Cannot append Uint8Array to a core

[sammacbeth]

In node appending Uint8Array data to a core fails, and the callback is never called:

Given the following code:

const ram = require('random-access-memory')
const Hypercore = require('hypercore')
const core = new Hypercore(ram)
core.ready(() => {
  core.append(new Uint8Array(10), (err) => {
    if (err) {
      console.warn('append error', err);
      return
    }
    console.log('append successful');
  });
});

This code works fine in the browser, but when running in node we get the following:

    req.data.copy(page, rel, start, end)
             ^

TypeError: req.data.copy is not a function
    at RAM._write (/Users/sammacbeth/work/hyper-yjs/node_modules/random-access-memory/index.js:49:14)
    at Request._run (/Users/sammacbeth/work/hyper-yjs/node_modules/random-access-storage/index.js:207:40)
    at drainQueue (/Users/sammacbeth/work/hyper-yjs/node_modules/random-access-storage/index.js:246:48)
    at Request._unqueue (/Users/sammacbeth/work/hyper-yjs/node_modules/random-access-storage/index.js:167:23)
    at Request.callback (/Users/sammacbeth/work/hyper-yjs/node_modules/random-access-storage/index.js:172:8)
    at nextTickCallback (/Users/sammacbeth/work/hyper-yjs/node_modules/random-access-storage/index.js:272:7)
    at processTicksAndRejections (internal/process/task_queues.js:81:21)

This error comes from a tick, and the append callback is never called.

The source of this error is that the Uint8Array is being treated as a Buffer when it is not one, and this can be fixed with Buffer.from(new Uint8Array(10)). However, the error handling here is problematic - it cannot be caught or detected, and the error message appears in the storage layer.

[martinheidegger]

random-access-* assumes data to be passed-in as Buffer's rather than Uint8Array.

What do you think is the better approach: Document everything in hyper-* assumes Buffer or changing the code to assume everything is Uint8Array?

[mafintosh]

Coercing uint8arrays to buffers inside append in hypercore is prob easy

@sammacbeth would that be ok for what you are doing?

[emilbayes]

Otherwise .copy can be replaced with .set?

[mafintosh]

@emilbayes across 8 billion repos tho

[sammacbeth]

My concern is largely around error reporting - the error is currently not surfaced to the caller, making it difficult to initially debug the cause of the error.

I see two options:

1. Fail fast - throw an error when non-buffers are appended to binary cores - this forces clients to convert to buffer.
2. Do Buffer conversion automatically when uint8array is passed to append. I'm not sure what the perf cost of type check and conversion is, and if this would be an issue? I guess this can't be worse than existing conversions for other value encodings?

[mafintosh]

@sammacbeth you can convert them for "free" using buf = Buffer.from(uint8.buffer, uint8.byteOffset, uint8.byteLength)

[sammacbeth]

In that case it could probably be handled by the binary codec here: https://github.com/mafintosh/codecs/blob/master/index.js#L13

[mafintosh]

I just added support for that right there actually!

[mafintosh]

(if you reinstall it should get picked up)