warn on strings that look like inline format strings but aren't

[matthiaskrgr]

What it does

fn main() {
    let number = 123;
    let s = "{number:?}"; // this is likely a bug.
    println!("{s}");
}

Lint Name

No response

Category

No response

Advantage

No response

Drawbacks

No response

Example

fn main() {
    let number = 123;
    let s = "{number:?}"; // this is likely a bug.
    println!("{s}");
}

Could be written as:

fn main() {
    let number = 123;
    let s = format!("{number:?}"); 
    println!("{s}");
}

[awused]

I've noticed myself running into this a fair bit with the new format syntax, especially when refactoring code. Before, "{}", arg would be two arguments and fail type checking if I put it into a function call that was expecting a &str, but now "{arg}" is one &str and type checks without issue.

[matthiaskrgr]

I just found an even more interesting case IRL in kani, where it looks like expect()would take a format string, but it does just print the raw strwithout applying any formatting 😄

model-checking/kani#2865

pub fn main() {
    let mut x = Some(3);
    let y = "aaa";
    x = None.expect("hello {y}");
}

will print literally hello {y} instead of hello aaa

[TTWNO]

Found some code in my crate that did the same thing:

role.rs:632: let encoded = to_bytes(ctxt, &from_role).expect("Unable to encode {from_role}");
role.rs:635: from_slice(&encoded, ctxt).expect("Unable to convert {encoded} into Role");
interface.rs:390: let serde_val = serde_plain::to_string(&iface).expect("Unable to serialize {iface}");
tests.rs:35: .expect("Could not set sender to {unique_bus_name:?}")
tests.rs:91: .expect("Could not set sender to {unique_bus_name:?}")
tests.rs:169: .expect("Could not set sender to {unique_bus_name:?}")

Could we get a comment on if this would be a reasonable lint or not?

[emilk]

This is an easy mistake when using UI toolkits:

let x = 42;
ui.label("The value is {x}"); // oops!

[emilk]

@GuillaumeGomez wdyt?

It requires some thought on exactly what patterns to look for, and I'm sure there will be false positives (so this should probably be an opt-in lint). But at looking for the regex \{\w*:?\??\} should catch quite a lot of the problems, without too many false positives. Looking for the regex \{.*} would catch all problems, but would perhaps have too many false positives

[GuillaumeGomez]

I think it's common enough to definitely deserve a lint. I'm at rustconf currently but I'm putting it in my todo list for next week when I'm back.

EDIT: as for the "level" of check, I'll give a try at different approaches but I think a global regex that then is checked individually (to see if it's actually valid rust syntax and not {{ escaped too) should allow to rule out most false-positive.

[GuillaumeGomez]

I opened #13410. Don't hesitate to give it a try as I'm interested to find out cases I might have missed. :)