Deduction failures in pointer interconvertibility trait tests #5730
Conversation
An upcoming MSVC version will fix a bug that caused us to accept some template argument deduction cases that should be deduction failures. Specifically, with a template parameter `R (C::*)` where `C` was explicity specified you can't deduce `R` from a pointer-to-member with a mismatched class type, even one that has an implicit conversion. The pointer interconvertibility trait tests included a few of these situations where a template argument was explicitly specified but a pointer-to-member of a base class type was provided.
|
Mirror of MSVC-PR-671469, @joemmett's compiler fix for DevCom-10963535 VSO-2573454 "[accepts invalid] Pointer-to-member of base class as template argument". Thanks! 😻 |
| ASSERT(!is_corresponding_member(&S5::v1, &S6::v2)); | ||
| ASSERT(!is_corresponding_member(&S5::v2, &S6::v1)); | ||
| ASSERT(!is_corresponding_member(&S5::v3, &S6::v3)); | ||
| ASSERT(!is_corresponding_member<NS, NS>(&NS::v1, &NS::w1)); |
There was a problem hiding this comment.
This assertion is directly translated from the example in the standard's [meta.member]. We might need to tell the committee that this example is problematic.
There was a problem hiding this comment.
The example does note that it fails, but not why it fails:
static_assert( is_corresponding_member<C, C>( &C::a, &C::b ) );
// Forces the use of class C, and fails.It's reasonable to assume that this would fail the static_assert, but it actually fails deduction.
There was a problem hiding this comment.
I guess it's better to use is_corresponding_member<C, C, int, int>( &C::a, &C::b ) to achieve implicit conversion. Correspondingly, it's perhaps better to modify the line in the test file to ASSERT(!is_corresponding_member<NS, NS, int, int>(&NS::v1, &NS::w1));.
StephanTLavavej
changed the title
|
I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed. (I forgot that this was being mirrored as part of the compiler fix but after checking with Jonathan I'll go ahead and mirror this as it can go in independently) |
|
Thanks for fixing our tests! 😻 🔨 🔧 |
4 participants
An upcoming MSVC version will fix a bug that caused us to accept some template argument deduction cases that should be deduction failures. Specifically, with a template parameter
R (C::*)whereCwas explicity specified you can't deduceRfrom a pointer-to-member with a mismatched class type, even one that has an implicit conversion.The pointer interconvertibility trait tests included a few of these situations where a template argument was explicitly specified but a pointer-to-member of a base class type was provided.