Report delivery spec algorithm can send reports erroneously about attribution-reporting-api

[apasel422]

Does it suffice to check for existence in the cache when running https://wicg.github.io/attribution-reporting-api/#attempt-to-deliver-a-report?

Today, yes, I think it suffices (modulo the race conditions called out for "in parallel"), but such a check in isolation would rely on the fields of a report being immutable, depending on whether or not report structures in the spec have identity or only value equivalence. I've long been unsatisfied with our use of set data-types for the source and report caches, as we do not specify anywhere what distinguishes two items in the set (is it all fields, or just one?).

[apasel422]

We should probably just change the source and report caches to be maps from identifier to structure.

[csharrison]

Does it suffice to check for existence in the cache when running https://wicg.github.io/attribution-reporting-api/#attempt-to-deliver-a-report?

[csharrison]

Ack - I suppose we could add an identifier in there for this purpose.

[apasel422]

Ack - I suppose we could add an identifier in there for this purpose.

I'd be fine with that (we already have such an identifier for sources), but https://infra.spec.whatwg.org/#sets doesn't even provide a means for specifying the set-member-equality operator.

[apasel422]

It may also be worth investigating parallel queues.