The fetch pull algorithm does not pass proper bytesWritten to ReadableByteStreamControllerRespond

[saschanaz]

(Talking about https://fetch.spec.whatwg.org/#concept-http-network-fetch step 10 that is modified by #1593)

Implementing the algorithm as-is forces the size of resulting array buffer view be equal to that of the byob request view.

The caller may do some subarraying, but maybe the enqueue algorithm should receive bytesWritten as an argument?

[saschanaz]

The caller may do some subarraying

It's complicated when the view is not Uint8Array, so better not. This shouldn't be problem actually.

[domenic]

I think you are talking about the following case:

• current BYOB request view is non-null
• current BYOB request view's byte length > available
• Thus, bytes's length = available is less than current BYOB request view's byte length, and only bytes's length is written

Indeed the caller could create a new Uint8Array wrapping the underlying AB, but that is kind of annoying, and there's no nice Web IDL helper for it. So yeah maybe Streams should provide more help here.

Ideas:

• An optional bytesWritten argument, as you suggest
• Something that abstracts most of steps 10.2.3.1-8 for the caller?

[saschanaz]

Something that abstracts most of steps 10.2.3.1-8 for the caller?

Gecko has an internal implementation that converts any internal byte stream to ReadableStream (that is also used for Fetch), so having an abstraction in the Streams spec would probably better match with Gecko.