Nonsense suggestion for concat!(x)

[ollie27]

The error for concat!(x) is:

error: expected a literal
 --> src/main.rs:2:13
  |
2 |     concat!(x);
  |             ^
help: you might be missing a string literal to format with
  |
2 |     concat!("{}", x);
  |             ^^^^^^^

That suggestion will clearly not work and gets worse when there are more arguments:

error: expected a literal
 --> src/main.rs:2:13
  |
2 |     concat!(x, y, z);
  |             ^
help: you might be missing a string literal to format with
  |
2 |     concat!("{}", x, y, z);
  |             ^^^^^^^

error: expected a literal
 --> src/main.rs:2:16
  |
2 |     concat!(x, y, z);
  |                ^
help: you might be missing a string literal to format with
  |
2 |     concat!(x, "{}", y, z);
  |                ^^^^^^^

error: expected a literal
 --> src/main.rs:2:19
  |
2 |     concat!(x, y, z);
  |                   ^
help: you might be missing a string literal to format with
  |
2 |     concat!(x, y, "{}", z);
  |                   ^^^^^^^

This suggestion was added by #51614. cc. @csmoe, @estebank

[estebank]

@csmoe, we might get away with println calling a specialized concat where the builtin is just one method with different diagnostics depending on the callee. Just a thought though, you'll likely find a better solution :)

[estebank]

It'd be ideal if we could also keep the expected a literal messages to a minimum, by emitting only one diagnostic with a MultiSpan primary span. (In the last example, instead of three errors, only one pointing at all three arguments.)

[csmoe]

@estebank I suspect this issue is more complex than I thought, here is a testcase:

fn main() {
    concat!(a);     // ERROR: ("{}", a)
    println!(a);    // FINE: ("{}", a)
}

playground
Since we are working with concat.rs here, when only ONE arg is passed into them, the emitted "{}", foo message ONLY helps println!, not concat!. And we cannot distinguish between println! and concat! here(only-one-arg-passed-in method fails), so I think such kind of help-message is blocked by the mixed literal and format string issue(print!s accept literal and format string, but concat! ONLY accepts literal), #51614 should be retrieved.

---

The only way I can think of is splitting concat! from print!s, and rewrite them into compiler-builtin macros(print.rs) like conat.rs for concat!.

[ollie27]

What I think should be done is for the suggestion to be moved to format_args! where it should always make sense. It would then work for macros like write!, print!, format! and format_args! itself. Then to get it working for the *ln! macros fixing #30143 would be enough. Maybe the prospect of better diagnostics would persuade people that it's worth fixing.