Use a proxy to intercept responses, rather than a clone

[Rich-Harris]

This bypasses the buffering issues encountered in #591

[benmccann]

@stalkerg could you take a look at this?

[benmccann]

Ah, so if I'm understanding this correctly, if the user calls await response.text() then that call is now made only a single time and we're not making our own call? If so, then I think that will work for the common case of getting the whole response body

I do not think that this approach will be compatible with streaming however. Are we comfortable with shutting off approaches that will let us do streaming in the future? Or should we also try to solve for the streaming use case now while we're working on this?

[Rich-Harris]

Correct, text() only gets called once. (Calling it a second time would be an error regardless of the proxy.)

I do not think that this approach will be compatible with streaming however

Why? All it means is that the streamed response doesn't get inlined. Think of the inlining as a progressive enhancement

[benmccann]

I guess you can skip the serialized data with the way hydration works today. I was thinking you wouldn't be able to do that longer-term. E.g. if we have a mode where you trust the contents of the page instead of reconciling every element that might not work. But maybe that's too far away or too speculative to worry about right now

[stalkerg]

@benmccann yes, this fix is working for me, thanks!

[Rich-Harris]

excellent, will get this merged and released. thanks for verifying