Reduce Buffer operations in EventSource

[Uzlopak]

In the EventSourceStream code of the initial EventSource implementation, alot of Buffer Operations are used to process the incoming Stream. The Buffer operations are concentrated in the _transform and parseLine method of the EventSourceStream.

I am sure, we can reduce the overhead of the buffer operations by using some simple integer variables, marking start and end of the Buffer we want to extract the subarray and then pass it accordingly parseLine method.

Maybe the Buffer.concat operation to buffer the incoming chunks can be avoided. But be careful, as the incoming stream is containing utf8 code. So you have to concat the buffers at one point to avoid utf8 issues.

In the parseLine method the field and value are also processed via .subarray. Maybe first calling toString('utf8') and then processing it makes more sense? Anyhow, some benchmarking would be useful to make a informed decision.

[KhafraDev]

In websocket I was inspired by lpinca/ws to add in a utility that returned a buffer that contained n bytes, rather than concatenating every buffer stored. https://github.com/nodejs/undici/commit/5165d6712088c41ebbf66cf71ba856924008e899

[Uzlopak]

@KhafraDev

Maybe use similar logic as in eventsource npm package? Especially https://github.com/EventSource/eventsource/pull/239

[KhafraDev]

It looks as though they behave similarly, with the benefit of ws' approach being able to lazily set the buffers and allocate an exact buffer size. Both would have to be benchmarked before I could give a definitive answer.

[metcoder95]

The solution from the PR seems to be of help for the usage of concat, but a way to iterate over the buffer to consume it while avoiding subarrays might require the approach @KhafraDev suggests