nanojson

nanojson is a JSON serializer and pull-parser with zero dependencies. No serde, no syn, no proc-macro2 — it adds nothing to your build graph. The derive macros are hand-rolled without any macro-writing libraries, so they compile instantly. The API is immediate-mode: you call methods to produce JSON and call methods to consume it, one value at a time. No intermediate tree, no Value type, no dynamic dispatch.

// Annotate your types once:
#[derive(nanojson::Serialize, nanojson::Deserialize)]
struct Point { x: i64, y: i64 }

// Then serialize and parse with a single call:
let json: String = nanojson::stringify(&Point { x: 3, y: 4 })?;
let point: Point = nanojson::parse(&json)?;

Works with std (default — ergonomic one-liners) and without it (everything on the stack, useful in embedded and firmware contexts).

Derive macros

Use derive feature (enabled by default) and annotate your types:

use nanojson::{Serialize, Deserialize};

#[derive(Serialize, Deserialize, Debug, PartialEq)]
struct Vec2 {
    x: i64,
    y: i64,
}

#[derive(Serialize, Deserialize, Debug, PartialEq)]
struct Entity {
    id: i64,
    #[nanojson(rename = "is_active")]
    active: bool,
    position: Vec2,   // nested derived struct — works automatically
    health: i64,
}

// Unit enums serialize as JSON strings:
#[derive(Serialize, Deserialize, Debug, PartialEq)]
enum Team {
    Red,
    Blue,
    #[nanojson(rename = "spectator")]
    Spectator,
}

// Struct-variant enums use externally-tagged format: {"VariantName": {...}}
#[derive(Serialize, Deserialize, Debug, PartialEq)]
enum Event {
    Spawn { entity_id: i64, x: i64, y: i64 },
    Death { entity_id: i64 },
}

std tier roundtrip

let entity = Entity { id: 42, active: true, position: Vec2 { x: 10, y: -5 }, health: 100 };

let json: String = nanojson::stringify(&entity)?;

let entity2: Entity = nanojson::parse(&json)?;
assert_eq!(entity, entity2);

no_std tier roundtrip

let mut buf = [0; 256];
let json = nanojson::stringify_sized(&mut buf, &entity)?;

let mut buf = [0; 64];
let entity2: Entity = nanojson::parse_sized(&mut buf, json)?;
assert_eq!(entity, entity2);

Feature tiers

Two API tiers

std tier (default feature)

No buffer choices. The output String grows as needed; the scratch buffer for parsing is auto-allocated to src.len() bytes (a safe upper bound).

Serialization

// One-liner for a derived type
let json: String = nanojson::stringify(&entity)?;

// Closure form for hand-written JSON
let json: String = nanojson::stringify_manual(|s| {
    s.object_begin()?;
      s.member_key("name")?; s.string("Alice")?;
      s.member_key("age")?;  s.integer(30)?;
    s.object_end()
})?;

Deserialization

// One-liner for a derived type
let entity: Entity = nanojson::parse(&json)?;
let entity: Entity = nanojson::parse_bytes(json.as_bytes())?;

// Closure form for manual parsing
let json = r#"{"x": 3, "y": 4}"#;
let (x, y) = nanojson::parse_manual(json.as_bytes(), |p, buf| {
    p.object_begin()?;
    let mut x = 0i64; let mut y = 0i64;
    while let Some(k) = p.object_member(buf)? {
        match k {
            "x" => x = p.number_str()?.parse().unwrap(),
            "y" => y = p.number_str()?.parse().unwrap(),
            _   => {}
        }
    }
    p.object_end()?;
    Ok((x, y))
})?;

no_std tier

All memory on the stack. You choose N (output buffer size) and STR_BUF (string scratch size — only needs to fit the longest single field value after escape-decoding, typically 32–128 bytes).

