Endilll wrote:

> Taking a step back, I think what we should be aiming for here is to, at some 
> point, being able to autogenerate the diagnostics expectations for a test 
> file, so that tests can be automatically updated.

What you're proposing is regression testing, which is (I hope) different from 
what we do today, because my understanding is that we don't just write tests 
and then put whatever `expected` directives needed to make them pass, but we 
actively consider whether it's correct to have diagnostics where they are 
issued (and, conversely, whether some examples ought to be accepted), and 
whether issued diagnsotics make sense.

I'd be opposed moving towards regression testing. From what I've heard how LLVM 
is tested these days, I don't want Clang to end up there. Even if we do, this 
won't apply to C++ DR tests anyway.

> What would the design for that look like? How does the direction this patch 
> pushes us to, compare to that final direction?

I don't think this work precludes from going in the direction of regression 
tests. Every test that passes with `-verify-directives` passes without it 
(kinda by definition), so if we find in the future that it gets in the way, we 
can easily drop it.

https://github.com/llvm/llvm-project/pull/179835
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to