macro_rules attempts to expand tokens it needs to only pass through.

[eddyb]

macro_rules! expr {
    ($a:expr) => ($a)
}

macro_rules! str_tts {
    ($b:tt) => {
        expr!(stringify!($b))
    }
}

fn main () {
    // This works ("$ c").
    let x = expr!(stringify!($c));
    // This errors with "unknown macro variable `c`".
    let y = str_tts!($c);
    println!("{} {}", x, y);
}

On playpen.

It seems that while expanding the str_tts invocation, the $c token ends up being interpreted as a macro variable, even though it should be packaged in a tt non-terminal.

Also to be noted is the fact that stringify!($c) produces "$ c", implying there are two tokens, a dollar sign and an identifier, whereas when passed to a macro_rules macro, it "fits" in a single tt.

Maybe it was a bad idea to ever have a $ token and a $var token type, the $ token would suffice, although macro_rules expansion may become more complicated.

[Mark-Simulacrum]

stringify!($c), along with any other macro invocation with $foo as an argument, now errors with "error: unknown macro variable c", so this can either be closed or renamed/repurposed; a decision as to what we want would be nice.

[Mark-Simulacrum]

Not sure what I was talking about before, but today the following error is reported. I believe this indicates that $c is now two tokens, so accepting a single tt doesn't work. @jseyfried Can you confirm that interpretation as correct?

error: no rules expected the token `c`
  --> test.rs:15:23
   |
15 |     let y = str_tts!($c);
   |             ----------^- in this macro invocation

[eddyb]

I believe this has been fixed by one of @jseyfried's many token/macro refactors.
(Feel free to wait for proper confirmation, of course)

[jseyfried]

@Mark-Simulacrum
Yeah, that interpretation is correct -- that changed was caused by fixing #39403 in #39419.