Named format arguments can be used as positional

[CAD97]

I tried this code:

fn main() {
    let what = "what?";
    println!("{a}{}", a=what);
}

I expected to see this happen:

error: 1 positional argument in format string, but no arguments were given
 --> <source>:3:18
  |
3 |     println!("{a}{}", a=what);
  |                  ^^

Instead, this happened: explanation

Standard Error

   Compiling playground v0.0.1 (/playground)
    Finished dev [unoptimized + debuginfo] target(s) in 1.94s
     Running `target/debug/playground`

Standard Output

what?what?

Meta

This happens on current 1.60.0-nightly (2022-01-27 21b4a9c), and back to rustc 1.12.0¹. On rustc 1.11.0, it gives

<std macros>:3:11: 3:36 error: invalid reference to argument `0` (no arguments given)
<std macros>:3 print ! ( concat ! ( $ fmt , "\n" ) , $ ( $ arg ) * ) ) ;
                         ^~~~~~~~~~~~~~~~~~~~~~~~~
<std macros>:3:11: 3:36 note: in this expansion of concat!
<std macros>:2:27: 2:58 note: in this expansion of format_args!
<std macros>:3:1: 3:54 note: in this expansion of print! (defined in <std macros>)
<source>:3:5: 3:37 note: in this expansion of println! (defined in <std macros>)

Given how old this "support" is, I don't know if this can be classified as a bug anymore, or if it's just more of a misfeature at this point. IMHO, this should at least get a lint, or even a future-compat lint if that's considered reasonable.

Footnotes

1. So I guess this would be considered regression-from-stable-to-stable? 🙃 ↩

[klensy]

Perhaps #93378?

[CAD97]

Looks to be it, beat me to it by a day it looks like. (I still think this working even with explicit named parameters should be linted against, though.)

[CAD97]

I restricted this just to the "named arguments can be used as positional" case, since #93378 is getting fixed for the more obviously bad case of println!("{what}{}").

[CAD97]

Note that if this gets upgraded to a hard error, #93394 can be backed out, as it exists just so that implicit captures don't count as positional arguments (but named arguments still do).

[m-ou-se]

I agree with @CAD97 said: While it can be confusing, I'm not sure it's something we should consider a bug, as this behaviour has been stable for almost six years.

A lint that warns about it could be reasonable. Maybe something like this:

warning: positional argument in format string refers to named argument
 --> <source>:3:18
  |
3 |     println!("{a}{}", a=what);
  |                  ^^
help: use its name to avoid confusion
  |
3 |     println!("{a}{a}", a=what);
  |                   +

[Veetaha]

this behaviour has been stable for almost six years.

@m-ou-se I don't think this behavior was documented. I did stumble upon it even before opening #93378 when writing a derive with thiserror, but thought it could be a bug in that library. It surprised me to see the same unexpected behavior in standard formatting macros!

Since this behavior wasn't documented we shouldn't consider it as a valid usage of formatting macros. If we make it a hard error I am sure a bunch of code will break (probably revealing real bugs though!), but we can make it at least a compatibility warning similar to how we got rid of panic!("{}") behavior, saying that it is considered a bug and we intend to remove this in future versions of the language.

But, I'd rather give it to crater before we make a decision.

[m-ou-se]

make it at least a compatibility warning similar to how we got rid of panic!("{}") behavior, saying that it is considered a bug and we intend to remove this in future versions of the language.

Sure, that's basically the same lint that was already suggested. Whether we tag that as a future compaibility lint or not is debatable. I don't see any reason to make this stronger than a warning. A warning helps against bugs, but anything stronger doesn't really seem to have much extra value. I don't think it makes sense to warn if any of your dependencies trigger the lint, for example.

But, I'd rather give it to crater before we make a decision.

As Rust usage outside of open-source grows, Crater is becoming less and less a reliable indicator for these sort of things. Especially since it mostly tests libraries, while print!()ing is probably way more common in binaries.

[estebank]

#98580 will lint against named args used only as named, so it won't trigger for this case.

[estebank]

rust-lang/rust-clippy#9040 will cover this case.