[−][src]Crate ordcode
A set of primitives and Serde serializers for fast, prefix-free encoding which preserves lexicographic ordering of values.
It is intended for encoding keys and values in key-value databases.
OMG! Yet another serialization format?
In most existing designs, prefix-free encoding of byte sequences is performed by escaping
"end-of-sequence" bytes. This takes extra space, and makes it difficult to know sequence length
without processing the whole input buffer; this also complicates memory allocation for
deserialized data. Instead, we take advantage of the fact that exact record size is always
known in key-value databases. This implementation relies on "two-sided" buffer design:
sequence lengths are varint-encoded and pushed to the tail end of the buffer, so
it is possible to get original length of serialized byte sequence(s) by deserializing
a few bytes only.
For serialization, this implementation provides (very fast) calculation of exact size
of serialized data length before serialization itself. These features
enable effective and predictable buffer management for repetitive scans and no-heap
(#[no-std]
) targets.
Features
- encodings in both ascending and descending lexicographic orderings are supported
- encoding puts lengths of variable-size sequences to the end of serialized data, so resulting encoding is prefix-free and friendly to lexicographic ordering
- zero allocations, supports
#[no_std]
environments - method to cheaply get exact size of serialized data without doing actual serialization, for effective buffer management
- space-efficient varint encoding for sequence lengths and discriminants
- easily customizable (endianness, encoding of primitive types etc.), with useful pre-sets
- reader/writer traits for double-ended buffers, so you can implement your own or use implementations provided by the crate
- no unsafe code
Cargo.toml features and dependencies
serde
(on by default): includeserde
serializer and deserializer. If you need only primitives, you can opt out.std
(on by default): opt out for#[no-std]
use, you will lose some convenience methods which useVec<u8>
Stability guarantees
The underlying encoding format is simple and unlikely to change.
As a safeguard, Serializer
and Deserializer
implement FormatVersion
trait for all serializer parameter
pre-sets (params::AscendingOrder
, params::PortableBinary
, params::NativeBinary
).
Note: serializing with descending lexicographic order is particularly useful for key-value databases like rocksdb, where reverse iteration is slower than forward iteration.
Re-exports
pub use buf::DeBytesReader; |
pub use buf::DeBytesWriter; |
pub use buf::ReadFromTail; |
pub use buf::WriteToTail; |
Modules
buf | Types and traits for working with serialization and deserialization buffers |
bytes_esc | Prefix-free encoding and decoding of byte sequences, with escaping |
params | Serialization parameters traits and types |
primitives | Ordered serialization/deserialization for primitive types and byte arrays. |
varint | Fast variable length serialization of unsigned integers with |
Structs
Deserializer |
|
Serializer |
|
SizeCalc | Serialized object size calculator |
Enums
Error | Serialization and deserialization errors |
Order | lexicographic ordering for serialization |
Traits
FormatVersion | Current version of data encoding format for |
Functions
calc_size | Calculate exact size of serialized data for a |
calc_size_asc | Convenience method: same as |
de_from_bytes_asc | Deserialize value from byte slice with |
de_from_bytes_ordered | Deserialize value from mutable byte slice. |
new_de_asc | Create new default deserializer instance (with |
new_ser_asc | Create new default serializer instance (with |
ser_to_buf_asc_exact | Serialize |
ser_to_buf_ordered | Serialize |
ser_to_vec_ordered | Serialize |
Type Definitions
Result | A convenient Result type |