nom, eating data byte by byte

nom is a parser combinator library with a focus on safe parsing, streaming patterns, and as much as possible zero copy.

Example

extern crate nom;

use nom::{
IResult,
bytes::complete::{tag, take_while_m_n},
combinator::map_res,
sequence::tuple};

#[derive(Debug,PartialEq)]
pub struct Color {
pub red:     u8,
pub green:   u8,
pub blue:    u8,
}

fn from_hex(input: &str) -> Result<u8, std::num::ParseIntError> {
u8::from_str_radix(input, 16)
}

fn is_hex_digit(c: char) -> bool {
c.is_digit(16)
}

fn hex_primary(input: &str) -> IResult<&str, u8> {
map_res(
take_while_m_n(2, 2, is_hex_digit),
from_hex
)(input)
}

fn hex_color(input: &str) -> IResult<&str, Color> {
let (input, _) = tag("#")(input)?;
let (input, (red, green, blue)) = tuple((hex_primary, hex_primary, hex_primary))(input)?;

Ok((input, Color { red, green, blue }))
}

fn main() {
assert_eq!(hex_color("#2F14DF"), Ok(("", Color {
red: 47,
green: 20,
blue: 223,
})));
}

The code is available on Github

There are a few guides with more details about the design of nom macros, how to write parsers, or the error management system.

Looking for a specific combinator? Read the "choose a combinator" guide

If you are upgrading to nom 5.0, please read the migration document.

Parser combinators

Parser combinators are an approach to parsers that is very different from software like lex and yacc. Instead of writing the grammar in a separate syntax and generating the corresponding code, you use very small functions with very specific purposes, like "take 5 bytes", or "recognize the word 'HTTP'", and assemble them in meaningful patterns like "recognize 'HTTP', then a space, then a version". The resulting code is small, and looks like the grammar you would have written with other parser approaches.

This gives us a few advantages:

the parsers are small and easy to write
the parsers components are easy to reuse (if they're general enough, please add them to nom!)
the parsers components are easy to test separately (unit tests and property-based tests)
the parser combination code looks close to the grammar you would have written
you can build partial parsers, specific to the data you need at the moment, and ignore the rest

Here is an example of one such parser, to recognize text between parentheses:

use nom::{
IResult,
sequence::delimited,
// see the "streaming/complete" paragraph lower for an explanation of these submodules
character::complete::char,
bytes::complete::is_not
};

fn parens(input: &str) -> IResult<&str, &str> {
delimited(char('('), is_not(")"), char(')'))(input)
}

It defines a function named parens which will recognize a sequence of the character (, the longest byte array not containing ), then the character ), and will return the byte array in the middle.

Here is another parser, written without using nom's combinators this time:

#[macro_use]
extern crate nom;

use nom::{IResult, Err, Needed};

# fn main() {
fn take4(i: &[u8]) -> IResult<&[u8], &[u8]>{
if i.len() < 4 {
Err(Err::Incomplete(Needed::Size(4)))
} else {
Ok((&i[4..], &i[0..4]))
}
}
# }

This function takes a byte array as input, and tries to consume 4 bytes. Writing all the parsers manually, like this, is dangerous, despite Rust's safety features. There are still a lot of mistakes one can make. That's why nom provides a list of function and macros to help in developing parsers.

With functions, you would write it like this:

use nom::{IResult, bytes::streaming::take};
fn take4(input: &str) -> IResult<&str, &str> {
take(4u8)(input)
}

With macros, you would write it like this:

#[macro_use]
extern crate nom;

# fn main() {
named!(take4, take!(4));
# }

nom has used macros for combinators from versions 1 to 4, and from version 5, it proposes new combinators as functions, but still allows the macros style (macros have been rewritten to use the functions under the hood). For new parsers, we recommend using the functions instead of macros, since rustc messages will be much easier to understand.

