add an appveyor config for aarch64-pc-windows-msvc

[froydnj]

This is purely a cargo-cult of things to solicit feedback from humans and/or automation failures. Not sure that the build artifacts would get packaged properly to start providing nightly tarballs for libstd, but this is at least a start.

Fixes #53864.

[rust-highfive]

r? @alexcrichton

(rust_highfive has picked a reviewer for you, use r? to override)

[alexcrichton]

Thanks! We've currently maxed out our parallelism with AppVeyor which means we can't easily add another entry to the job matrix at the moment. (as it'd take another 2 hours beyond the initial two hours). Our hard time limit is 3 hours per job but we tend to shoot for 2 hours max per job. Building a standard library, however, is pretty cheap so we can try to slot it in with an existing build!

Looking at a recent successful build it looks like the job that builds the i586 msvc target has 30 extra minutes, so mind slotting it in there with an extra --target argument?

[froydnj]

Thanks for the explanation!

Looking at a recent successful build it looks like the job that builds the i586 msvc target has 30 extra minutes, so mind slotting it in there with an extra --target argument?

I can try! I'm not sure how the "cross-compile 64-bit target on 32-bit platform" is going to go, but MSVC appears to support it, so we'll try it. Is it possible to trigger an appveyor run (ideally just the build being modified) for the changes?

[alexcrichton]

@bors: r+

Alright let's see how this goes!

[bors]

📌 Commit 215c649917797ebc276b220883c6f941c2b8cef0 has been approved by alexcrichton

[bors]

⌛ Testing commit 215c649917797ebc276b220883c6f941c2b8cef0 with merge b01953bf4fe77973d7759e0d14d6e4362bf1127e...

[bors]

💔 Test failed - status-appveyor

[froydnj]

configure: processing command line
configure: 
configure: rust.dist-src        := False
configure: llvm.clang-cl        := C:\clang-rust\bin\clang-cl.exe
configure: build.print-step-timings := True
configure: rust.remap-debuginfo := True
configure: error: Option 'target' provided more than once
Command exited with code 1

Well, that didn't go as planned!

[alexcrichton]

Ah right I think the targets need to be comma separated in one argument, my bad!

[froydnj]

Ah right I think the targets need to be comma separated in one argument, my bad!

No problem, let's try this again!

[alexcrichton]

@bors: r+

[bors]

📌 Commit 6d2fd35432bdda56df761d2c98b94bf3e3fc2122 has been approved by alexcrichton

[bors]

⌛ Testing commit 6d2fd35432bdda56df761d2c98b94bf3e3fc2122 with merge aca87cdaf6312156fe9ff996af470828e3611de0...

[bors]

💔 Test failed - status-appveyor

[froydnj]

3 hour timeout in dist-i686-msvc.

This is probably an actual problem, since we're adding more work to the dist-i686-msvc job. Apparently it's not as cheap as expected to build just a stdlib?

[kennytm]

Ah okay. Could we add the target to dist-x86_64-msvc instead?

[froydnj]

Ah okay. Could we add the target to dist-x86_64-msvc instead?

We could try! The timings Alex referenced above suggest that there's slightly more headroom in the dist-x86_64-msvc job, and maybe building a 64-bit stdlib on a 64-bit host will lead to better results anyway?

[alexcrichton]

@bors: r+

Nah I think this is fine still or rather it is "probably fine". By updating Clang it busted all the sccache caches, which can often cause us to overflow the time limit. The caches should be primed now, so let's see if it works this time!

[bors]

📌 Commit ebf6507 has been approved by alexcrichton

[froydnj]

By updating Clang it busted all the sccache caches, which can often cause us to overflow the time limit. The caches should be primed now, so let's see if it works this time!

Ah, good point, I didn't think about that!

[bors]

⌛ Testing commit ebf6507 with merge d458ce5e099ba1dfd4ebb38410a8c7a10f713415...

[bors]

💔 Test failed - status-appveyor

[kennytm]

@bors retry

[bors]

⌛ Testing commit ebf6507 with merge d0c869c...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: alexcrichton
Pushing d0c869c to master...

[alexcrichton]

The timing of this looks good to me, the builder for this PR with aarch64 came in at just around 2 hours, which is perfect!