The Slice encoding

Compact binary serialization format

The Slice encoding defines how each Slice language construct is encoded into a stream of bytes. It's a non-self describing format: the Slice encoding relies on the encoder and decoder sharing the same contract (Slice definitions) to achieve compactness.

For example:

compact struct Item {
    name: string
    count: int32
}

The encoder encodes a string followed by an int32 into a stream of bytes. Later on, a decoder decodes these bytes. Since the decoder has the same definitions, it expects the stream to hold an encoded string followed by an encoded int32. The stream does not encode "the following bytes represent a string". If the byte stream holds some other encoded type, the decoding fails—or if by happenstance it succeeds, the decoded data is gibberish.

Slice encoding versions

There are currently two versions of the Slice encoding: Slice1 and Slice2. The compilation mode of a Slice file determines the Slice encoding version to use when encoding the arguments and return values of operations defined in that file:

All types defined in a Slice1 file can be encoded with the Slice1 encoding, and all types defined in a Slice2 file can be encoded with the Slice2 encoding.

Furthermore, a Slice2-compatible type defined in a Slice1 file can be encoded with the Slice2 encoding.

Byte ordering

Slice is little-endian: when encoding integers and floating point numbers into multiple bytes, the first byte encoded into the byte stream holds the least significant portion of the value.