A parser in nom is a function which, for an input type I, an output type O and an optional error type E, will have the following signature:

fn parser(input: I) -> IResult<I, O, E>;

Or like this, if you don't want to specify a custom error type (it will be u32 by default):

fn parser(input: I) -> IResult<I, O>;

IResult is an alias for the Result type:

use nom::{Needed, error::ErrorKind};

type IResult<I, O, E = (I,ErrorKind)> = Result<(I, O), Err<E>>;

enum Err<E> {
Incomplete(Needed),
Error(E),
Failure(E),
}

It can have the following values:

a correct result Ok((I,O)) with the first element being the remaining of the input (not parsed yet), and the second the output value;
an error Err(Err::Error(c)) with c an error that can be built from the input position and a parser specific error
an error Err(Err::Incomplete(Needed)) indicating that more input is necessary. Needed can indicate how much data is needed
an error Err(Err::Failure(c)). It works like the Error case, except it indicates an unrecoverable error: we cannot backtrack and test another parser

Please refer to the "choose a combinator" guide for an exhaustive list of parsers. See also the rest of the documentation here. .

Making new parsers with function combinators

nom is based on functions that generate parsers, with a signature like this: (arguments) -> impl Fn(Input) -> IResult<Input, Output, Error>. The arguments of a combinator can be direct values (like take which uses a number of bytes or character as argument) or even other parsers (like delimited which takes as argument 3 parsers, and returns the result of the second one if all are successful).

Here are some examples:

use nom::IResult;
use nom::bytes::complete::{tag, take};
fn abcd_parser(i: &str) -> IResult<&str, &str> {
tag("abcd")(i) // will consume bytes if the input begins with "abcd"
}

fn take_10(i: &[u8]) -> IResult<&[u8], &[u8]> {
take(10u8)(i) // will consume and return 10 bytes of input
}

Combining parsers

There are higher level patterns, like the alt combinator, which provides a choice between multiple parsers. If one branch fails, it tries the next, and returns the result of the first parser that succeeds:

use nom::IResult;
use nom::branch::alt;
use nom::bytes::complete::tag;

let alt_tags = alt((tag("abcd"), tag("efgh")));

assert_eq!(alt_tags(&b"abcdxxx"[..]), Ok((&b"xxx"[..], &b"abcd"[..])));
assert_eq!(alt_tags(&b"efghxxx"[..]), Ok((&b"xxx"[..], &b"efgh"[..])));
assert_eq!(alt_tags(&b"ijklxxx"[..]), Err(nom::Err::Error((&b"ijklxxx"[..], nom::error::ErrorKind::Tag))));

The opt combinator makes a parser optional. If the child parser returns an error, opt will still succeed and return None:

use nom::{IResult, combinator::opt, bytes::complete::tag};
fn abcd_opt(i: &[u8]) -> IResult<&[u8], Option<&[u8]>> {
opt(tag("abcd"))(i)
}

assert_eq!(abcd_opt(&b"abcdxxx"[..]), Ok((&b"xxx"[..], Some(&b"abcd"[..]))));
assert_eq!(abcd_opt(&b"efghxxx"[..]), Ok((&b"efghxxx"[..], None)));

many0 applies a parser 0 or more times, and returns a vector of the aggregated results:

# #[macro_use] extern crate nom;
# #[cfg(feature = "alloc")]
# fn main() {
use nom::{IResult, multi::many0, bytes::complete::tag};
use std::str;

fn multi(i: &str) -> IResult<&str, Vec<&str>> {
many0(tag("abcd"))(i)
}

let a = "abcdef";
let b = "abcdabcdef";
let c = "azerty";
assert_eq!(multi(a), Ok(("ef",     vec!["abcd"])));
assert_eq!(multi(b), Ok(("ef",     vec!["abcd", "abcd"])));
assert_eq!(multi(c), Ok(("azerty", Vec::new())));
# }
# #[cfg(not(feature = "alloc"))]
# fn main() {}

Here are some basic combining macros available:

opt: will make the parser optional (if it returns the O type, the new parser returns Option<O>)
many0: will apply the parser 0 or more times (if it returns the O type, the new parser returns Vec<O>)
many1: will apply the parser 1 or more times

There are more complex (and more useful) parsers like tuple!, which is used to apply a series of parsers then assemble their results.

