Clean up bytes module

[kt3k]

Most APIs of bytes module are simple functions that work on Uint8Arrays, but only BytesList has exceptional design and purpose.

BytesList was introduced for optimizing readDelim method of io (then bufio) module. #867 It was more like internal utility for optimization rather than a public API.

I think we should deprecate this API before stabilizing std/bytes (but we might better to check the actual usages in the ecosystem)

part of #3489

[lino-levan]

I use BytesList quite a lot as it's a pretty nice way of doing operations on a group of Uint8Arrays. I'm not sure I support deprecating it.

[kt3k]

@lino-levan Could you share your use cases?

[lino-levan]

I use it a lot for efficient encoding. It doesn't look super used in the ecosystem outside of some of my repos and iuioiua's r2d2, but maybe that's just due to it's nature. I'm for keeping this around.

[iuioiua]

I only use BytesList for concatenating Uint8Arrays. According to benchmarks, BytesList and concat() are equal in performance when the amount of Uint8Arrays to be concatenated is low. However, as you increase the amount of Uint8Arrays, BytesList becomes more and more slower than concat(). Beyond performance, I prefer functions over classes and methods because they're simpler. In fact, I will switch from using BytesList to using concat(), irrespective of whether ByteList gets deprecated.

Benchmark:

import { BytesList } from "https://deno.land/std@0.200.0/bytes/bytes_list.ts";
import { concat } from "https://deno.land/std@0.200.0/bytes/concat.ts";

const ARRAYS_LENGTH = 100;
const DATA_LENGTH = 100;

function genData(length: number) {
  const data = new Uint8Array(length);
  crypto.getRandomValues(data);
  return data;
}

const arrays = Array.from(
  { length: ARRAYS_LENGTH },
  () => genData(DATA_LENGTH),
);

Deno.bench("ByteList", { baseline: true }, () => {
  const lines = new BytesList();
  for (const array of arrays) lines.add(array);
  lines.concat();
});

Deno.bench("concat", () => {
  concat(...arrays);
});

Results:

benchmark      time (avg)        iter/s             (min … max)       p75       p99      p995
--------------------------------------------------------------- -----------------------------
ByteList        7.42 µs/iter     134,789.1   (6.38 µs … 186.42 µs)   7.12 µs  10.17 µs  36.58 µs
concat          2.82 µs/iter     355,109.3     (2.67 µs … 3.33 µs)   2.86 µs   3.33 µs   3.33 µs

summary
  ByteList
   2.63x slower than concat

I'm +1 for deprecating BytesList. My main concern would be: For each BytesList method, what must a dev use instead if BytesList were to be deprecated?

[johnspurlock]

Heads up - this seems to have broken DelimiterStream: see #3609