[PATCH v1 4/4] libstdc++: Add tests for std::extents.

[Luc Grosheintz]

>> diff --git a/libstdc++-v3/testsuite/23_containers/mdspan/extents/assign_copy_01_neg.cc b/libstdc++-v3/testsuite/23_containers/mdspan/extents/assign_copy_01_neg.cc
>> new file mode 100644
>> index 00000000000..8a514f2207b
>> --- /dev/null
>> +++ b/libstdc++-v3/testsuite/23_containers/mdspan/extents/assign_copy_01_neg.cc
>> @@ -0,0 +1,15 @@
>> +// { dg-do compile { target c++23 } }
>> +#include <mdspan>
>> +
>> +constexpr auto dyn = std::dynamic_extent;
>> +
>> +void
>> +test_assigment()
>> +{
>> +  auto e1 = std::extents<int, 1, 2>();
>> +  auto e2 = std::extents<int, 1, dyn>();
>> +
>> +  e1 = e2;
>> +}
>> +
>> +// { dg-error "no match for 'operator=" "" { target *-*-* } 12 }
> 
> 
> Could most of these negative *_neg.cc tests be replaced by checks like
> static_assert( ! std::is_copy_assignable_v<...> ) in the corresponding
> positive test, e.g. assign_copy.cc in this case?
> 
> Every separate test file has a small but measurable startup cost for
> the testsuite, and we already have 10k test files!
> If we can verify the same behaviour using concepts or traits, we don't
> need a separate neg.cc test.
> Sometimes we really do want a separate test, because we want to check
> that there's a specific error message, or we want to check for a
> static_assert inside a function body which can't be checked using
> concepts and so really does need to use dg-error, but I don't think
> that appies here. The string you're matching is just the generic "no
> match" error from the compiler.
> 
> 
Yes, I'll restructure the tests as proposed.