Link-time dependency on core: use of `#[inline]` functions

[fee1-dead]

See also rust-lang/rust#100759 (comment). This is undesirable as it causes a confusing linker error when modifying the library functions. #472 was opened to fixed this, but it was closed because it looked like adding #[inline] to fix this had been more appropriate. I encountered this issue again in the link above and I am creating this issue for further discussion as to whether this is an issue or not.

[Amanieu]

compiler-builtins depends on a ton of inline functions in core, such as wrapping_add. I don't think it is reasonable for us to avoid using these functions entirely.

[RalfJung]

I just spent almost an hour debugging strange linker errors until I realized they are caused by compiler_builtins and this bug. The current situation seems pretty fragile and it'd be great to find some proper solution here.

[tgross35]

We now get an error "error: compiler_builtins cannot call functions through upstream monomorphizations; encountered invalid call from foo to rustc_std_workspace_core::bar", so I think this situation has been improved (even if the error sometimes lacks details).

[Urhengulas]

Is it possible to configure compiler-builtins to be able to call libcore functions through upstream monomorphisation and therefore be able to not inline the core functions?

I understand this is bad for performance, but for our usecase this would be acceptable. Or is there another reason this is necessary apart from performance?

[bjorn3]

No, there can't be a cyclic dependency between libcore and compiler-builtins. If there is ld.bfd will error out.

[tgross35]

I understand this is bad for performance, but for our usecase this would be acceptable. Or is there another reason this is necessary apart from performance?

What usecase is this?