Allow browser parse API to consume Uint8Array

[imirkin]

Summary

The current types accepted by e.g. esm/sync.d.ts::parse() are Buffer | string. But there's no way to access the Buffer class as it is not exported.

Motivation

I have a Uint8Array that I would like to pass in. Given that the built-in Buffer class is an overlay on top of Uint8Array, seems like that should be easy to hook up.

Alternative

I've been implementing this by importing a different Buffer implementation which apparently is close enough to fool the typescript types and the internal processing of csv-parse to work out. This is not ideal.

Draft

Change esm/sync.js::parse() and related APIs to explicitly allow a Uint8Array to be passed in, and wrap it in the local Buffer type directly (or copy it to a fresh one if that's preferable). Perhaps it makes sense to include other TypedArray types as well.

[wdavidw]

Would you share a sample illustrating the usage, we could integrate it to the test suite.

[imirkin]

To adapt one of your tests in api.sync.js, you could do something like

  it("take a Uint8Array and return records", function () {
    const records = parse(new TextEncoder().encode("field_1,field_2\nvalue 1,value 2"));
    records.should.eql([
      ["field_1", "field_2"],
      ["value 1", "value 2"],
    ]);
  });

I realize I could use a TextDecoder to convert the Uint8Array into a string, but I'm hoping to avoid the overhead since it should be possible to take the Uint8Array without copying.

[wdavidw]

Your test is passing without any adjustment.

Also, there is no particular overhead. When passing a string, it is converted as a Node.js Buffer. When passing a Uint8Array, the bytes are directly read from the array.

[imirkin]

The parse() function is only advertised to take string + Buffer, not Uint8Array. Perhaps the types can just be widened?

I also wasn't sure if the Buffer.from() implementation that's distributed as part of the library would do a copy into a new buffer, or be able to wrap the Uint8Array directly.

[imirkin]

Also, to be clear, I'm talking about the browser environment, not node.js.

[wdavidw]

I am fixing the types, the JS part is good.

[imirkin]

OK great. For the avoidance of doubt, I'm talking about the type spec here: https://github.com/adaltas/node-csv/blob/master/packages/csv-parse/dist/esm/sync.d.ts