Parser for the Nix Expressions Language.
-
✔️ Fast
It parses all the files in Nixpkgs in under 25 seconds, single-threaded. 1
It's written in Rust and a little bit of C++, Flex and GNU Bison.
-
✔️ Correct
This library is a copy-paste of the original lexer and parser of Nix, with some types adapted for better ergonomy.
No parser can get closer to the original implementation than this.
-
✔️ Reliable
High coverage, battle-tested, and memory-safe2.
-
✔️ Useful
It gives you comments, whitespace, starting and end positions, automatic string un-escaping, multiline string indentation handling, a typed API, and everything you need to parse the Nix language!
You can check out the documentation at docs.rs/nixel.
This is a full usage example:
let input: String = String::from(
r#"
# Greet the user
"Hello, World!"
# Bye!
"#,
);
let parsed: nixel::Parsed = nixel::parse(input);
match &*parsed.expression {
nixel::Expression::String(string) => {
assert_eq!(
&string.span,
&nixel::Span {
start: nixel::Position { line: 3, column: 9 }.into(),
end: nixel::Position { line: 3, column: 24 }.into(),
}
.into()
);
assert_eq!(
&parsed.trivia_before(&string.span.start)[1],
&nixel::Trivia::Comment(nixel::TriviaComment {
content: "# Greet the user".into(),
span: nixel::Span {
start: nixel::Position { line: 2, column: 9 }.into(),
end: nixel::Position { line: 2, column: 25 }.into(),
}
.into()
})
);
assert_eq!(
&string.parts[0],
&nixel::Part::Raw(nixel::PartRaw {
content: "Hello, World!".into(),
span: nixel::Span {
start: nixel::Position { line: 3, column: 10 }.into(),
end: nixel::Position { line: 3, column: 23 }.into(),
}
.into()
})
);
assert_eq!(
&parsed.trivia_after(&string.span.end)[1],
&nixel::Trivia::Comment(nixel::TriviaComment {
content: "# Bye!".into(),
span: nixel::Span {
start: nixel::Position { line: 4, column: 9 }.into(),
end: nixel::Position { line: 4, column: 15 }.into(),
}
.into()
})
);
},
expression => unreachable!("Expected a String, got: {expression:#?}"),
}
Or from the CLI using Rust's Debug trait:
$ echo '1 + 2' | nix run github:kamadorueda/nixel -- --format=debug
BinaryOperation(
BinaryOperation {
left: Integer(
Integer {
value: "1",
span: Span {
start: Position {
line: 1,
column: 1,
},
end: Position {
line: 1,
column: 2,
},
},
},
),
operator: Addition,
right: Integer(
Integer {
value: "2",
span: Span {
start: Position {
line: 1,
column: 5,
},
end: Position {
line: 1,
column: 6,
},
},
},
),
},
)
Or from the CLI using JSON format:
$ echo '1 + 2' | nix run github:kamadorueda/nixel -- --format=json
{
"BinaryOperation": {
"left": {
"Integer": {
"value": "1",
"span": {
"start": {
"line": 1,
"column": 1
},
"end": {
"line": 1,
"column": 2
}
}
}
},
"operator": "Addition",
"right": {
"Integer": {
"value": "2",
"span": {
"start": {
"line": 1,
"column": 5
},
"end": {
"line": 1,
"column": 6
}
}
}
}
}
}
You can check out more examples in the tests folder.
Please read LICENSE.md.
Footnotes
-
Running on a machine with:
- CPU: 4 physical, 4 logical, 11th Gen Intel(R) Core(TM) i7-1165G7 @ 2.80GHz
- MHz: from 400 to 4700 MHz
- BogoMips: 5606.40
- Cache L3: 12 MiB
The following command takes around 1 minute:
↩$ nix build --system x86_64-linux $ time find /data/nixpkgs -type f -name '*.nix' \ -exec ./result/bin/nixel --format=none {} \; real 0m24.293s user 0m15.066s sys 0m8.955s
-
Tested under real-life workloads using Valgrind, and by running an infinite loop of parsing cycles over Nixpkgs :).
↩$ nix build --system x86_64-linux $ valgrind ./result/bin/nixel $file LEAK SUMMARY: definitely lost: 0 bytes in 0 blocks indirectly lost: 0 bytes in 0 blocks possibly lost: 0 bytes in 0 blocks suppressed: 0 bytes in 0 blocks