[Resource Timing] Handling duplicate requests

I wanted to continue the discussion about Resource Timing from the
workshop...

The relevant piece of the spec is step 6 of the processing model [1]:

"If the resource is not to be fetched from the networking layer, such as
being fetched from an in-memory cache, abort the remaining steps."

My question is what to do in situations where multiple objects are
requesting the same resource. There are two important cases:

1. The easier case is when two requesters are in the same document.
2. The harder case is when two requesters are in different documents.

In case #1, I think it's clear the first one should "win" and the second
request is ignored. This should be explicitly called out in the spec.
Otherwise, technically, the second requester is waiting for the resource
"to be fetched from the networking layer." That means the current spec _is_
saying we should have duplicates.

In case #2, neither one should "win" IMO. The second document may be
blocked on that resource and it'd be unfortunate if it didn't show up in
its timeline.

However, if we do show it in case #2, then the timeline may be confusing.
For instance, the second document may have been created after the resource
began fetching, which would mean some of the high res times would be
negative.

Another option would just be to show the time spent waiting in each phase.
So if the DNS lookup occurred before the second document started fetching,
that'd be 0, but the rest of the fields would be populated.

Thoughts?

James

[1] http://w3c-test.org/webperf/specs/ResourceTiming/#processing-model