[llvm] [NFCI][Coroutines] update tests to autogenerated formatting (PR #178358)

[Christian Ulmann via llvm-commits]

Dinistro wrote:

> > I'm not aware of any guidelines, so I might miss something, but doesn't this just make the test exceptionally rigid and hard to evolve?
> > I had a lot of problems evolving features in some other parts of the mono repo, because dozens of tests strictly check everything, even when that wasn't really the point of the test.
> 
> For the guidelines, see the first point in https://llvm.org/docs/TestingGuide.html#best-practices-for-regression-tests.
> 
> Generated tests make it much, much easier to evolve something, because it's a matter of regenerating tests (e.g. by passing `--update-tests` to lit) and then reviewing the diff, instead of spending a few hours updating each test by hand. The downside is that it is harder to understand what the exact intent of the test is.

Thanks for the link. I understand that this makes sense for classical regression tests that attempt to cover all the IR produced. In the case of many of the coroutine tests changed in this PR, this is not really the case. Most of these tests only check specific subparts without caring about other parts of the IR.

> The downside is that it is harder to understand what the exact intent of the test is.

I completely agree with this. We had issues like this in LLVM forks as well as CIRCT at some point. Unfortunately, this problem seems to be quite inherent for passes that produce a lot of IR (which is also the case for coro passes).

https://github.com/llvm/llvm-project/pull/178358