# Serde fast flatten
A faster `flatten` implementation for serde.
## Summary
The `Deserialize` implementation generated by `serde`'s auto-derive
macro for flattened struct fields, is quite slow compared to the
unflattened version, because it needs to store the unrecognized fields
in the parent structure for subsequent deserilization by the child
structure.
This crate provides traits that allow a more performant implementation
of struct flattening, along with some tests and a benchmark. The
intention is to provide an auto-derive macro to generate these
implementations.
## Compatibility
The test suite currently includes roundtrip tests for `serde_json`,
`serde_yaml`, `serde_cbor` (normal and packed repr), `bincode` and
`bitcode`.
For `serde_json` and `serde_yaml`, the serialized output is the same
as the standard `serde` implementation. Using `serde_cbor`, the output
differs. `bincode` and `bitcode` both fail on `serde`'s auto-derived
flatten implementation, and work correctly using the
`serde_fast_flatten`-based implementation.
## Benchmarks
A simple benchmark can be run using `cargo bench`. It serializes and
deserializes a simple structure with `serde_json`. On my machine
(Intel Ultra 9 185H) this gives the following results:
| serialize/serde_flatten | 53.415 µs | 53.522 µs | 53.634 µs |
| serialize/serde_unflattened | 76.089 µs | 76.279 µs | 76.483 µs |
| serialize/fast_flatten | 54.318 µs | 54.467 µs | 54.647 µs |
| | | | |
| deserialize/serde_flatten | 279.46 µs | 279.89 µs | 280.32 µs |
| deserialize/serde_unflattened | 122.75 µs | 123.06 µs | 123.43 µs |
| deserialize/fast_flatten | 124.03 µs | 124.20 µs | 124.38 µs |
When _serializing_, there is little difference between the
`Deserialize` implementation auto-derived by `serde` and the one
written using `serde_fast_flatten`. The unflattened representation
serializes slower, probably due to the increased output size.
For deserialization, however, there is a big difference between the
performance of `serde`'s implementation and the one by
`serde_fast_flatten`. While `serde`'s implementation takes about twice
as long to deserialize the test values from the flattened
representation as deserializing from an unflattened representation,
`serde_fast_flatten`'s performance approaches that of the unflattened
representation.
## License
Licensed under either of [Apache License, Version 2.0](LICENSE-APACHE)
or [MIT license](LICENSE-MIT) at your option.
Unless you explicitly state otherwise, any contribution intentionally
submitted for inclusion in this crate by you, as defined in the
Apache-2.0 license, shall be dual licensed as above, without any
additional terms or conditions.