Parsing of floats should be more permissive with regards to NaN and +/- inf syntax

Issue by dhardy
Monday Jan 26, 2015 at 11:04 GMT

For earlier discussion, see https://github.com/rust-lang/rust/issues/21652

This issue was labelled with: A-libs in the Rust repository

When parsing an f32, f64, etc., more special syntaxes for infinity and not-a-number should be expected:

fn f64_parse(text: &str){
    match text.parse::<f64>(){
        Some(v) => println!("Parsing {} yields {}", text, v),
        None => println!("Parsing {} fails", text)
    };
}
fn main(){
    f64_parse("NaN");
    f64_parse("nan");
    f64_parse("NAN");
    f64_parse("-nan");
    f64_parse("inf");
    f64_parse("INF");
    f64_parse("infinity");
}

Result:

Parsing NaN yields NaN
Parsing nan fails
Parsing NAN fails
Parsing -nan fails
Parsing inf yields inf
Parsing INF fails
Parsing infinity fails

It is debatable whether the default parser should support all these syntaxes, but it would be very useful to have this functionality somewhere. Links:

• Boost's nonfinite_num_get
• C's strtof, etc.
• C's printf (see notes at bottom of table):

The floating point conversion functions convert infinity to inf or
infinity. Which one is used is implementation defined.
Not-a-number is converted to nan or nan(char_sequence). Which one is
used is implementation defined.
The conversions F, E, G, A output INF, INFINITY, NAN instead.

I'm sure I've come across -nan (possibly different case) as well.

Review the earlier discussion in rust-lang/rust#21652 and the examples using text.parse::() as the starting point. Determine which NaN and infinity spellings are in scope, then use the listed sample inputs as the completion check; the issue names no implementation files or tests.