[libcxx] [test] Fix experimental/memory.resource.adaptor.mem/db_deallocate on Windows

[mstorsjo]

Summary

The checks within the libc++experimental memory_resource class uses this
limit:

_MaxAlign = _LIBCPP_ALIGNOF(max_align_t);

Therefore, only use max_align_t for this limit instead of using
__STDCPP_DEFAULT_NEW_ALIGNMENT__ if available.

Alternatively, we could make _MaxAlign in memory_resource be defined
using __STDCPP_DEFAULT_NEW_ALIGNMENT__ if available.

[mstorsjo]

Rebase past CI config changes

[• Quuxplusone]

Looking at the git history of <experimental/memory_resource>, I can see that the calculation has always been

    static const size_t _MaxAlign = alignof(max_align_t);
[...]
    size_t __max_size() const _NOEXCEPT {
        return numeric_limits<size_t>::max() - _MaxAlign;
    }

Obviously __STDCPP_DEFAULT_NEW_ALIGNMENT__ has never been relevant to this behavior, and shouldn't have been mentioned in its tests.

This particular bogus test code was introduced by @joerg in D73245. It might be worth (either @joerg or @mstorsjo) auditing all other places where D73245 changed test code, to see whether they're also referring to __STDCPP_DEFAULT_NEW_ALIGNMENT__ in counterfactual ways. Alternatively, maybe resource_adaptor::_MaxAlign should be updated to actually care about __STDCPP_DEFAULT_NEW_ALIGNMENT__, in which case the test would become correct.

Requesting "changes" in the form of investigation/discussion. If the result of investigation ends up being "yeah, _MaxAlign should continue not referring to __STDCPP_DEFAULT_NEW_ALIGNMENT__, and this is the only affected test case," then I'd happily approve this PR in its current form, since it's just bringing the test suite into line with libc++'s actual behavior.

[mstorsjo]

@ldionne - What’s your opinion on this one?

[ldionne]

In D105905#2880877, @mstorsjo wrote:

@ldionne - What’s your opinion on this one?

std::pmr::memory_resource is clearly spec'd to use std::max_align_t, so I agree this is the intended behavior (http://eel.is/c++draft/mem.res#class.general-1).

I audited D73245, and the following changes don't look right to me in retrospect:

libcxx/test/std/containers/sequences/array/size_and_alignment.pass.cpp (I don't see what __STDCPP_DEFAULT_NEW_ALIGNMENT__ has to do with the alignment in std::array)
libcxx/test/std/language.support/support.types/max_align_t.pass.cpp (I think we should remove `#ifdef __STDCPP_DEFAULT_NEW_ALIGNMENT__`)

What do you both think?

[mstorsjo]

In D105905#2883789, @ldionne wrote:

In D105905#2880877, @mstorsjo wrote:

@ldionne - What’s your opinion on this one?

std::pmr::memory_resource is clearly spec'd to use std::max_align_t, so I agree this is the intended behavior (http://eel.is/c++draft/mem.res#class.general-1).

Thanks!

I audited D73245, and the following changes don't look right to me in retrospect:

libcxx/test/std/containers/sequences/array/size_and_alignment.pass.cpp (I don't see what __STDCPP_DEFAULT_NEW_ALIGNMENT__ has to do with the alignment in std::array)
libcxx/test/std/language.support/support.types/max_align_t.pass.cpp (I think we should remove `#ifdef __STDCPP_DEFAULT_NEW_ALIGNMENT__`)

What do you both think?

Can someone else than me pick up moving forward with those - separately from this patch? I presume this particular patch could be OK'd then?

libcxx/test/libcxx/experimental/memory/memory.resource.adaptor/memory.resource.adaptor.mem/db_deallocate.pass.cpp

 //===----------------------------------------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 // UNSUPPORTED: c++03
 
 // <experimental/memory_resource>
 
 // template <class T> class polymorphic_allocator
 
 // T* polymorphic_allocator<T>::deallocate(T*, size_t size)
 
-// XFAIL: LIBCXX-WINDOWS-FIXME
-
 int AssertCount = 0;
 
 #define _LIBCPP_ASSERT(x, m) ((x) ? (void)0 : (void)::AssertCount++)
 #define _LIBCPP_DEBUG 0
 #include <experimental/memory_resource>
 #include <type_traits>
 #include <cassert>
 
 #include "test_memory_resource.h"
 
 #include "test_macros.h"
 
 namespace ex = std::experimental::pmr;
 
 int main(int, char**)
 {
     using Alloc = NullAllocator<char>;
 
     AllocController P;
     ex::resource_adaptor<Alloc> r(Alloc{P});
     ex::memory_resource & m1 = r;
 
-#ifdef __STDCPP_DEFAULT_NEW_ALIGNMENT__
-    std::size_t maxSize = std::numeric_limits<std::size_t>::max()
-                            - __STDCPP_DEFAULT_NEW_ALIGNMENT__;
-#else
     std::size_t maxSize = std::numeric_limits<std::size_t>::max()
                             - alignof(std::max_align_t);
-#endif
 
     m1.deallocate(nullptr, maxSize);
     assert(AssertCount == 0);
     m1.deallocate(nullptr, maxSize + 1);
     assert(AssertCount >= 1);
 
   return 0;
 }