run-make -Zbuild-std tests fail on compiler_builtins on older branches

[cuviper]

I tried: merging backports on 1.82-beta (#130569)

I expected to see this happen: Test successful

Instead, this happened: The job dist-various-1 failed!

failures:
    [run-make] tests/run-make/compiler-builtins
    [run-make] tests/run-make/thumb-none-cortex-m

Both tests use -Zbuild-std and failed to build compiler_builtins v0.1.126:

error[E0433]: failed to resolve: could not find `naked_asm` in `arch`
##[error]  --> /cargo/registry/src/index.crates.io-6f17d22bba15001f/compiler_builtins-0.1.126/src/arm.rs:26:21
   |
26 |         core::arch::naked_asm!(
   |                     ^^^^^^^^^ could not find `naked_asm` in `arch`

error[E0433]: failed to resolve: could not find `naked_asm` in `arch`
##[error]  --> /cargo/registry/src/index.crates.io-6f17d22bba15001f/compiler_builtins-0.1.126/src/arm.rs:39:21
   |
39 |         core::arch::naked_asm!(
   |                     ^^^^^^^^^ could not find `naked_asm` in `arch`

error[E0433]: failed to resolve: could not find `naked_asm` in `arch`
##[error]  --> /cargo/registry/src/index.crates.io-6f17d22bba15001f/compiler_builtins-0.1.126/src/arm.rs:54:21
   |
54 |         core::arch::naked_asm!(
   |                     ^^^^^^^^^ could not find `naked_asm` in `arch`

error[E0433]: failed to resolve: could not find `naked_asm` in `arch`
##[error]  --> /cargo/registry/src/index.crates.io-6f17d22bba15001f/compiler_builtins-0.1.126/src/arm.rs:66:21
   |
66 |         core::arch::naked_asm!(
   |                     ^^^^^^^^^ could not find `naked_asm` in `arch`

Note that beta's library/Cargo.lock only has compiler_builtins v0.1.123, so it appears cargo isn't heeding that with -Zbuild-std.

arch::naked_asm! was added in #130146 and started being used in rust-lang/compiler-builtins#686 for 0.1.126.

To unblock beta, I'm going to add //@ only-nightly to those tests, as suggested on my PR, especially since -Zbuild-std is unstable anyway. This particular failure shouldn't be an issue when the current nightlies are promoted to the next beta, but we should think about this kind of failure mode in the long term too.

[workingjubilee]

We should probably add a test using the beta compiler to the compiler_builtins CI.

[workingjubilee]

We could also add running the beta compiler with -Zbuild-std to rust-lang/rust's CI?

[cuviper]

I think this kind of thing would also be an issue once we get to the stable promotion. I don't know if compiler_builtins wants to promise any kind of MSRV, but really I would have expected the library lock file to be used.

[saethlin]

I'm very surprised there isn't a lockfile being used here. Does -Zbuild-std not use one automatically? Should be modify the tests to cp in the one from library/Cargo.lock?

[workingjubilee]

well at least now libstd has its own lockfile, so that makes things easier.

[cuviper]

Those run-make tests are using stage0 cargo, which is 1.81.0 on the beta branch, and @weihanglo pointed out that this version doesn't have the logic to find the relatively new library/Cargo.lock.

Maybe run-make should always use the current stage's cargo as a prereq? (i.e. the same stage as rustc)

[saethlin]

Maybe run-make should always use the current stage's cargo as a prereq? (i.e. the same stage as rustc)

Definitely. I thought using bootstrap cargo was sus when I wrote that but I had already spent a lot of time trying to get any cargo to run.

[jieyouxu]

Maybe run-make should always use the current stage's cargo as a prereq?

Yes, I should've done that even if bootstrap cargo somehow worked