jiter

Fast iterable JSON parser.

Documentation is available at docs.rs/jiter.

jiter has three interfaces:

• JsonValue an enum representing JSON data
• Jiter an iterator over JSON data
• PythonParse which parses a JSON string into a Python object

JsonValue Example

See the JsonValue docs for more details.

use jiter::JsonValue;

let json_data = r#"
    {
        "name": "John Doe",
        "age": 43,
        "phones": [
            "+44 1234567",
            "+44 2345678"
        ]
    }"#;
let json_value = JsonValue::parse(json_data.as_bytes(), true).unwrap();
println!("{:#?}", json_value);

returns:

Object(
    {
        "name": Str("John Doe"),
        "age": Int(43),
        "phones": Array(
            [
                Str("+44 1234567"),
                Str("+44 2345678"),
            ],
        ),
    },
)

Jiter Example

To use Jiter, you need to know what schema you're expecting:

use jiter::{Jiter, NumberInt, Peek};

let json_data = r#"
    {
        "name": "John Doe",
        "age": 43,
        "phones": [
            "+44 1234567",
            "+44 2345678"
        ]
    }"#;
let mut jiter = Jiter::new(json_data.as_bytes());
assert_eq!(jiter.next_object().unwrap(), Some("name"));
assert_eq!(jiter.next_str().unwrap(), "John Doe");
assert_eq!(jiter.next_key().unwrap(), Some("age"));
assert_eq!(jiter.next_int().unwrap(), NumberInt::Int(43));
assert_eq!(jiter.next_key().unwrap(), Some("phones"));
assert_eq!(jiter.next_array().unwrap(), Some(Peek::String));
// we know the next value is a string as we just asserted so
assert_eq!(jiter.known_str().unwrap(), "+44 1234567");
assert_eq!(jiter.array_step().unwrap(), Some(Peek::String));
// same again
assert_eq!(jiter.known_str().unwrap(), "+44 2345678");
// next we'll get `None` from `array_step` as the array is finished
assert_eq!(jiter.array_step().unwrap(), None);
// and `None` from `next_key` as the object is finished
assert_eq!(jiter.next_key().unwrap(), None);
// and we check there's nothing else in the input
jiter.finish().unwrap();

Benchmarks

There are lies, damned lies and benchmarks.

In particular, serde-json benchmarks use serde_json::Value which is significantly slower than deserializing to a string.

For more details, see the benchmarks.

benchmark	jiter iter	jiter value	serde value	serde/jiter
strings
x100	8ns	9ns	40ns	4.4x
sentence	232ns	124ns	293ns	2.4x
unicode	268ns	159ns	308ns	1.9x
unicode_dense	152ns	152ns	177ns	1.2x
string_array	466ns	751ns	3.0µs	4.0x
pass2	304ns	785ns	576ns	0.7x
json_cases_strings	-	14.92ms	58.85ms	3.9x
json_cases_escapes	-	1.62ms	2.44ms	1.5x
json_cases_non-ascii	-	715.5µs	995.3µs	1.4x
numbers
short_numbers	-	10.9µs	37.6µs	3.4x
floats_array	15.2µs	19.2µs	117.2µs	6.1x
doubles_array	13.2µs	17.6µs	110.9µs	6.3x
short_floats	9.7µs	12.6µs	46.6µs	3.7x
long_significand_floats	12.1µs	13.9µs	94.7µs	6.8x
bigints_array	12.2µs	12.5µs	69.3µs	5.6x
massive_ints_array	73.2µs	77.9µs	284.9µs	3.7x
big	2.93ms	3.96ms	20.32ms	5.1x
json_cases_numbers	-	7.81ms	50.42ms	6.5x
json_cases_ints	-	7.82ms	50.27ms	6.4x
json_cases_floats	-	3.04ms	17.68ms	5.8x
constants
true_array	194ns	477ns	1.1µs	2.3x
true_object	2.4µs	1.4µs	6.0µs	4.2x
json_cases_constants	-	25.6µs	66.3µs	2.6x
documents
pass1	-	1.8µs	5.3µs	3.0x
medium_response	-	2.2µs	6.8µs	3.1x
json_cases_all	-	22.15ms	90.38ms	4.1x
json_cases_arrays	-	10.78ms	45.96ms	4.3x
json_cases_objects	-	12.98ms	53.66ms	4.1x
json_cases_deep	-	4.63ms	12.83ms	2.8x
json_cases_whitespace	-	6.24ms	16.58ms	2.7x
json_cases_error	-	2.38ms	11.87ms	5.0x

Part of the Pydantic Stack

The Pydantic Stack is everything you need to ship production-grade AI agents:

• Pydantic AI - Type-safe agent framework
• Pydantic Logfire - AI-first, full-stack observability
• Logfire AI Gateway - Unified LLM proxy