HySnappy is a lightweight, high-performance Snappy decompression library compiled to WebAssembly. It provides:
- Very fast Snappy decompression suitable for web and Node.js environments.
- A minimal footprint with no external dependencies.
- Seamless integration with tools like Hyparquet.
The Snappy compression format, originally released by Google, is designed for high-speed and reasonable compression ratios. HySnappy leverages these strengths by providing a WebAssembly build that can be included directly in your JavaScript bundle for optimal performance.
The snappyUncompress
function requires arguments:
compressed
: aUint8Array
with compressed data.outputLength
: the uncompressed size of the data.
The length is needed to know how much wasm memory to allocate. For formats like parquet, this length will generally be known in advance.
To decompress a Uint8Array
with known output length:
const { snappyUncompress } = await import('hysnappy')
const compressed = new Uint8Array([
0x0a, 0x24, 0x68, 0x79, 0x70, 0x65, 0x72, 0x70, 0x61, 0x72, 0x61, 0x6d
])
const outputLength = 10
const output = snappyUncompress(compressed, outputLength) // hyperparam
Hysnappy was built specifically to accelerate the the hyparquet parquet parsing library.
Hysnappy exports a loader function snappyUncompressor()
which loads the WASM module once, and returns a pre-loaded version of snappyUncompress
function.
To use hysnappy with hyparquet:
import { parquetQuery } from 'hyparquet'
import { snappyUncompressor } from 'hysnappy'
await parquetQuery({
file,
compressors: {
SNAPPY: snappyUncompressor(),
},
})
Alternatively, check out hyparquet-compressors which includes hysnappy decompression.
The build uses clang without emscripten, in order to produce the smallest possible binary.
Run make
to build from source. The build process consists of:
- Compile from
snappy.c
tohysnappy.wasm
usingclang
. - Encode
hysnappy.wasm
as base64 tohysnappy.wasm.base64
. - Insert base64 string into
hysnappy.js
for distribution.
By keeping hysnappy.wasm
under 4kb, we can include it directly in the hysnappy.js
file and load the WASM blob synchronously, which is faster than loading a separate .wasm
file. [web.dev]