Experimental Soft Navigation support

[tunetheweb]

Warning: This branch is likey to stay in draft for some time, and not to be merged into main until the Soft Navs experiment completes

Fixes #119

Implements experimental support of Soft Navigations to the library:

• - Updated README
• - TTFB
• - FCP (not currently emitted by Chrome at this time).
• - LCP
• - CLS
• - INP
• - Attribution (particularly navigationTiming)
• - Test Cases (needs to wait for a supporting version of Chrome to be available in WDIO)

[mmocny]

Hey, this is excellent work. Quite a bit to get right, thanks for the effort!

I only review the CLS changes, and I think there's a potentially cleaner way to do it. I didn't review the specifics of the other metrics since I the change would alter a lot there too, so wanted to discuss that first.

Basically:

• I think the observe() helper, which all metrics use, should be the only place where we look at navigationID and split entries whenever it changes. See: https://github.com/GoogleChrome/web-vitals/blob/main/src/lib/observe.ts#L52
• Whenever navigationID changes (or, the first time for initial nav), that helper can look up pageURL, etc, that way we have shared cache for all metrics
• Each metric implementation right now already uses helpers like onFCP, onHidden, onBFCacheRestore... in order to "finalize" and/or "reset" its metrics. I think we could have another helper like onNavigate(), which can be used to flush/reset metrics.

One tricky bit may be that the onNavigate() helper needs to be called before any new Entry that uses that navigationID, and you could have entries with different navigationID in a single PerformanceObserver callbacks, especially when using buffered entries.

That said, I think all this code made be fragile to such cases and needs careful testing.

Keep at it!

[tunetheweb]

That’s for the review @mmocny ! I’ve cleaned it up quite a bit further so it’s a bit smaller now. Have another look if you get a chance and see what you think now.

[alekseykulikov]

Hey! I'm super excited about this work. Could you republish the branch?
The current 3.1.1-soft-navs doesn't report LCP and instead prints BARRY.

[tunetheweb]

Well that's embarrassing! Republished to the soft-navs tag (which is an alias to 3.1.1-soft-navs-1 now).

[alekseykulikov]

Thanks, that was fast.
Maybe, I'm doing something wrong, but there's still no LCP report, despite actual values are available (Chrome 110):

[karreiro]

Hi @tunetheweb,

Thanks a lot for introducing support for Soft Navigations in the library!

While exploring the new features, I noticed something interesting. It seems that not all changes are being reported when reportAllChanges is activated, as the onINP module is only tracking the longest interactions.

To handle that scenario, I've slightly updated the onINP module (#411) to ensure that all changes are reported when reportAllChanges is turned on (demo). I've directed my PR to your branch, as my changes are somewhat related to soft navs. I'd be glad to hear your thoughts on this.

[tunetheweb]

As discussed in #411 that is working as intended. But I raised #413 to make it clearer what exactly that intention is.

[tunetheweb]

Updated this branch to v4 code.