Status: Production deployed (in my own systems, and also in Oxide Computer systems).

hubpack is an algorithm for converting Rust values to bytes and back. It was originally designed for encoding messages sent between embedded programs. It is designed for use with serde, and its encoding is similar to a subset of ssmarshal.

Some of the nice things about hubpack include:

• Its encoding format is relatively compact.

• Its encoding format is predictable. In particular, there are no variable-length integer encodings.

• Because the size is predictable, hubpack provides a SerializedSize trait. Any type that implements SerializedSize can report the maximum number of bytes necessary to encode it using hubpack. This means you can allocate a fixed-size buffer without worry. (You can #[derive(SerializedSize)] for your own types.)

• The encode/decode implementations generate fairly small, efficient code.

• The implementation uses no unsafe code.

• The encoding format tends to play well with COBS for framing (I think it goes particularly well with corncobs but of course I'd say that).

You might not want to use hubpack because of the following limitations:

• hubpack is designed for fixed-size small data structures, and cannot encode things like Vec, str, and maps. (Though there are patterns for doing similar things if you need it; see below.)

• hubpack does not support enum types with more than 256 variants.

• hubpack aims for predictability over compactness, so certain types of data -- like lots of integers whose values are small relative to their types -- can be more compactly encoded using formats like bincode.

hubpack itself won't serialize variable length types like slice or Vec or String. You can still use hubpack to build encodings or protocols that use variable length chunks, you just have to use some care.

The basic pattern is: include a size field in the fixed-length hubpack-encoded portion of the message, and then concatenate the variable length data onto the end.

To support this, hubpack operations always give you "change." The deserialize function hands you a slice representing the unused part of its input buffer, which you can use to extract any trailing data (or another hubpack message). serialize instead gives you the serialized size, and you can slice your output buffer with that information.