@yume-chan/no-data-view

@yume-chan/no-data-view

Plain methods to avoid creating DataViews.

Why?

If you have many short Uint8Arrays and you want to read numbers from them, creating DataViews for each of them is not a good idea:

Each DataView needs to allocate some memory, it impacts the performance, increases the memory usage, and adds GC pressure.

(If you are using DataViews for large ArrayBuffers, it's fine)

How does it work?

This package provides a set of methods to read/write numbers from Uint8Arrays without creating DataViews.

Because they are very basic number operations, the performance between a JavaScript implementation and the native DataView is nearly identical.

(Except for getBigUint64 and setBigUint64, Chrome uses an inefficient implementation, so this JavaScript implementation is even faster than the native one).

Check the benchmark for more details.

Why there is no setInt8?

Assign a negative number to a Uint8Array will treat it as an unsigned number, so there is no need to provide a setInt8 method.

(Assigning a number to Uint8Array directly translates to byte storing CPU instruction so it's very fast)

In fact, setIntXX and setUintXX is the same method, they are both provided only for consistency.

import { getInt8 } from "@yume-chan/no-data-view";

const array = new Uint8Array(1);
array[0] = -1;
console.log(array[0]); // 255
console.log(new Int8Array(array.buffer)[0]); // -1
console.log(getInt8(array, 0)); // -1

Why setIntXX doesn't need & 0xFF?

Similarly, Uint8Array only stores the lowest 8 bits of a number, so there is no need to mask the number.

const array = new Uint8Array(2);
array[0] = 0x1234;
console.log(array[0]); // 52 (0x34)
console.log(array[1]); // 0

But why setBigUint64 have & 0xFFn?

Because BigInt can contain a super huge value, that even when shifted to the right, is still not representable by Number. So they must be masked with 0xFFn before converting to Number.

(Converting between BigInt and Number has very little performance impact, because they only need one number extension or truncation CPU instruction)

const value = (BigInt(Number.MAX_SAFE_INTEGER) + 2n) << 8n;
console.log((value >> 8n) & 0xffn); // 1n
console.log(Number(value >> 8n) & 0xff); // 0