7 releases (breaking)

0.8.0 Feb 19, 2022
0.7.0 Feb 13, 2022
0.6.0 Feb 11, 2022
0.5.0 Feb 8, 2022
0.3.1 Feb 5, 2022

#345 in #serialization-deserialization

31 downloads per month
Used in 2 crates (via deser)

Apache-2.0

58KB
1K SLoC

deser: an experimental serialization and deserialization library for Rust

Crates.io License Documentation

Deser is an experimental serialization system for Rust. It wants to explore the possibilities of serialization and deserialization of structural formats such as JSON or msgpack. It intentionally does not desire to support non self describing formats such as bincode.

This is not a production ready yet.

use deser::{Serialize, Deserialize};

#[derive(Debug, Serialize, Deserialize)]
#[deser(rename_all = "camelCase")]
pub struct Account {
    id: usize,
    account_holder: String,
    is_deactivated: bool,
}

This generates out the necessary Serialize and Deserialize implementations.

To see some practical examples of this have a look at the examples.

Design Goals

  • Fast Compile Times: deser avoids excessive monomorphization by encouraging dynamic dispatch. The goal is to avoid generating a lot of duplicate code that produces bloat the compiler needs to churn through.
  • Simple Data Model: deser simplifies the data model on the serialization and deserialization interface. For instance instead of making a distinction between u8 and u64 they are represented the same in the model. To compensate for this, it provides type descriptors that provide auxiliary information for when a serializer wants to process it. This helps with compile times and makes using the crate easier.
  • Native Bytes Support: deser has built-in specialization for serializing bytes and byte vectors. A Vec<u8> is serialized as bytes and does not need special handling for text-only formats such as JSON.
  • Unlimited Recursion: the real world is nasty and incoming data might be badly nested. Deser does not exhaust the call stack no matter how deep your data is. It accomplishes this by an alternative trait design to serde where handles to "sinks" or "serializable" objects are returned. This means that it's up to the caller to manage the recursion.
  • Native Optionals: the serialization system has a built-in understanding of the concept of optional data. This means that with a single attribute a struct serializer can skip over all fields currently set to null.
  • Native Flattening Support: deser's serialization and deserialization support has native support for flattening of structs. This means no internal buffering is required for #[deser(flatten)].
  • Stateful Processing: deser compensates the simplified data model with providing a space to hold meta information. Out of the box it provides information about the types that are being serialized. The additional space can be used to keep track of the "path" to the current structure during serialization and deserialization. (See deser-path for a practical example)

Deser does not intend on replacing serde but it attempts to address some if it's shortcomings. For more information there is a document about Serde Learnings with more details.

Future Plans

  • Extensible Data Model: deser wants to make it possible to extend the data model with types that are not native to the serialization interface. For instance if a data format wants to support arbitrarily sized integers this should be possible without falling back to in-band signalling.

Known Limitations

The current design of this system is very allocation heavy. This is the consequence of a certain level of flexibility paired with the dynamic dispatch nature. For instance for JSON parsing, Serde is more than 3 times faster than Deser and for deserialization 2.5 times.

Crates

  • deser: the core crate providing the base functionality
  • deser-json: basic JSON implementation for deser
  • deser-path: a crate that extends deser to track the path during serialization
  • deser-debug: formats a serializable to the std::fmt debug format

Inspiration

This crate heavily borrows from miniserde, serde and Sentry Relay's meta system. The general trait design was modelled after miniserde.

Safety

Deser (currently) uses excessive amounts of unsafe code internally. It is not vetted and it is likely completely wrong. If this design turns out to be useful there will be need to be a re-design of the internals.

Dependencies

~1.5MB
~33K SLoC