Watching for 'scroll' event may end LCP before a valid browser entry is created about web-vitals HOT 6 CLOSED

Awesome, thanks for your work @philipwalton!

from web-vitals.

Interesting, @npm1 can you speak to Chrome's kUserScroll and kCompositorScroll usage here? Could there be some scroll events that addEventListener('scroll') would catch but wouldn't end LCP observation?

from web-vitals.

We should stop on scrolling caused by user, which is not the case here. Does the listener check that the event has isTrusted set?

from web-vitals.

Although window.scroll() does create a trusted event despite not being caused by the user so that seems insufficient. @flackr do you know if there's a reliable way to detect user-initiated scrolls?

from web-vitals.

Ping @flackr

Unless we find a better way to detect only user-initiated scrolls, we'll likely have to drop this check and only listen for keypress and click. Ignoring scroll events won't affect the reliability of the results, but it will likely affect how soon tools like the Web Vitals Chrome Extension can report the metric as "final" (cc: @addyosmani)

from web-vitals.

There is currently no reliable way to detect user-initiated scrolls. This is because the scroll event itself is an observation of the scroll position having change rather than a direct event causing it.

You could however try to observe all possible user initiated scroll sources (mousewheel, pointerdown / pointermove, down arrow / page down / space bar?) in combination with a scrollable page.

Could you explain more the rationale around no longer observing LCP after the user scrolls? This doesn't seem like it would be guaranteed IMO.

from web-vitals.