#serde #ckb #blockchain

no-std serde_molecule

Implementation of Molecule using Serde

4 stable releases

new 1.1.1 Oct 24, 2024
1.1.0 Oct 11, 2024
1.0.1 Sep 14, 2024
1.0.0 Aug 29, 2024

#279 in Encoding

Download history 142/week @ 2024-08-25 8/week @ 2024-09-01 154/week @ 2024-09-08 74/week @ 2024-09-15 20/week @ 2024-09-22 40/week @ 2024-09-29 154/week @ 2024-10-06 74/week @ 2024-10-13 156/week @ 2024-10-20

425 downloads per month

MIT license

67KB
2K SLoC

Introduction

The molecule is a serialization format used on CKB. It has several implementations in different languages (C, Rust, JavaScript). In Rust, the implementation introduces numerous types and functions, making it difficult to remember. However, with the power of the serde framework, this process can be greatly simplified. This project offers an implementation of Molecule using Serde.

How to use

Here is a simple case about how to use:

use serde::{Serialize, Deserialize};
use serde_molecule::{to_vec, from_slice};

#[derive(Serialize, Deserialize)]
pub struct Table1 {
    pub f1: u8,
    pub f2: u16,
}

fn main() {
    let t1 = Table1{f1: 0, f2: 0};
    // serialize
    let bytes = to_vec(&t1, false).unwrap();
    // deserialize
    let t2: Table1 = from_slice(&bytes, false).unwrap();
    assert_eq!(t1.f1, t2.f1);
}

First step is to add dependency in Cargo.toml:

serde = { version = "???", features = ["derive"] }
serde_molecule = { version = "???" }

Then to annotate the types with #[derive(Serialize, Deserialize)]. After that, use to_vec or from_slice to serialize/deserialize.

Types mapping

Rust types are mapping to molecule types, according to the RFC:

Rust Type Molecule Type Fixed Size
i8,u8 byte yes
i16, u16, i32, u32, i64, u64, i128, u128 array yes
f32 array([byte; 4]) yes
f64 array([byte; 8]) yes
[u8; N] array([byte; N]) yes
[T; N] array([T; N]) yes
Vec fixvec no
struct table no
#[serde(with = "struct_serde")] struct yes
#[serde(with = "dynvec_serde")] dynvec no
Option option no
enum union no
String fixvec no
BTreeMap dynvec no
HashMap dynvec no
BinaryHeap fixvec no
LinkedList fixvec no
VecDeque fixvec no
HashSet fixvec no
Tuple Struct table no
Tuple Variant table no
Struct Variant table no
Tuple not supported N/A

By default, Vec-like containers (such as Vec, BinaryHeap, etc.) are serialized into fixvec. Every element in the Vec must have a fixed size (like a molecule struct, array, or primitive type). If the element is not of a fixed size, it should be annotated with #[serde(with = "dynvec_serde")]:

use serde_molecule::dynvec_serde;
#[derive(Serialize, Deserialize)]
struct RawTransaction {
    // ...
    #[serde(with = "dynvec_serde")]
    pub outputs: Vec<CellOutput>,
}

By default, every field is considered as molecule table. If it is molecule struct, we should annotate it explicitly.

use serde_molecule::struct_serde;

#[derive(Serialize, Deserialize)]
pub struct CellInput {
    pub since: u64,
    #[serde(with = "struct_serde")]
    pub previous_output: OutPoint,
}

If the top-level type is a molecule struct, the second argument to to_vec or from_slice should be true. If the value is false, the top-level type is considered a molecule table.

For all Molecule structs, their inner and descendant fields should be "fixed size" (see the table above).

Map

The Rust map types (like BTreeMap and HashMap) can be mapped to the following Molecule schemas:

table MapEntry {
    key: KEY_TYPE,
    value: VALUE_TYPE,
}

vector Map <MapEntry>;

It is not recommended to use HashMap because its key-value pairs are stored in arbitrary order.

Union with customized id

For molecule union with customized id, see example.

no_std support

To use this library in a no_std environment:

  1. Disable default features for both serde_molecule and serde
  2. Enable the alloc feature for serde_molecule
  3. Enable the derive feature for serde

Add the following to your Cargo.toml:

serde_molecule = { version = "x.x.x", default-features = false, features = ["alloc"] }
serde = { version = "x.x.x", default-features = false, features = ["derive"] }

See the no_std example for more details.

Big Array Support

The Serde framework doesn't support arrays with element sizes greater than 32. See this solution. This limitation can be addressed using a new serde with annotation (big_array_serde):

use serde::{Deserialize, Serialize};
use serde_molecule::big_array_serde;

#[derive(Serialize, Deserialize)]
struct BigArray {
    f1: u8,
    #[serde(with = "big_array_serde")]
    f2: [u8; 33],
    #[serde(with = "big_array_serde")]
    f3: [u8; 64],
}

Drawback of Deserialization

Compared to the Rust version of the Molecule implementation, deserialization with serde_molecule consumes at least double the memory. In memory-limited scenarios, such as on-chain scripts, it's not recommended to use.

Tuple Support

Serde treats tuples (e.g., (u8, String)) similarly to arrays (e.g., [u8; 2]), which leads to limited support in the context of Molecule serialization. This is due to Serde's implementation.

However, there are some exceptions:

  1. The unit tuple () is supported and maps to an empty array.
  2. Tuple structs and tuple variants are supported and map to Molecule tables.

For general tuples, it's recommended to use a rust struct instead. For example:

use serde::{Deserialize, Serialize};

// Instead of this:
// type MyTuple = (u8, String);

// Use this:
#[derive(Serialize, Deserialize)]
struct MyStruct {
    field1: u8,
    field2: String,
}

Example

Here is an example definition of CKB types.

Dependencies

~110–340KB