaluates to false
quux.C: In instantiation of 'void rvalue_const_int_variant(T&&) [with T = const
int]':
quux.C:64:30: required by substitution of 'template<class T> void
rvalue_const_int(T&&) [with T = const int&&]'
64 | #pragma omp declare variant (rvalue_const_int_variant) match
(implementation={vendor("gnu")})
| ^~~~~~~~~~~~~~~~~~~~~~~~
quux.C:79:32: required from here
79 | rvalue_const_int<int const&&>(static_cast<int const&&>(0));
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
quux.C:60:42: error: static assertion failed
60 | static_assert(is_same<T, int const&&>::value);
| ^~~~~
quux.C:60:42: note: 'is_same<const int, const int&&>::value' evaluates to false
The template base functions rvalue_int and rvalue_const_int are being
explicitly instantiated with T "int &&" and "int const &&", respectively, but
the corresponding variants with "int" and "const int".
(FWIW, I'm by no means an expert on this, just filing a bug report so I can
mention it in an xfail in the test case.)
