Bencodex.js

This library is an alternative take on implementing Bencodex in JavaScript.
It focused to address the following problems from the existing JavaScript
implementation(s):
-
No Node.js-only APIs: It does not depend on Node.js-only APIs such as
Buffer
. Usable with web browsers, Deno, and Node.js.
-
Static-time and runtime type safety: It is fully written in TypeScript,
and promise us, we use any
nowhere, and it never trusts any data from
external sources thoughtlessly. Instead, it tries to parse data and
validate them whenever it read data from external sources. It's not only
written in TypeScript, it also has runtime checks here and there so that
you could use it in vanilla JavaScript without fear.
-
Well-tested: Passing only the Bencodex spec test suite is not enough to
guarantee the whole functionality of a library, as every library has their
own part to interpret data in JavaScript (how data should look like in
JavaScript values, where data are read from and written into, whether data
are streamed or given in a single big chunk, etc). It has its own unit
tests besides spec test suite, and we've done the best to reach
near-100% test coverage.
-
Comprehensive docs: Every non-trivial library needs the three types of
docs: tutorials for getting stated, in-depth manuals for common scenarios,
and complete API references with concise examples. It has all of them!
-
Reducing unnecessary memcpy: Instead of creating chunks of
Uint8Array
s and then allocating a large buffer to concatenate the whole
data once again, it allocates only the necessary buffer only once.
It is not just memory-efficient, also time-efficient as memcpy quite
takes time.
-
Proper binary keys: It does not simply depend on Map
, which compares
its Uint8Array
keys by reference equality. Instead, it implements its
own proper dictionary which compares Uint8Array
by content equality.
Distributed under LGPL 2.1 or later.
Getting started
See also the complete API references as well.
Deno
It's available on deno.land/x/bencodex:
import { decode, encode } from "https://deno.land/x/bencodex/mod.ts";
const value = new Map([
["foo", true],
["bar", null],
["baz", new Uint8Array([0x73, 0x70, 0x61, 0x6d])],
[
new Uint8Array([0x68, 0x65, 0x6c, 0x6c, 0x6f]),
[
123456n,
"Unicode string",
false,
],
],
]);
const encoded = encode(value);
const decoded = decode(encoded);
Node.js
Add @planetarium/bencodex to your dependencies using your favorite package
manager:
npm install @planetarium/bencodex
The API usage is equivalent to the example above for Deno, except that you
need to import things from "@planetarium/bencodex"
instead of the
deno.land/x/ URL:
import { decode, encode } from "@planetarium/bencodex";
Types
Bencodex | TypeScript (Value ) |
---|
Null | null |
Boolean | boolean |
Integer | bigint (not number ) |
Binary | Uint8Array |
Text | string |
List | Value[] |
Dictionary | Dictionary |
Benchmarks
$ deno bench
cpu: 12th Gen Intel(R) Core(TM) i5-1235U
runtime: deno 1.30.3 (x86_64-unknown-linux-gnu)
file:///home/dahlia/src/planetarium/bencodex.js/bench/comparison.bench.ts
benchmark time (avg) (min … max) p75 p99 p995
------------------------------------------------------------------- -----------------------------
encoding (bencodex.js) 28.96 µs/iter (17.24 µs … 10.43 ms) 22.25 µs 117.5 µs 221.98 µs
encoding (disjukr/bencodex) 66.01 µs/iter (39.36 µs … 4.38 ms) 55.58 µs 215.2 µs 313.14 µs
summary
encoding (bencodex.js)
2.28x faster than encoding (disjukr/bencodex)
decoding (bencodex.js) 7.73 µs/iter (6.35 µs … 16.97 ms) 6.92 µs 17.39 µs 30.93 µs
decoding (disjukr/bencodex) 45.76 µs/iter (31.2 µs … 5.08 ms) 46.35 µs 135.29 µs 185.85 µs
summary
decoding (bencodex.js)
5.92x faster than decoding (disjukr/bencodex)
You can run the benchmarks by deno bench
.