Do not emit diagnostic with extended device lambdas with preserved re…

[Revaj]

Description

Corresponding from #1457

Allows extended lambdas with queryable return types for invoke_result_t

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[jrhemstad]

@miscco could you help @Revaj with adding a test that ensures the return type for a __device__ and __host__ __device__ lambda can be successfully queried with cuda::std::invoke_result_t iff it uses a trailing return type?

[Revaj]

I'll be adding a test like this

auto d_lm = [] __device__ () -> float { return 3.14f; };
auto dh_lm = [] __host__ __device__ () -> float { return 3.14f; };
using Td = decltype(d_lm);
using Tdh = decltype(dh_lm);

ASSERT_SAME_TYPE(cuda::std::is_same_v<cuda::std::invoke_result_t<Td>,
                                   float>);
ASSERT_SAME_TYPE(cuda::std::is_same_v<cuda::std::invoke_result_t<Tdh>,
                                   float>);

Is there a specific location where should be add it?

[miscco]

Is there a specific location where should be add it?

The tests are here libcudacxx/test/libcudacxx/cuda/proclaim_return_type.pass.cpp

[Revaj]

I noticed that std/utilities/meta/meta.trans/meta.trans.other/result_of.fail.cpp it supposed it should fail

FAIL: libcu++ :: std/utilities/meta/meta.trans/meta.trans.other/result_of.fail.cpp (18 of 2029)
******************** TEST 'libcu++ :: std/utilities/meta/meta.trans/meta.trans.other/result_of.fail.cpp' FAILED ********************
Command: ['/usr/local/cuda/bin/nvcc', '-o', '/dev/null', '-x', 'cu', '/home/coder/cccl/libcudacxx/test/libcudacxx/std/utilities/meta/meta.trans/meta.trans.other/result_of.fail.cpp', '-c', '-std=c++17', '-ftemplate-depth=270', '-ccbin=/usr/bin/g++', '-include', '/home/coder/cccl/libcudacxx/test/support/nasty_macros.h', '-I/home/coder/cccl/libcudacxx/include', '-D__STDC_FORMAT_MACROS', '-D__STDC_LIMIT_MACROS', '-D__STDC_CONSTANT_MACROS', '-Xcompiler', '-fno-exceptions', '-Xcompiler', '-fno-rtti', '-D_LIBCUDACXX_NO_RTTI', '-I/home/coder/cccl/libcudacxx/test/support', '-D_CCCL_NO_SYSTEM_HEADER', '-include', '/home/coder/cccl/libcudacxx/test/libcudacxx/force_include.h', '--compiler-options=-Wall', '--compiler-options=-Wextra', '--extended-lambda', '-gencode=arch=compute_80,code=sm_80', '-c']
Exit Code: 0

Expected compilation to fail!

Probably related to the PR since the compiler diagnostic was updated. Should I update the test to expect to pass instead?

[miscco]

Probably related to the PR since the compiler diagnostic was updated. Should I update the test to expect to pass instead?

We should keep that test as failing, as it does fail in older nvcc versions that do not support the trailing return type

You can add a // UNSUPPORTED: nvcc-12.X && nvcc-12.Y line where X would be the first CTK version that would pass

[Revaj]

Test added in proclaim_return_type.pass.cpp

[jrhemstad]

/ok to test

[Revaj]

I updated the test because the support of CUDA 12.3 below and std version.

[gonidelis]

@Revaj please rebase so we oktotest it :)

[Revaj]

@Revaj please rebase so we oktotest it :)

Done 😀

[jrhemstad]

/ok to test

[jrhemstad]

/ok to test

[Revaj]

For some reason, one of the tests failed from CI

The test

~/cccl/libcudacxx/test/libcudacxx/std/utilities/time/time.traits/time.traits.duration_values/max.pass.cpp

[Revaj]

I merged by error the branch, I'm going to fix it when I can :(

[jrhemstad]

/ok to test

[Revaj]

I have fixed the merge error in the branch so now it should be ok

[jrhemstad]

/ok to test

[jrhemstad]

The windows CI failures look unrelated. Rerunning those jobs.

[Revaj]

It seems the tests passed and it's ready to merge 😃

[jrhemstad]

Thanks @Revaj !