Bad span in suggestion for unsafe_attr_outside_unsafe in macro expansion with cfg_attr

[ehuss]

The following code:

#![warn(unsafe_attr_outside_unsafe)]

macro_rules! m {
    () => {
        #[cfg_attr(all(), link_section = ".text.startup")]
        unsafe extern "C" fn __ctor() {}
    };
}

m! {}

generates a suggestion which when applied looks like:

#[cfg_attr(all(), liunsafe(nk_section = ".text.startup)")]

Somehow the span is off by two bytes on both the start and end.

This doesn't happen outside of a macro expansion, and it seems to also require the cfg_attr.

Meta

rustc --version --verbose:

rustc 1.84.0-nightly (143ce0920 2024-11-10)
binary: rustc
commit-hash: 143ce0920a2307b19831160a01f06f107610f1b2
commit-date: 2024-11-10
host: aarch64-unknown-linux-gnu
release: 1.84.0-nightly
LLVM version: 19.1.3

[traviscross]

cc @carbotaniuman

[ehuss]

I'm pretty sure this is related to this line:

rust/compiler/rustc_parse/src/validate_attr.rs

Line 172 in bfe809d

attr.span.with_lo(attr.span.lo() + BytePos(2)).with_hi(attr.span.hi() - BytePos(1))

This is assuming that the span is around #[no_mangle], and the offsets are to go inside the #[ and ]. However, with cfg_attr, that's not a valid assumption.

I'm looking at this in relation to #132906, as I would like to figure out a way to prevent the lint from firing on macro_rules generated from a proc-macro. However, I haven't cracked that problem. Thought I would note that I see where the problem here is.

EDIT: Correction: This only happens from a macro, so I'm not sure why it is different between the two.

[ehuss]

Posted a fix in #133270.