Example with tuple:

# #[macro_use] extern crate nom;
# fn main() {
use nom::{error::ErrorKind, Needed,
number::streaming::be_u16,
bytes::streaming::{tag, take},
sequence::tuple};

let tpl = tuple((be_u16, take(3u8), tag("fg")));

assert_eq!(
tpl(&b"abcdefgh"[..]),
Ok((
&b"h"[..],
(0x6162u16, &b"cde"[..], &b"fg"[..])
))
);
assert_eq!(tpl(&b"abcde"[..]), Err(nom::Err::Incomplete(Needed::Size(2))));
let input = &b"abcdejk"[..];
assert_eq!(tpl(input), Err(nom::Err::Error((&input[5..], ErrorKind::Tag))));
# }

But you can also use a sequence of combinators written in imperative style, thanks to the ? operator:

# #[macro_use] extern crate nom;
# fn main() {
use nom::{IResult, bytes::complete::tag};

#[derive(Debug, PartialEq)]
struct A {
a: u8,
b: u8
}

fn ret_int1(i:&[u8]) -> IResult<&[u8], u8> { Ok((i,1)) }
fn ret_int2(i:&[u8]) -> IResult<&[u8], u8> { Ok((i,2)) }

fn f(i: &[u8]) -> IResult<&[u8], A> {
// if successful, the parser returns `Ok((remaining_input, output_value))` that we can destructure
let (i, _) = tag("abcd")(i)?;
let (i, a) = ret_int1(i)?;
let (i, _) = tag("efgh")(i)?;
let (i, b) = ret_int2(i)?;

Ok((i, A { a, b }))
}

let r = f(b"abcdefghX");
assert_eq!(r, Ok((&b"X"[..], A{a: 1, b: 2})));
# }

Streaming / Complete

Some of nom's modules have streaming or complete submodules. They hold different variants of the same combinators.

A streaming parser assumes that we might not have all of the input data. This can happen with some network protocol or large file parsers, where the input buffer can be full and need to be resized or refilled.

A complete parser assumes that we already have all of the input data. This will be the common case with small files that can be read entirely to memory.

Here is how it works in practice:

use nom::{IResult, Err, Needed, error::ErrorKind, bytes, character};

fn take_streaming(i: &[u8]) -> IResult<&[u8], &[u8]> {
bytes::streaming::take(4u8)(i)
}

fn take_complete(i: &[u8]) -> IResult<&[u8], &[u8]> {
bytes::complete::take(4u8)(i)
}

// both parsers will take 4 bytes as expected
assert_eq!(take_streaming(&b"abcde"[..]), Ok((&b"e"[..], &b"abcd"[..])));
assert_eq!(take_complete(&b"abcde"[..]), Ok((&b"e"[..], &b"abcd"[..])));

// if the input is smaller than 4 bytes, the streaming parser
// will return `Incomplete` to indicate that we need more data
assert_eq!(take_streaming(&b"abc"[..]), Err(Err::Incomplete(Needed::Size(4))));

// but the complete parser will return an error
assert_eq!(take_complete(&b"abc"[..]), Err(Err::Error((&b"abc"[..], ErrorKind::Eof))));

// the alpha0 function recognizes 0 or more alphabetic characters
fn alpha0_streaming(i: &str) -> IResult<&str, &str> {
character::streaming::alpha0(i)
}

fn alpha0_complete(i: &str) -> IResult<&str, &str> {
character::complete::alpha0(i)
}

// if there's a clear limit to the recognized characters, both parsers work the same way
assert_eq!(alpha0_streaming("abcd;"), Ok((";", "abcd")));
assert_eq!(alpha0_complete("abcd;"), Ok((";", "abcd")));

// but when there's no limit, the streaming version returns `Incomplete`, because it cannot
// know if more input data should be recognized. The whole input could be "abcd;", or
// "abcde;"
assert_eq!(alpha0_streaming("abcd"), Err(Err::Incomplete(Needed::Size(1))));

// while the complete version knows that all of the data is there
assert_eq!(alpha0_complete("abcd"), Ok(("", "abcd")));