Emit errors or warning to user about recursion of display

[gftea]

Code

code cause recursion stack overflow

struct Hello;

impl std::fmt::Debug for Hello {
  fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
    write!(f, "{}", self)
  }
}

impl std::fmt::Display for Hello {
  fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
    write!(f, "{:}", self)
  }
}

fn main() {
    let h = Hello;
    println!("{}", h);
}

Current output

Runtime error instead of compile time error at present

thread 'main' has overflowed its stack
fatal runtime error: stack overflow

Desired output

In a large code base, such small mistake is hard to locate because it does not point out which recursion call cause stack overflow.

Rationale and extra context

1. It is calling Display in a implementation for Display, this seems to be abvious recursion can be detected by compiler. I guess there may be general effort to detect recursion call, but if such obvious recursion can be prevented by compiler earlier
2. there is no use case for using : as standalone formatter, and it is better the compiler to emit error to use : as standalone formatter because it is likely user want to use :? but drop ? by mistake,

Other cases

No response

Rust Version

rustc v1.78.0

Anything else?

https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=15e6928ea61ea7e8ac7072fa82875e97

[workingjubilee]

Macros, like the one you call there, are pretty good at obscuring these impl details, but yeah we should probably warn.

[estebank]

I could have sworn there was a similar report about using to_string within an impl Display for from maybe 2017/2018, but can't find it 😅

@workingjubilee in the wild I've seen that simple cases where the chain repeats after just one or two jumps represent the bulk of them, so beyond the obfuscation issues, I'm convincing myself that a very limited call-graph analysis to look for unconditional recursivity is warranted. It's funny that it's always involving Display or Debug though 😄

Edit: found it https://github.com/rust-lang/rust/issues/45838

[so-schen]

And regarding the 2nd suggestion, what is your view

there is no use case for using : as standalone formatter, and it is better the compiler to emit error to use : as standalone formatter because it is likely user want to use :? but miss ? by mistake,

[tbu-]

there is no use case for using : as standalone formatter, and it is better the compiler to emit error to use : as standalone formatter because it is likely user want to use :? but miss ? by mistake,

Unlikely to work, there's probably a crate out there that uses it, so it can't be removed due to backward compatibility guarantees.

[gftea]

there is no use case for using : as standalone formatter, and it is better the compiler to emit error to use : as standalone formatter because it is likely user want to use :? but miss ? by mistake,

Unlikely to work, there's probably a crate out there that uses it, so it can't be removed due to backward compatibility guarantees.

can be a warnings?

[workingjubilee]

Yes, we can lint on that. In fact we already lint on a few bad format_args! inputs.