This crate provides encoding and decoding of integers to and from bytestring representations.
The format is described here: Google's protobuf integer encoding technique.
Please feel free to use cargo bench
to determine the rate at which your
machine can encode and decode varints and fixedints. Note that one iteration
comprises each eight rounds of encoding (or decoding) a signed and an unsigned
integer each -- divide the resulting benchmark time by 16 in order to have a
rough estimate of time per operation. The integers are very large, so the
results represent the worst case.
If you use Tokio v0.2 and you use the asynchronous types in this crate (feature
tokio_async
), you may be interested in the v2.0
branch. It is still
maintained with the occasional fix for edge cases and depends on Tokio v0.2.
FixedInt
casts integers to bytes by either copying the underlying memory or
performing a transmutation. The encoded values use are little-endian.
However, a trait method is implemented for all integer types allowing convenient conversion between little and big endian. That is, if you receive a big-endian on the wire and decode it, it will first be interpreted as little-endian; converting will recover the correct value.
VarInt
encodes integers in blocks of 7 bits; the MSB is set for every byte but
the last, in which it is cleared.
Signed values are first converted to an unsigned representation using zigzag encoding (also described on the page linked above), and then encoded as every other unsigned number.