Ignore target feature test when LLVM fails to compile minicore

[TimNN]

llvm/llvm-project#111334 made LLVM reject certain target feature + ABI mismatches (ARM hard / soft float) which means that minicore now fails to build with errors like error: <unknown>:0:0: in function _RNvXsc_Cs4Af2OiBEA1T_8minicoreiNtB5_3Add3add i32 (i32, i32): calling convention is hard-float, but floating-point registers are unavailable

This PR ignores the test where this currently happens on LLVM 24, assuming that if LLVM is now rejecting this, we should turn the rust-side warning into an error as well. Created #161276 to track that.

r? @RalfJung (since you worked ~recently on these, including migrating them to minicore)

@rustbot label llvm-main

[rustbot]

RalfJung is not on the review rotation at the moment.
They may take a while to respond.

[RalfJung]

What exactly happens when LLVM "rejects" these?

[TimNN]

In the linked CI failure we see error: <unknown>:0:0: in function _RNvXsc_Cs4Af2OiBEA1T_8minicoreiNtB5_3Add3add i32 (i32, i32): calling convention is hard-float, but floating-point registers are unavailable when the test tries to compile minicore.

[RalfJung]

If these are turning into hard errors in LLVM, I'd say it is time to make them hard errors in Rust so it's impossible to hit those LLVM errors via rustc. Would you be willing to look into that or do you just want to get the CI green ASAP?

[TimNN]

The main priority was getting CI green again, because it's annoying if multiple failures accumulate, but that was effectively achieved by applying the llvm-main label.

I don't have time to work on making this a hard error on the Rust side at the moment. (I assume it's going to be a bit of a process with crater + fcp).

Ignoring the ARM version of the test on LLVM 24 seems like a reasonable interim solution, I'll update the PR tomorrow.

@rustbot author

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.

[RalfJung]

(I assume it's going to be a bit of a process with crater + fcp).

Crater won't help here, but yes this will need FCP.

[RalfJung]

Thanks!
@bors r+ rollup

Thinking about it some more, if we make this a hard error I guess we'll have to remove the minicore anyway as otherwise the error occurs in the wrong spot. That's a bit annoying. Ideally we'd have a way for minicore to not inherit the flags from the test itself (#147881)... we'll cross that bridge when we get there.

[rust-bors]

📌 Commit 0a1912f has been approved by RalfJung

It is now in the queue for this repository.