Proc-macro definition attributes are accepted in arbitrary non-item positions

[petrochenkov]

For example,

#![feature(proc_macro)]

fn main() {
    #[proc_macro]
    let x = 10;
    #[proc_macro_attribute]
    let x = 11;
    #[proc_macro_derive]
    let x = 12;
}

A lot of stable built-in attributes have this issue as well, but if we can fix it in this specific case, then why not.
I think these kinds of checks are usually performed in librustc/hir/check_attr.rs.

[csmoe]

For some built-in attributes like:

rust/src/libstd/panicking.rs

Lines 176 to 186 in 64f7de9

	#[cfg(feature = "backtrace")]
	let log_backtrace = {
	let panics = update_panic_count(0);
	if panics >= 2 {
	Some(backtrace::PrintFormat::Full)
	} else {
	backtrace::log_enabled()
	}
	};

rust/src/libstd/sys/unix/thread.rs

Lines 273 to 276 in 64f7de9

	#[cfg(target_os = "freebsd")]
	let e = libc::pthread_attr_get_np(libc::pthread_self(), &mut attr);
	#[cfg(not(target_os = "freebsd"))]
	let e = libc::pthread_getattr_np(libc::pthread_self(), &mut attr);

are they legal in this issue?

[petrochenkov]

@csmoe
Yes, a number of attributes is valid in non-item positions.
This issue is only about three #[proc_macro*] attributes.

[petrochenkov]

Closing, see #52367 (comment)