Serialization

// One-liner for a derived type
let mut buf = [0; 256];
let json = nanojson::stringify_sized(&mut buf, &entity)?;

// Closure form
let json = nanojson::stringify_manual_sized(&mut buf, |s| {
    s.object_begin()?;
      s.member_key("name")?; s.string("Alice")?;
      s.member_key("age")?;  s.integer(30)?;
    s.object_end()
})?;

Deserialization

// One-liner for a derived type (STR_BUF = 64)
let entity: Entity = nanojson::parse_sized(&json, &mut [0; 64])?;

// Low-level parser for hand-written code
let json = r#"{"x": 3, "y": 4}"#.as_bytes();
let (x, y) = nanojson::parse_manual_sized(json, &mut [0; 64], |p, buf| {
    p.object_begin()?;
    let mut x = 0i64; let mut y = 0i64;
    while let Some(k) = p.object_member(buf)? {
        match k {
            "x" => x = p.number_str()?.parse().unwrap(),
            "y" => y = p.number_str()?.parse().unwrap(),
            _   => {}
        }
    }
    p.object_end()?;
    Ok((x, y))
})?;

Size estimation

let n = nanojson::measure(|s| entity.serialize(s));
// n is the exact byte count — use it to pick N in stringify_sized / parse_sized.

Pretty-printing

Pass an indent width to the _pretty variants of any serialization function:

// std tier
let json = nanojson::stringify_pretty(2, &entity)?;
let json = nanojson::stringify_manual_pretty(2, |s| { ... })?;

// no_std tier
let mut buf = [0; 256];
let json = nanojson::stringify_sized_pretty(&mut buf, 2, &entity)?;
let json = nanojson::stringify_manual_sized_pretty(&mut buf, 2, |s| { ... })?;

{
  "name": "Alice",
  "active": true,
  "level": 3
}

Error handling

Serialization errors are SerializeError<W::Error>:

Parse errors are ParseError { kind: ParseErrorKind, offset: usize }. The offset is a byte position in the source slice.

Workspace layout

nanojson/
├── Cargo.toml                    workspace root
├── nanojson/                     core library (#![no_std], no dependencies)
│   ├── src/
│   │   ├── lib.rs
│   │   ├── write.rs              Write, SliceWriter, SizeCounter
│   │   ├── serialize.rs          Serializer, SerializeError, Serialize
│   │   ├── deserialize.rs        Parser, Deserialize
│   │   └── error.rs              Error types
│   ├── examples/
│   └── tests/
│       ├── readme.rs             tests for README.md code
│       ├── non_derive.rs         tests for manual parsing with derive trait
│       └── derive_roundtrip.rs   integration tests
└── nanojson-derive/              proc-macro crate (no syn/quote/proc_macro2)

Core concepts

Limitations

• Scratch buffer is reused per string. Parser::string() and Parser::object_member() both write into the same &mut [u8]. The returned &str is invalidated by the next string parse. Copy the value immediately if you need to keep it.
• No streaming / async. The serializer writes synchronously to Write; the parser requires the entire input to be in memory at once.
• No serde compatibility. nanojson is its own trait ecosystem. If you need serde interop, use serde.
• Non-finite floats are an error. Serializing f32::NAN, f64::INFINITY, etc. returns SerializeError::InvalidValue. JSON has no representation for these values.
• Nesting depth limit. The serializer's DEPTH const generic (default 32) limits how deeply you can nest objects and arrays. Use Serializer<W, 64> directly for deeper structures.
• No tuple enum variants Use enum Kind { A { number: i32 } } instead of enum Kind { A(i32) }

Running the examples

cargo run --example simple        # simple derive example
cargo run --example nostd         # almost same as simple but no_std
cargo run --example big           # big derive example
cargo run --example manual        # hand-written serialize + parse
cargo run --example derive        # derive-macro workflow
cargo run --example sensor_log    # embedded sensor log
cargo run --example recursive     # recursive tree + depth limits

Running the tests

cargo test                    # std feature (default)
cargo test --no-default-features  # no_std mode