JSON reference

json.encode / json.decode are available in v0.0.18 and share the stringify / parse implementations. Compatibility names remain available. The previous JSON implementation remains the default; yyjson is not used.

Import and interfaces

import json from "std.json"

Interface	Returns	Meaning
json.stringify(&value:any)	str!	Read only encoding into new compact JSON text
json.parse(text:str)	KuValue!	Strict parsing into an independently owned dynamic value
json.try_parse(text:str)	KuValue!	Compatibility alias for parse
json.encode(&value:any)	str!	Development branch alias for stringify
json.decode(text:str)	KuValue!	Development branch alias for parse

! denotes Result. Use ? to propagate errors or try/catch to handle them. Encoding does not consume or mutate the input.

Encode and read values

This is a complete basic example. The development branch has a broader regression example in examples/json.ku.

import json from "std.json"

fn main(): null! {
    value = { name: "Ku", count: 7 }
    text = json.stringify(value)?
    println(text)

    decoded = json.parse(text)?
    println(decoded["name"]?.as_str()?)
    println(decoded["count"]?.as_int()? + 1)
    return ok(null)
}

Object keys are sorted deterministically. Expected output:

{"count":7,"name":"Ku"}
Ku
8

A parsed KuValue cannot participate directly in integer arithmetic. Convert with .as_int()? or .as_str()? first. A mismatched type returns an error. Missing object keys also return errors rather than automatic defaults.

Data and boundaries

Supported inputs include strings, booleans, integers, finite floats, null, nested arrays, objects with string keys, and representable structs. Functions, Tasks, resource handles, Results, enums, and cycles cannot be encoded.

Rule	Current implementation
JSON syntax	Strict; rejects comments, trailing commas, extra trailing input, and invalid Unicode
Input / output	Each limited to 1,000,000 bytes
Depth budget	Root depth is 0; rejects depth > 32; 33 empty container levels are accepted, 34 rejected
Integers	Signed 64-bit; overflow is an error, without automatic float conversion
Floating point	Finite binary64; original spelling, float tags, and negative zero are not preserved
Unicode	Chinese, emoji, escapes, embedded NUL
Duplicate object keys	Parsing retains the last value
Key order	Encoding sorts by UTF-8 byte order

A float such as 1.0 may encode as 1. When an integer-looking float is outside int64, the encoder adds .0 to keep decoding on the float path: 1e20 encodes as 100000000000000000000.0 and decodes again. Finite binary64 values do not preserve original spelling, a float tag or the sign of negative zero.

Errors

Recoverable errors use the json domain. Common codes are parse_error, stringify_error, and out_of_memory. A parse message’s byte N is a zero based UTF-8 byte offset. See errors and troubleshooting for a minimal invalid input handler.

Backend support

The interpreter, default packaged runner, and synchronous native C backend have JSON implementations. The restricted native Task subset and LLVM do not support JSON. Default runner imports still depend on their source locations; JSON support does not imply a fully independent package.

Signatures and limits on this page were checked against source and contracts. Execution verification must name the version, commit, and backend. See the complete JSON syntax contract.