2 unstable releases

0.2.0 Jun 29, 2021
0.1.0 Oct 15, 2020

#1853 in Encoding

Download history 12/week @ 2023-12-04 24/week @ 2023-12-11 41/week @ 2023-12-18 104/week @ 2024-01-01 132/week @ 2024-01-08 23/week @ 2024-01-15 32/week @ 2024-01-22 36/week @ 2024-02-12 42/week @ 2024-02-19 16/week @ 2024-02-26 2/week @ 2024-03-04 48/week @ 2024-03-11 40/week @ 2024-03-18

107 downloads per month

MIT license

42KB
661 lines

declio

A declarative I/O serialization library.

declio provides a pair of traits, [Encode] and [Decode], that facilitate bidirectional conversions between binary data streams (the std::io traits) and arbitrary data types.

These traits are implemented for many of the types in std; integer and floating-point primitives can be encoded and decoded as their big- or little-endian binary representations, and collections and other container types encode and decode the data they contain.

However, there are some notable exceptions; for example, there are no implementations for bool and str/String. This is because these types have several common representations, so to avoid accidental misuse, you are required to explicitly declare their representation. Some of the common representations are provided in the [util] module, implemented as both wrapper types and helper modules.

This crate also provides a pair of derive macros, via the default feature derive, that can implement Encode and Decode for arbitrary compound data types. By default it will encode and decode all of its fields in order, but it is highly configurable, intended to target the many different patterns found in binary formats.

Inspiration for this crate largely comes from deku, but incorporating some changes based on my own opinions and preferences. For example, declio uses byte-wise data streams from std::io instead of the bit-wise BitVecs used by deku.

Examples

Let's start with a simple example - encoding a single integer into a byte buffer:

use declio::Encode;
use declio::ctx::Endian;

let mut buf: Vec<u8> = Vec::new();
u32::encode(&0xdeadbeef, Endian::Big, &mut buf)
    .expect("encode failed");

assert_eq!(buf, [0xde, 0xad, 0xbe, 0xef]);

In this example, [Endian::Big] is a "context" value. declio provides these as an easy way to configure or alter an implementation of Encode or Decode at runtime, instead of using wrapper types or helper functions. In this case, it tells the implementation of Encode for u32 to encode the bytes in big-endian order. It can instead be set to [Endian::Little] to reverse the byte order, and in general, [Endian] can be passed to any of the integer and floating-point primitive types for similar effects.

Context can also be used to abstract some fields that are necessary to encode and decode some types. For example, containers with variable length like Vec accept a [Len] context value during decoding, which tells the Decode implementation how many values it should decode. This can be a compile-time constant like Len(1024), or it can be created from another value at runtime, like this:

use declio::Decode;
use declio::ctx::{Len, Endian};

let mut bytes = &[
    // len
    0x00, 0x02,

    // words[0]
    0xde, 0xad,

    // words[1]
    0xbe, 0xef,
][..];

let len = u16::decode(Endian::Big, &mut bytes)
    .expect("decode len failed");

let words: Vec<u16> = Vec::decode((Len(len as usize), Endian::Big), &mut bytes)
    .expect("decode bytes failed");

assert!(bytes.is_empty()); // did we consume the whole buffer?
assert_eq!(words, [0xdead, 0xbeef]);

The reason for this is that the Decode implementation for Vec does not know how to read the length value. It doesn't know what integer size or byte order the binary format uses to encode the length; it doesn't even know if the length is encoded at all! It might be some fixed length defined as part of the format.

Vec::decode can also accept an additional context value to pass to the element decoder, using a 2-tuple like (Len(len as usize), Endian::Big). However, in this example, only a Len is passed, which is also valid and will pass () as context to the element decoder.

Deriving

Here is an example which makes use of derive macros to encode and decode a user-defined data type. This is not a complete demonstration of the features of the derive macros; for a more complete reference, see the [mod@derive] module docs.

use declio::{Encode, Decode};
use declio::ctx::{Endian, Len};
use std::convert::TryInto;

#[derive(Debug, PartialEq, Encode, Decode)]
struct WithLength {
    // Context can be passed to the field decoder with a `ctx` attribute.
    #[declio(ctx = "Endian::Little")]
    len: u16,

    // Context may be different for encode and decode,
    // though they should generally be as symmetric as possible.
    // For example, `Vec` requires a `Len` when decoding, but it is
    // optional for encoding. However, it should be provided anyway
    // because it will be used to check that the encoded length is
    // the same as the actual length.
    //
    // Fields declared before this one can be accessed by name
    // (or by `field_0`, `field_1`, etc for tuple structs):
    #[declio(ctx = "Len((*len).try_into()?)")]
    bytes: Vec<u8>,
}

let bytes: Vec<u8> = vec![0xde, 0xad, 0xbe, 0xef];

let with_length = WithLength {
    len: bytes.len().try_into().expect("length out of range"),
    bytes,
};

let encoded: Vec<u8> = declio::to_bytes(&with_length)
    .expect("encode failed");
assert_eq!(encoded, [0x04, 0x00, 0xde, 0xad, 0xbe, 0xef]);

let decoded: WithLength = declio::from_bytes(&encoded)
    .expect("decode failed");

assert_eq!(decoded, with_length);

Dependencies

~0–255KB