New Resource Timing issues

[annevk]

@noamr when looking at whatwg/xhr#319 I started looking at fetch() again and I don't understand a couple things. Let's start with this bit:

Let handleFetchDone be to finalize and report timing with response, globalObject, and "fetch".

Where does response come from here? I guess https://fetch.spec.whatwg.org/#finalize-response has to pass it on.

How does this interact with callers of HTTP-network fetch that might take this response and determine to return a network error to the caller of fetch?

Another problem here is that this might not be the final response. We end up calling these things for redirects and such too.

One way to potentially solve this is that in "fetch finale" we declare that this response is the one we need to hear about in terms of a done callback.

I suppose if "fetch finale" got a network error we could at that point invoke "done" as well, if it should happen for network errors. Do you get an entry if there's a CORS failure? I'm not sure that's currently adequately covered.

[noamr]

In "fetch finale" we attach the timing info to the response. Intermediate redirect responses don't get that, so their timing info is null, and gets discarded when the fetch params are gone, so report and finalize will be a noop.

But I do see what you're saying, that responseDone is reported for intermediate responses - it should only be reported for responses with an attached timing info (we can make that more explicit with a flag)

[annevk]

From IRC:

if I do fetch("https://www.google.com") and then check the last entry of window.performance.getEntriesByType("resource") it's there, so network errors do result in entries afaict

And note that in this case the network error is due to CORS. I think we should either report all network errors or none of them.

Doing this would also require some kind of integration with "fetch finale" as only there do we know if a response became a network error.

[noamr]

OK this is actually my oversight, CORS errors should be reported as an RT entry, but with some of the metrics zeroed. I'll post a PR to that effect.

[noamr]

#1202 addresses both issues now.

[annevk]

Part of this was split into #1215.