Upgrade pre-built Clang used in MSVC and MacOS builds, move MSVC builds to Server 2022

[dpaoliello]

Fixes #92948

Example working MacOS and Windows builds: https://github.com/rust-lang/rust/actions/runs/8989360201

There is a bug in Clang 18 that causes issues when building for Arm64 in later parts of the build (specifically libgit2). As a workaround, we will still use the pre-built Clang to build LLVM but will use MSVC for the rest of the Arm64 build.

[rustbot]

r? @Kobzol

rustbot has assigned @Kobzol.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[Kobzol]

I wonder why the build is slower now.

Master:

Windows 2022:

Could you please add the jobs to the try build section so that we can run them on rust-lang-ci, twice in a row, just to see if maybe some caching will start kicking in?

[dpaoliello]

Could you please add the jobs to the try build section so that we can run them on rust-lang-ci, twice in a row, just to see if maybe some caching will start kicking in?

Done

[Kobzol]

@bors try

[bors]

⌛ Trying commit 768e382 with merge 00258dc...

[dpaoliello]

I wonder why the build is slower now.

Looks like a little from git, but the majority was building LLVM. Comparing the x86-64 builds:

Part	Before	After	Delta
Total	1:29:12	1:56:52	+27:40
"checkout the source code"	0:00:22	0:01:54	+01:32
"checkout submodules"	0:01:47	0:04:48	+03:01
"run the build"	1:24:41	1:47:44	+22:57
LLVM	0:05:22	0:27:58	+22:36

The LLVM build slowdown makes sense, given that we wrap Clang in SCCache to speed up the build, but we wouldn't have a cached build using the updated version of Clang.

[Kobzol]

Yeah, that makes sense. Let's see if sccache helps.

[dpaoliello]

First run is complete, can we try again and see if it picks up the cached build from the last attempt (assuming that try builds update the cache?)

[Kobzol]

@bors try

[bors]

⌛ Trying commit f8770d8 with merge 21e0471...

[bors]

☀️ Try build successful - checks-actions
Build commit: 21e0471 (21e04717e36d43996a01589abec8505839fea807)

[Kobzol]

Ok, it's fast again, great. Could you please remove the CI modifications?

Otherwise PR looks good to me. - &job-windows-2019-16c seems to be unused after this PR (?), if it is unused, I would just remove it from jobs.yml.

[dpaoliello]

Ok, it's fast again, great. Could you please remove the CI modifications?

Otherwise PR looks good to me. - &job-windows-2019-16c seems to be unused after this PR (?), if it is unused, I would just remove it from jobs.yml.

Done - thanks so much for helping out with this :)

[Kobzol]

@bors r+ rollup=never

[bors]

📌 Commit 4aaadd1 has been approved by Kobzol

It is now in the queue for this repository.

[Kobzol]

Thanks for preparing and testing this! Let's see if it works.

[bors]

⌛ Testing commit 4aaadd1 with merge 87f5aa7...

[bors]

⌛ Testing commit 4aaadd1 with merge ef685d1...

[bors]

💔 Test failed - checks-actions

[dpaoliello]

Ah, interesting, it's Clang 18 that breaks aarch64-msvc, not VS 2022's STL - I guess I'll have to apply the workaround to this PR...

[dpaoliello]

@Kobzol Workaround for aarch64-msvc has been implemented: please have a look at builder.rs

[Kobzol]

@bors r+

[bors]

📌 Commit 5212e07 has been approved by Kobzol

It is now in the queue for this repository.

[bors]

⌛ Testing commit 5212e07 with merge a6e87c5...

[bors]

☀️ Test successful - checks-actions
Approved by: Kobzol
Pushing a6e87c5 to master...

[rust-timer]

Finished benchmarking commit (a6e87c5): comparison URL.

Overall result: ✅ improvements - no action needed

@rustbot label: -perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.3%	[-0.3%, -0.3%]	1
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

This benchmark run did not return any relevant results for this metric.

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 673.028s -> 675.734s (0.40%)
Artifact size: 315.78 MiB -> 315.98 MiB (0.06%)