Implement Display for rustc_target::callconv::Conv

[tiif]

Follow up of #133103 (comment)

[rustbot]

r? @fmease

rustbot has assigned @fmease.
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

[rustbot]

Some changes occurred to the CTFE machinery

cc @rust-lang/wg-const-eval

The Miri subtree was changed

cc @rust-lang/miri

Some changes occurred to the CTFE / Miri interpreter

cc @rust-lang/miri

[tiif]

It's miri related, and Ralf looks busy, so r? @oli-obk

[oli-obk]

@bors r+ rollup

has no test changes as for almost all cases the Debug impl has the same output as the Display impl, adding a new test just for risc doesn't really bring much value imo, so shipping as is

[bors]

📌 Commit bbde683 has been approved by oli-obk

It is now in the queue for this repository.

[workingjubilee]

@oli-obk This error is user-facing, correct?

If so, the proposed formatting is not desirable, it should be using ExternAbi's formatting instead.

[RalfJung]

This isn't about ExternAbi though, it is about the underlying calling convention.

[tiif]

I hope it'd help by providing more context. Initially we indeed used ExternAbi here for check_abi, but after #133103 we passed FnAbi instead of ExternAbi, so we can check the number of fixed args like what is done in rust-lang/miri#4122.

[RalfJung]

Also, this is how we allow e.g. calling a C function through a C-unwind fn pointer.

[workingjubilee]

This isn't about ExternAbi though, it is about the underlying calling convention.

The underlying calling convention is still mappable to one of the ExternAbi strings, and we can still express conflicts in terms of that. Dropping data randomly, like erasing "rust-cold" to "Cold", or "x86-interrupt" to "X86Intr", will mostly be confusing.

[RalfJung]

The concrete error this comment is attached to does not have a ui test; the other one changed in this PR is tested at

rust/src/tools/miri/tests/fail/function_calls/check_callback_abi.rs

Line 12 in 522af10

//~^ ERROR: calling a function with calling convention C using calling convention Rust

and

https://github.com/rust-lang/rust/blob/e8ff9b1836fc8e69ab4fe78c35a6799277c8e700/src/tools/miri/tests/fail/function_calls/exported_symbol_abi_mismatch.rs

[workingjubilee]

huh, those two say "calling convention" but this error says "ABI"? is that what you mean by "the concrete error"... does not have a UI test?

[RalfJung]

Yeah this one here is triggered when you call a built-in shim the wrong way (rather than some normal Rust function), which we don't have a test for. We should add one.

And indeed it should say "calling convention".

[tiif]

So we should change the error message in check_abito use "calling convention" and add a test that will invoke the check_abi here.

Other than that, should we continue with impl Display for Conv or map the calling convention to ExternAbi?

[workingjubilee]

Please implement Display by mapping Conv back to the matching ExternAbi variant (without unwind). This will sometimes give a surprising result, but it will still be comprehensible. If this is properly tested, then I can elaborate on the error message in a followup that gives the rest of the information necessary to fix the problem.

Of course, if even this is considered impossible by my peers, then I can relent for now and simply make possible the impossible later.

[oli-obk]

@bors r-