#nom

nom-span

Wrap your nom input into a span to get line, column and byte offset

3 releases

Uses old Rust 2015

0.1.2 Oct 9, 2024
0.1.1 Dec 1, 2023
0.1.0 Dec 1, 2023

#793 in Parser implementations

Download history 9/week @ 2024-09-20 5/week @ 2024-09-27 131/week @ 2024-10-04 31/week @ 2024-10-11 2/week @ 2024-10-18

66 downloads per month

MIT license

16KB
292 lines

Nom Span   Build Status Latest Version

This library expose Spanned, a struct that will wraps your input and allow you to keep track of the line number, the column number and the byte offset

How to use it?

Here is a basic example of how to create the input and how to retrieve all the informations you need.

use nom_span::Spanned;

type Span<'a> = Spanned<&'a str>;

fn main() {
    let span = Span::new(
      r#"{"hello": "world 🙌"}"#,
      // Supporting UTF-8
      true
    );

    assert_eq!(span.line(), 1);
    assert_eq!(span.col(), 1);
    assert_eq!(span.byte_offset(), 0);
}

You can notice that supporting UTF-8 is optional. The reason is that UTF-8 strings need to be handled in a different way than pure ASCII strings, and thus, there can be a performance gap with UTF-8 support (see the benchmark below)

UTF-8 and ASCII comparison

A UTF-8 char can be made of 1 to 4 bytes, so counting it the ASCII way would result in counting each byte of the UTF-8 char, and will result in unexpected column number:

use nom_span::Spanned;

type Span<'a> = Spanned<&'a str>;

fn utf8_vs_ascii() {
    let utf8 = Span::new("🙌", true);
    let ascii = Span::new("🙌", false);

    let utf8_after: IResult<Span<'_>, Vec<char>> = many1(anychar)(utf8);
    let ascii_after: IResult<Span<'_>, Vec<char>> = many1(anychar)(ascii);

    let (utf8_after, _) = utf8_after.unwrap();
    let (ascii_after, _) = ascii_after.unwrap();

    assert_eq!(utf8_after.col(), 2);
    assert_eq!(ascii_after.col(), 5);
}

What about nom_locate?

I was initially using nom_locate, but I faced some huge performance issue while building a json parser, so I decided to implement my own input. I basically cloned nom_locate and modified the counting function that was causing the performance issue. So thanks a lot for this awesome crate and please go add a star to it!

What's the difference with nom_locate?

nom_locate is recounting all the chars of your entire input (even if you already consumed it) when you're calling get_column. If you're calling get_column every char, runtime would be: O(N^2) With this crate, it's counting lines and columns everytime you're consuming your input. If you're calling col every char, runtime would be: O(2N)

So if you're planning to get the column only a few times, for example, only when an error occur, it may be better to use nom_locate, but if you need it quite often, this crate should be better.

Dependencies

~1MB
~20K SLoC