musli

Müsli

Müsli is a flexible and generic binary serialization framework.

The central components of the framework are the Encode and Decode derives. They are thoroughly documented in the derives module.

Müsli currently depends on GATs and is nightly-only

I've chosen to internally use the term "encoding", "encode", and "decode" because it's common terminology when talking about binary formats. It's also distinct from serde's use of "serialization" allowing for the ease of using both libraries side by side if desired.

Design

Müsli is designed with similar principles as serde. Relying on Rust's powerful trait system to generate code which can largely be optimized away. The end result should be very similar to a handwritten encoding.

The heavy lifting in user code is done through the Encode and Decode derives. They are both documented in the derives module.

Where Müsli differs in approach is that we don't make as heavy use of the visitor pattern. Instead the encoding interacts with the framework through encoding interfaces that describe "what it wants" and leverages GATs to make the API efficient and ergonomic.

Usage

Add the following to your Cargo.toml:

musli = "0.0.37"
musli-wire = "0.0.37"

Formats

Formats are currently distinguished by supporting various degrees of upgrade stability. A fully upgrade stable encoding format must tolerate that one model can add fields that an older version of the model should be capable of ignoring.

Partial upgrade stability can still be useful as is the case of the musli-storage format below, because reading from storage only requires decoding to be upgrade stable. So if correctly managed with #[musli(default)] this will never result in any readers seeing unknown fields.

The available formats and their capabilities are:

recorder? determines whether fields must occur in exactly the order in which they are specified. So reordering fields in such a struct would cause an error. This is only suitable for byte-oriented IPC where data models are strictly synchronized.

missing? determines if the reader can handle missing fields, as exemplified above. This is suitable for on-disk storage.

unknown? determines if the format can skip over unknown fields. This is suitable for network communication.

For every feature you drop, the format becomes more compact and efficient. musli-storage #[musli(packed)] for example is roughly as compact and efficient as bincode while musli-wire is comparable to something like protobuf*.

Examples

The following is an example of full upgrade stability using musli-wire:

use musli::{Encode, Decode};

#[derive(Debug, PartialEq, Encode, Decode)]
struct Version1 {
    name: String,
}

#[derive(Debug, PartialEq, Encode, Decode)]
struct Version2 {
    name: String,
    #[musli(default)]
    age: Option<u32>,
}

let version2 = musli_wire::to_buffer(&Version2 {
    name: String::from("Aristotle"),
    age: Some(62),
})?;

let version1: Version1 = musli_wire::decode(version2.as_slice())?;

assert_eq!(version1, Version1 {
    name: String::from("Aristotle"),
});

The following is an example of partial upgrade stability using musli-storage:

use musli::{Encode, Decode};

let version2 = musli_storage::to_buffer(&Version2 {
    name: String::from("Aristotle"),
    age: Some(62),
})?;

assert!(musli_storage::decode::<_, Version1>(version2.as_slice()).is_err());

let version1 = musli_storage::to_buffer(&Version1 {
    name: String::from("Aristotle"),
})?;

let version2: Version2 = musli_storage::decode(version1.as_slice())?;

assert_eq!(version2, Version2 {
    name: String::from("Aristotle"),
    age: None,
});

Unsafety

This library currently has two instances of unsafe:

• A mem::transcode in Tag::kind. Which guarantees that converting into the Kind enum which is #[repr(u8)] is as efficient as possible. (Soon to be replaced with an equivalent safe variant).

• A largely unsafe SliceReader which provides more efficient reading than the default Reader impl for &[u8] does (which uses split_at). Since it can perform most of the necessary comparisons directly on the pointers.

Performance

The following are the results of preliminary benchmarking and should be taken with a big grain of 🧂.

Preliminary benchmarking indicates that Müsli roundtrip encodings for large objects are about 10x faster than using JSON through serde, 5x faster than serde_cbor, and 12% faster than bincode. Note that the JSON comparison obviously isn't apples-to-apples since the Müsli encoding isn't self-descriptive, but it's included here to give a general idea of how it compares. CBOR and bincode on the other hand have comparable configurations.

For small objects the difference in encoding performance is even more significant. Müsli producing code that's 100x faster than JSON and CBOR, 20x faster than bincode (despite doing similarly oversized pre-allocation). This holds for both the wire and storage format.

json/roundtrip-large    time:   [91.263 us 91.756 us 92.239 us]
cbor/roundtrip-large    time:   [51.289 us 51.696 us 52.215 us]
bincode/roundtrip-large time:   [10.225 us 10.328 us 10.431 us]
musli-storage/roundtrip-large
                        time:   [9.0467 us 9.0881 us 9.1329 us]
musli-wire/roundtrip-large
                        time:   [11.906 us 11.933 us 11.964 us]

cbor/roundtrip-small    time:   [138.40 ns 147.94 ns 158.60 ns]
json/roundtrip-small    time:   [137.06 ns 137.93 ns 139.16 ns]
bincode/roundtrip-small time:   [16.978 ns 17.425 ns 18.057 ns]
musli-wire/roundtrip-small
                        time:   [1.0177 ns 1.0227 ns 1.0277 ns]
musli-storage/roundtrip-small
                        time:   [802.38 ps 803.95 ps 805.65 ps]

Note that these bencmarks include no "waste", like extra unrecognized fields. This is an area where Müsli's current encoding indeed is expected to lag behind since it needs to perform a fair bit of work to walk over unrecognized data.

License: MIT/Apache-2.0