Badly placed 'Unexpected end of macro invocation' when calling a macro with no arguments

[thepowersgang]

When a macro that expects arguments is not passed any, the "unexpected end of macro invocation" message is placed at the start of the crate root file, instead of the invocation.

// Test case

macro_rules! mac {
    ($e:expr) => ("hello");
}

fn main() {
    println!("{}", mac!());
}

<anon>:1:1: 1:1 error: unexpected end of macro invocation
<anon>:1 // Test case
         ^
playpen: application terminated with error code 101

[thepowersgang]

Mostly fixed in #27584, above test case still produces a slightly unexpected result on playpen nightly, but points to the correct line (just to the println! invocation instead of mac!)

<anon>:8:5: 8:28 error: unexpected end of macro invocation
<anon>:8     println!("{}", mac!());
             ^~~~~~~~~~~~~~~~~~~~~~~

[elinorbgr]

There are still some pathological cases, for example with this file:

fn main() {}

fn foo() {
    for _ in 0..3 {
        /*
            some
            code
            here
        */
        println!();
        /*
            some
            code
            here
        */
    }
}

The generated error (on nigthly playpen) is:

<anon>:4:5: 16:6 error: unexpected end of macro invocation
<anon>:4     for _ in 0..3 {
<anon>:5         /*
<anon>:6             some
<anon>:7             code
<anon>:8             here
<anon>:9         */
         ...

[TimNN]

Both of theses examples probably have the same root cause: The "unexpected end of macro invocation" error apparently always reports the span of the outermost ast node which requires expansion. See this gist (playground) for some examples of which span is reported in different contexts.

[TimNN]

The situation seems to have improved in the latest nightly: now only nested macro invocations have an incorrect spam and not macro invocations inside of other expanded things. Updated gist (playground).

[TimNN]

Looking over some pr's since my last comment, the improvement probably happened as part of #28857.

[Mark-Simulacrum]

As noted above, non-nested macro invocations are now fixed, and those are tracked separately (#33871). Closing.