Some followup changes regarding epochs #126
Conversation
| but retaining browsing history, | ||
| a [=user agent=] can [=map/set=] the value | ||
| in the [=privacy budget store=] to 0 for all [=epochs=] | ||
| for all affected [=sites=]. | ||
|
|
||
| When clearing site data at the request of a [=site=], | ||
| through the use of the [:Clear-Site-Data:] header, |
There was a problem hiding this comment.
Do we need to address the points from #125 (comment)?
| However, the only real requirement is that an [=epoch=] | ||
| can be used to index the [=maplike=] structures | ||
| used for the [=impression store|impression=] | ||
| and [=epoch start store|epoch start=] stores. |
There was a problem hiding this comment.
@csharrison suggests that we keep "epoch" as a concept that refers to a fixed span of time. Then we use - in all the algorithms - an "epoch index". Both are interchangeable if you understand the start time for a site, but we get more precision if we are clearer.
| The start time for a given pair of [=user agent=] and [=site=] | ||
| is chosen randomly by the [=user agent=] when an epoch is first needed-- | ||
| as a side effect of invoking the [=get the current epoch=] algorithm. | ||
| The definition of an [=epoch=] is also reset |
There was a problem hiding this comment.
| The definition of an [=epoch=] is also reset | |
| The [=epochs=] for a given [=site=] will change |
| To <dfn>get the current epoch</dfn> | ||
| given a [=site=] |site|, | ||
| and [=moment=] |now|: | ||
| and [=moment=] |now|, |
There was a problem hiding this comment.
Since we invoke this algorithm with non "now" times (e.g. the clear time), I think we should rename it to something like get the epoch, and change the variable now to moment or something similarly abstract.
|
|
||
| To <dfn>get the starting epoch for attribution</dfn> | ||
| given [=site=] site: | ||
| given [=site=] site, | ||
| returning an [=epoch index=]: | ||
|
|
||
| 1. Let |startEpoch| be a [=user agent=]-defined value |
There was a problem hiding this comment.
I am still confused by this. Let's say my earliest epoch index is 0. Now consider a site that first used PPA 5 years ago, so its epoch start store[0] is 5 years ago. This means I need to support attribution with impressions from 5 years ago right?
Similarly, setting this to a negative number doesn't make sense since there will never be impressions < 5 years old. Setting it to a positive number doesn't make much sense because then sites that just started using the API have to wait arbitrarily long to get attribution, etc.
It seems like |startEpoch| needs to be a function of the epoch start store (which is one of the reasons I was confused about the use of an indexing-based approach in the first place)
Looking at @csharrison here for a review.
I threw in a fix for #125 with this as a bonus.
Preview | Diff