Re: [whatwg/streams] Add a proof of concept of optimized pipe (#511)

> IdentityTransformStream's readable side and writable side themselves work as ReadableStream and WritableStream, and they can be separately passed around. Even pipeThrough() may be given some writable / readable pair which are build without considering the transform streams concept at all. I'd like to confirm that this understanding of mine is correct, first. Or, do you have some idea in your mind that we brand check that a given pair of writable / readable to pipeThrough is "transform stream"?

I haven't thought this through fully. But my idea was that yes, we would add an unobservable brand check and fast path if we know it's a transform stream created with the TransformStream constructor (or with `TransformStream.identity()` or whatever).

> Once pipeTo() is invoked on both the readable side and the writable side of the transform stream with pending chunks, we can consider optimizing this pair of pipeTo()-s. But we need to have the transform stream flush the chunks before establishing such skipped transfer.

Got it. I understand the problem now.

> Hmm, we can also resolve this by having a requirement enforced on all pipeTo() implementations to stop writing when the normal backpressure signal is exerted as well as the current pipeTo() reference implementation is doing. Yeah, it may work.

The draft in https://github.com/whatwg/streams/pull/512 says "While WritableStreamDefaultWriterGetDesiredSize(writer) is ≤ 0 or is null, the user agent must not read from reader." I think this is what we want, although I guess it is talking about reading, not writing. Maybe it should say both.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/whatwg/streams/pull/511#issuecomment-245103518