Update to clang 20.1.0

[sbomer]

See dotnet/runtime#109939 (comment). The ppc64le build break has been fixed: dotnet/runtime#112777.

[jkoritzinsky]

We should update the version in the optimization images as well.

[richlander]

I'm assuming we will not merge this change until we do a test run. We need confidence that we will not break CI, particularly since one break has already been found.

@jkotas

[am11]

In the past, we have waited for the official release (and sometimes for update 2; which gives them enough time to fix major issues with the release). llvm 20.0 is scheduled for Mar 11th and 20.1 two weeks after that. Can this not wait?

[richlander]

We could wait. What concerns do you have about "major issues"? Do you think the issues might block us shipping a preview?

It's also a question of whether we want to help finding those issues. I think that would be useful to the community.

[am11]

We could wait. What concerns do you have about "major issues"? Do you think the issues might block us shipping a preview?

In the first few weeks after the release, they usually prioritize fixing and backporting bugfixes discovered by large consumer base (most folks do not test the release candidate branches). Empirically speaking, the .1, .2 llvm-toolchain releases are usually more stable than RTM (just like .NET).

https://github.com/llvm/llvm-project/issues?q=is:issue%20is:closed%20label:regression:19

It's also a question of whether we want to help finding those issues. I think that would be useful to the community.

I think it is a good plan. We can update it during RC period each year.

[richlander]

Focusing on RC was an elaborate ploy to get us to adopt the GA release week-of!

[richlander]

Thanks @sbomer!

[am11]

BTW, new images are not built because the heuristics aren't perfect. We would need to touch crossdeps-llvm etc. or someone will need to manually evict the cache in ACR #1312.

[mthalman]

I'll kick the build to disable caching.