There was some discussion of this last year in this review: https://reviews.llvm.org/D128750

In D146168#4231457, @aaron.ballman wrote:

LGTM aside from a minor nit, though I have some minor concerns that we may silently break people's code by calling different overloads or instantiating templates differently so I wonder if this is a potentially breaking change or not. I think we can leave it out of the potentially breaking changes section unless we get some reports about a behavior change organically during this release cycle?

fix unintended indentation & rebase

Add some ((packed)) tests

In D146329#4203174, @shafik wrote:

I would have loved to test the case from https://github.com/llvm/llvm-project/issues/61335 directly but I think in order to do it nicely I need __builtin_memset to be usable in a constant expression context. I will add this to my todo list. I am open to other alternatives for testing this.

Add the test case from the GH issue.

rebase + release notes.

slimmer codegen test

Oops, sorry for the bad codegen test and thanks for the comments.

Fix the codegen test, add a standard ref to the comment.

Add a test case, slightly simpler modeling.

In D145851#4189158, @erichkeane wrote:

Generally looks good to me. Do we do anything special if there are multiple initializers? Also, can we have a codegen test that validates that we actually construct it correctly (and perhaps a constexpr test for the same!)?

Add more tests, small nit.

Small cleanup

Handle 0-length unnamed bit fields as well.

Change to a simpler fix.

In D140996#4185795, @bolshakov-a wrote:

@royjacobson, I've added some test cases for using the new NTTP arguments in clang modules. It uses serialization, in principle. Or more specialized tests are still needed?

Thanks! Do you need me to land this for you? If so, could you please provide name + email to sign the commit with?

In D143851#4153270, @carlosgalvezp wrote:

LGTM, thanks for the contribution! Do you have commit rights or would you like that we land it for you? If so, please provide name and email for attribution.

There are no AST [de]serialization tests in this PR, right? Would be nice to add some.

friendly ping

In D144572#4145614, @cor3ntin wrote:

@aaron.ballman is the plan to backport that to clang 16? I think it probably should.
We ought to try to improve the situation for clang 17

+1 from me, at least as long as we don't define the feature test macro.

It's an interesting discussion, I just want to remind that this change is for triviality rules from P0848 (conditionally trivial member functions) which is only going to ship in Clang 16 anyway.

In D143851#4126453, @ClockMan wrote:

Thing is that same issue may happen with all other members, what about copy/move constructors defaulted in .cpp (just to speed up compilation for classes with many members).
Best thing would simply to check if all definitions are available. In that case issues would be reported only for source file that corresponds to header file with defined class.
This would also reduce amount of generated warnings for a big project that include such header with class in many places.

Other good option would be to excluded classes defined in system headers, no point to check if some boost class or std::vector got proper constructors.

Landed in https://github.com/llvm/llvm-project/commit/be701ab08f12daf9437c2db6d08a1731cf1df34c

Address CR comments.

In D143891#4122660, @aaron.ballman wrote:

This is an ABI breaking change, isn't it? (The type trait now returns something different than it did before, which could change instantiations or object layout.)

friendly ping :)

Rebase to include the RISC-V changes.

15 -> Clang 15

In D142578#4088011, @tschuett wrote:

IDK. Clang 16 is fully is a fully conformant C++20 except for some DRs. Or beginning with Clang 16, we start a long-term project to overhaul the diagnostics.

I agree that breaking changes are important.

• Remove redundant paragraph from the intro
• Remove outdated bug fix bullet

In D142578#4087974, @tschuett wrote:

I am not the only one who has a different background. If the introduction says this is the introduction of the release notes, I am not super motivated to read on. I would prefer to see some highlights to motivate the readers to continue reading.

Thanks Erich and Emilia!

• Address comments from Erich and Emilia
• Validate that the rst actually builds an html file
• Fix some missing backticks and underscores

fix the test after diag change

Reject non class types, update doc.

rebase on main

In D141358#4039268, @ilya-biryukov wrote:

In D141358#4039265, @royjacobson wrote:

It's dr2237, see https://github.com/llvm/llvm-project/issues/55164

Thanks for the pointers! Clang does not seem to diagnose or forbid this even in trunk. I suspect the DR is not yet implemented, but there are other changes to the parser that cause these errors? 'Down with typename' maybe?

It's dr2237, see https://github.com/llvm/llvm-project/issues/55164

Since people run into this very quickly and the Clang workaround is pretty simple, LGTM.

So, I had a quick look at this. Turns out this fix changed the argument passing ABI for those classes on the Sony PS4. This is probably not acceptable :)

format

Rebase + remove extra class

In D140876#4023286, @ilya-biryukov wrote:

Should access checks should happen in the context where concept is written or where it's used? Is there a standard wording around it?

This seems to be stuck with no reviewers, is there any way to help here?

Thanks for picking this up! :)