Rust: ParseIntError and ParseFloatError does not allow to introspect the cause

[nagisa]

The only way to introspect reason of .parse failure is to compare to error messages returned by description or use the following code snippet:

let overflow = "1231123123123123".parse::<u32>().err().unwrap();
let underflow = "-1231123123123123".parse::<i32>().err().unwrap();
match "999999999999".parse::<u32>() {
    Ok(n) => println!("{}", n),
    Err(ref e) if *e == overflow => println!("Overflow"),
    Err(ref e) if *e == underflow => println!("Underflow"),
    _ => panic!("wow, no introspection, very convenience")
}

[alexcrichton]

Note that this was purposefully done to avoid adding extra #[stable] surface for now. It should be able to expose this information through an enum in the future, for example.

[steveklabnik]

Triage: no change here.

@rust-lang/libs is the intention still to change this to an enum of some kind in the future?

[Seeker14491]

Any reason to postpone this further?

[BurntSushi]

@Seeker14491 I don't think it's postponed. I think someone just needs to propose an API?

[phungleson]

Hey, we already have an enum IntErrorKind, just want to check if anything still need to be done to make it pub

https://github.com/rust-lang/rust/blob/fc02736d59252fe408dd6c2f7e2c4b6f229e4443/src/libcore/num/mod.rs#L2769-L2774

[gypsydave5]

I just brushed up against this issue while demoing the guessing game example from the Rust Book - seemed like an easy thing to show off matching on the error kind, but turned out not so much!

[frontsideair]

I also wanted to match IntErrorKind::Empty in the number game to abort on EOF, turned out to be impossible.

[sfackler]

@frontsideair you can just check if the string's empty before parsing, right?

[frontsideair]

I guess, but it was a fun exercise in pattern matching.

[dtolnay]

I see that checking for empty has come up in the context of the guessing game tutorial. Could somebody make a list of use cases from real code that needs to be able to distinguish between the following other ParseIntError cases?

• invalid digit
• number too large to fit in target type
• number too small to fit in target type

In the event that empty is the only one we ever need to introspect for, I would prefer to close this and instead recommend to check == "" before parsing.

[ranma42]

It can be convenient to distinguish too large/too small numbers when accepting configuration parameters that the code will later sanitize by clamping to a valid range. Specifically, they make it possible to perform saturation of the parsed value before it is passed on.

Example: the valid range of offsets is -10..+10, therefore the application stores it in an i8. When the value it outside of the -10..+10 range the app would like to warn the user and proceed with the nearest legitimate value.

Without the ability to discriminate if the number is too big/small, it is not possible to determine what is the nearest legitimate value.

[dtolnay]

@ranma42 do you think that use case would be solved by https://github.com/rust-lang/rfcs/issues/928?

use std::num::Saturating;

fn main() {
    let s = "-999";
    let n = match s.parse::<Saturating<i8>>()?.0 {
        std::i8::MIN ... -11 => { println!("saturating"); -10 }
        11 ... std::i8::MAX => { println!("saturating"); 10 }
        n => { println!("in range"); n }
    };
    println!("n={}", n);
}

[ranma42]

Yes, that use case would be covered if Rust had the ability to parse a Saturating<i8> (and in a convenient way, too!)

[amosbatto]

The only way I found around this problem was to do something ugly like this:

match my_str.parse::<i32>() {
    Ok(x) => {my_int = x;},
    Err(e) => {
        if e.to_string() == "invalid digit found in string" {
            println!("Only positive integers are allowed!");
        }
        else if e.to_string() == "number too large to fit in target type" {
            println!("Number is too large!");
        }
        else if e.to_string() == "number too small to fit in target type" {
            println!("Number is too small!");
        }
        else if e.to_string() == "cannot parse integer from empty string" {
            println!("Number is empty!");
        }
    }
}

What is so annoying is that println!("{:?}", e) shows me ParseIntError { kind: InvalidDigit }, but kind is private so we can't access it. Either kind should be made public or a kind() function should be implemented to access to.

[Clockwork-Muse]

I'm doing custom type parsing (something like this: X+123789Y-3453), and I'd actually like to map the errors into my own custom error type. Yes, I could implement my own integer parsing, but that would defeat the purpose of _having_ a built-in integer parser.

My map would be something like this:

• Empty -> MyInvalid (because the whole thing isn't empty, just one of the number portions)
• Underflow/Overflow -> MyUnderflow/MyOverflow (although I'd be fine with just overflow, a la checked_)
• Invalid -> MyInvalid

....except I can't do that at the moment without resorting to @amosbatto 's trick. It also makes testing very difficult, because I can't tell what error I got back (except by checking the string).

[ethanboxx]

Hello everyone. My PR fixes this issue and has been merged.

Use the feature flag #![feature(int_error_matching)] in nightly to try it out.

[tspiteri]

If this is to be stabilized, I think that the terminology has to be improved: IntErrorKind uses Underflow to refer to negative overflow, but underflow usually means when the absolute value of a floating-point number is so small it becomes 0 (or a subnormal).

[JarrettBillingsley]

Maybe PosOverflow and NegOverflow?

[getreu]

This will also solve my use case described in #39381.

[hellow554]

What is missing to make it stable?

[jyn514]

I would find this useful for NonZeroUsize::from_str() to distinguish between "0" and an invalid number.