Read blob in nodejs

[jimmywarting]

Nodejs have introduced buffer.Blob and we are about to get them from fs soon also

but they have not introduced the FileReader and likely never will. cuz there are some new file reading methods on the blob now with .stream(), .arrayBuffer(), and .text() that can return promises
so i was thinking maybe you could replace this hole function and just use blob.arrayBuffer() instead and maybe polyfill blob.prototype.arrayBuffer() if it dose not exist

file-type/browser.js

Lines 18 to 45 in 2cc0869

	/**
	Convert Web API File to Node Buffer.
	@param {Blob} blob - Web API Blob.
	@returns {Promise<Buffer>}
	*/
	function convertBlobToBuffer(blob) {
	return new Promise((resolve, reject) => {
	const fileReader = new FileReader();
	fileReader.addEventListener('loadend', event => {
	let data = event.target.result;
	if (data instanceof ArrayBuffer) {
	data = toBuffer(new Uint8Array(event.target.result));
	}
	resolve(data);
	});
	fileReader.addEventListener('error', event => {
	reject(new Error(event.message));
	});
	fileReader.addEventListener('abort', event => {
	reject(new Error(event.type));
	});
	fileReader.readAsArrayBuffer(blob);
	});
	}

I have also another concern, is it really necessary to read the hole blob into memory?
isn't the first and last 1000 bytes or something only needed?

it would seem more efficient to do something like:

chunk = new Blob([
  blob.slice(0, 4500),
  blob.slice(-4500)
]).arrayBuffer()

I would aslo suggest that you use Uint8Array, ArrayBuffer and DataView instead of using nodes Buffer as that would lower the browser bundle and not drag in any of nodes core modules.

I would also suggest that you make your fromStream method acceptable of a AsyncIterator that yields Uint8Arrays instead of relying on either node streams or whatwg streams or nodes EventEmitter since it could lower the bundle even a lot further
since node stream are already async iterable. and whatwg streams are too but no browser have implemented it.

[jimmywarting]

Hmm, i quite like what you are doing but i would like a un-opinionated layer of the magic number - kind of how mime-db is to mime. mime-db is just pure JSON and dose not contain any logic to remain as un-opinionated as possible. and mime adds a api layer on top of it.

Not saying you have to go and make some new npm module called magic-db or something - just a json file within your project would also be enough so one could do require('file-type/magic.json') or something - that would be nice 🙂

[Borewit]

I would also suggest that you use Uint8Array, ArrayBuffer and DataView instead of using nodes Buffer as that would lower the browser bundle and not drag in any of nodes core modules.

As you also correctly indicated, at the very beginning in the dependency tree: Borewit/token-types#245, I propose to investigate the suggestion starting from there.

I would also suggest that you make your fromStream method acceptable of a AsyncIterator that yields Uint8Arrays instead of relying on either node streams or whatwg streams or nodes EventEmitter since it could lower the bundle even a lot further
since node stream are already async iterable. and whatwg streams are too but no browser have implemented it.

Sounds interesting as well.

just a json file within your project would also be enough so one could do require('file-type/magic.json')

That would suite those file type detections based on simple magic number.