• constexpr -> _LIBCPP_CONSTEXPR_AFTER_CXX11 in front of invoke

• Make all constructors and members in callable_types constexpr.
• Make invoke use __invoke_constexpr after C++17.
• Update bind_front.pass.cpp to run all tests in a constexpr context as well.
• Remove inline prefix.

• Remove rejected files and files from incorrectly applied diff

• Rebase and address review comments
• Add _LIBCPP_INLINE_VISIBILITY

@ldionne thanks for your patience with this patch. It's great to finally get it in.

I think I originally created this as a precursor to D62274. But, I don't think it's needed anymore.

Resolved by 8aa2266fd8dd237c84daf12c2eec0df244558f49.

Not sure if this is the right patch or if it would be better as a follow-up, but there are a lot of places (vector, unique_ptr, valarray, etc.) where rvalues (or in the case of unique_ptr, lvalue references) are gated on _LIBCPP_CXX03_LANG rather than _LIBCPP_HAS_NO_RVALUE_REFERENCES. We should probably try to remove those workarounds as well.

• Remove "has builtin" check in __move.
• Replace #if in test with UNSUPPORTED.

Hopefully, I can land D65721 sometime this week, and then you can rebase this patch on master. Otherwise, feel free to take anything you need from that patch.

@rjmccall I updated this patch to introduce the function getCXXDestructorImplicitParam instead. This should be closer to D79942. It seems like there is never more than one implicit parameter so, I just have it return a single llvm::Value*.

• Remove emitCXXDestructorCall and add getCXXDestructorImplicitParam.

• Check all MemoryAccesses (fix segfault when one block is noop but another uses the store).

All tests now pass. dse.NumNoopStores still hasn't changed.

Does it pass without the patch? If it only fails with the patch, the patch is likely introducing a mis-compile.

Also, MultiSource/Applications/SPASS/SPASS.test is failing for some reason. The error it's giving me is:

/Users/zoe/Developer/llvm-source/stats-build/tools/fpcmp-target: files differ without tolerance allowance

It looks like this may be an issue with the time it took to execute the test? That may be because I'm running these tests on a laptop (so both compile time and execution time aren't going to be accurate metrics).

After running this on the LLVM test suite here are the stats:

Ping. This is a NFC.

Ping.

(You still need one of the libc++ maintainers to sign off on this)

Thanks for making this patch. Other than the noexcept issue, this LGTM.

Friendly ping.

• Mark as complete in cxx1z status

In C++03 __compressed_pair doesn't have an implicit move constructor. So, I disabled the tests in C++03.

• Rebase off master.
• Dissable in C++03 mode.

If we don't have proper support for array types in weak_ptr, we should add it separately (with tests too), not as part of fixing this simple LWG issue.

Do we have tests that shared_ptr isn't constructible from incompatible array types? If not, could we add some?

• Fix line wrapping

• Update status.
• Use TEST_STD_VER instead of the non-existant STD_TEST_VERSION.
• Use std::conditional (instead of logical and/or).

Closing this based on the discussion on the mailing list.

• Update 1z satus

I'm going to add a more comprehensive patch to fully support array types in weak_ptr later this week.

Friendly ping.

I am still having trouble to unterstand what benefit this has.

This passes all memory tests in C++03, C++11, and C++2a mode.

Oh right, I meant flipping the cl::init value of the variable https://github.com/llvm/llvm-project/blob/master/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp#L100 and rebuilding ;)

@fhahn I really appreciate all your help with this patch. The tests are updated.

• Update tests based on review

I'm not sure what huge benefit that would yield.

@ldionne I disagree. I think we can have all the things here :)

What if we did this in three patches?

1. Update everything to use data() instead of __elems_.
2. Use __array_storage in both implementations.
3. Remove the array<T, 0> specialization.

Might be interesting to see what happens with MSSA disabled for EarlyCSE.

Resolved by ca5bff18fc63d4f6b7fa2731d7a0a839efc984d8.

Resolved by c98648a1759b453a2ae019f4fcc5e38128c4cd4a.

This LGTM!

Thanks for fixing the nullptr thing, and thanks for all the work to improve tests here. Other than my one comment, this LGTM.

Also, it looks like the test suit does not have any noop stores (either that or I implemented the NumNoopStores statistic incorrectly).

• Rebase off master.
• Continue instead of break.
• Add noop store statistic.
• Remove phi argument support

Thanks for the review.