Prerendered pages and `timeOrigin`

[yoavweiss]

In https://github.com/WICG/nav-speculation/issues/64 we've settled on adding an activationTime timestamp for prerendered pages that would indicate when the user has activated the page - essentially when it becomes user visible.

One issue that came up on that front is that RUM scripts that are not aware of that would continue to register those prerendered pages based on the times in which they were initially prerendered, which can be significantly earlier, and cause folks to think their pages got slower, when in fact all that happened is that they were prerendered.

An alternative approach could be to change the timeOrigin on prerendered pages to represent the activation time, rather than the start time of the navigation request. There are two options there:

• We could then set all the timestamps that precede the activation time to be 0, indicating that they had no impact on the user experience
• We could set the timestamps that precede the activation time to be negative timestamps, representing the time in which they happened, but at the same time, running a risk of breaking existing scripts.

[yoavweiss]

Discussed on the WebPerfWG call.

Summary:

• We have multiple options: leave timeOrigin as is and let developers rely on activationStart explicitly, or move the timeOrigin to be activationStart (either with or without negative timestamps)
• Unanimous consensus on the call was that it’s better to have developers and RUM providers look at activationStart(even if that requires updates to existing code) rather than try to retrofit the timeOrigin, as it can have many undesirable side-effects. (what happens with PerformanceObservers pre-activation, etc)

Closing, but feel free to reopen if new arguments in favor of changing the timeOrigin arise.