#buffers #rope #zero-copy #io


A rope-like non-contiguous buffer for efficient data structure serialization and vectored output

7 releases

0.3.0 Jan 28, 2021
0.2.0 Oct 28, 2020
0.1.0 Jun 26, 2020

#1345 in Data structures

Download history 62/week @ 2023-08-05 86/week @ 2023-08-12 128/week @ 2023-08-19 143/week @ 2023-08-26 115/week @ 2023-09-02 136/week @ 2023-09-09 93/week @ 2023-09-16 147/week @ 2023-09-23 47/week @ 2023-09-30 17/week @ 2023-10-07 40/week @ 2023-10-14 163/week @ 2023-10-21 82/week @ 2023-10-28 57/week @ 2023-11-04 24/week @ 2023-11-11 26/week @ 2023-11-18

208 downloads per month
Used in parquet2json

MIT license

568 lines

Chunked Bytes

This crate provides two variants of ChunkedBytes, a non-contiguous buffer for efficient data structure serialization and vectored output.

In network programming, there is often a need to serialize data structures into a wire protocol representation and send the resulting sequence of bytes to an output object, such as a socket. For a variety of reasons, developers of protocol implementations prefer to serialize the data into an intermediate buffer in memory rather than deal with output objects directly in either synchronous or asynchronous form, or both. When a contiguous buffer like Vec is used for this, reallocations to fit a larger length of serialized data may adversely impact performance, while evaluating the required length to pre-allocate beforehand may be cumbersome or difficult in the context. The single contiguous buffer also forms a boundary for write requests, creating a need for copy-back schemes to avoid inefficiently fragmented writes of tail data.

Another important use case is passing network data through. If some of the data is received into Bytes handles, it should be possible to inject the data into the output stream without extra copying.

Enter ChunkedBytes, containers that can be used to coalesce data added as byte slices via the BufMut interface, as well as possible Bytes input, into a sequence of chunks suitable for implementations of the Write::write_vectored method. This design aims to deliver good performance regardless of the size of buffered data and with no need to pre-allocate sufficient capacity for it.


This project is licensed under the MIT license.


Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in chunked-bytes by you, shall be licensed as MIT, without any additional terms or conditions.