Universal IntersectionObserver polyfill

[dvoytenko]

Partial for: #27807

This is a two-stage polyfill. It uses the asynchronous nature of IntersectionObserver API:

1. v0.js creates a minimal stub. It only tracks requested intersection observers, but doesn't do any work.
2. v0.js requests the polyfill via a dedicated extension
3. Once the extension is loaded, it upgrades the stub to a full-featured polyfill.

TODO:

• Blocked by Polyfill: support same-origin iframe targets w3c/IntersectionObserver#419
• Tests for stub/upgrade logic
• Actual polyfill
• Polyfill for FIE
• Resolve how to skip the first version from inabox to simplify the initial submissions
• Module/nomodule instructions if needed.

[kristoferbaxter]

For the module build, can we ensure this polyfill is only loaded if the user-agent doesn't support IntersectionObserver and 'isIntersecting' in window.IntersectionObserverEntry.prototype?

[jridgewell]

We're loading the polyfill if needed in both modes. Note that module builds need it, too: https://prateekbh.lib.id/caniesmodule/?feature=intersectionobserver

[kristoferbaxter]

The vast majority of module browsers have support for IntersectionObserver, late loading it for a few small holdouts doesn't seem like a stretch.

[jridgewell]

That's what we're doing, for both modes.

[kristoferbaxter]

Missed that when reviewing, thanks for the confirmation.

[dvoytenko]

For the module build, can we ensure this polyfill is only loaded if the user-agent doesn't support IntersectionObserver and 'isIntersecting' in window.IntersectionObserverEntry.prototype?

IMHO, isIntersecting polyfill is so tiny, it'd probably be best to keep it in v0.js, nomodule if we can be certain module supports it everywhere.

[dvoytenko]

Heads up. This is slightly blocked by w3c/IntersectionObserver#419. But review is in progress and I expect it to be completed soon.

[dvoytenko]

@erwinmombay @jridgewell PTAL, I added the actual polyfill, including rewriting the installer for controlled installation.

[dvoytenko]

@jridgewell ptal again.

[dvoytenko]

Upgraded to polyfill 0.8.0 that supports same-origin iframes (FIE).

[dvoytenko]

All. This is now ready for review.

[dvoytenko]

@zhouyx PTAL again.

[rcebulko]

@danielrozenberg Do I just approve this? Shouldn't Dima have approval ability? I can't even tell why it's complaining, it looks like the check ran, but I can't tell what change it doesn't like

[danielrozenberg]

@rcebulko it's possible that you were added for a bundle-size bot failure on a previous commit, and the force-push caused the check to run again, successfully this time. The PR author can't approve their own bundle-size change though, so someone else from the two listed teams needs to approve this PR again

[dvoytenko]

@rcebulko I'll take you off the review here. I think the reason you were added has been already resolved.

[dreamofabear]

Just some readability suggestions. Curious to try this with "intersect-resources" experiment.

[dreamofabear]

Does this need to be a separate file from src/polyfills/intersection-observer.js? Following the code flow between the three files is a bit confusing.

[dvoytenko]

It is a bit confusing, I agree. I did this because otherwise src/polyfills are removed from compiler in many cases. I'd have to instead add a lot of overrides for js and mjs. And I figured that would be more dangerous rather than having a clear line between stubs and polyfill installers. But lmk if you think it'd still be better to live with exceptions.

[dreamofabear]

I think the word "install" is used to mean "stub", "fetch" and "upgrade" in different places which is tough for readability. Can we call this stub?

[dvoytenko]

I renamed all others to upgrade per your suggestions above. But I'd rather keep calling this one install() b/c I think this hides the fact of stubbing from most of places as an implementation choice. But I also redid the "fetch" part as well to reduce number of entities. PTAL.

[rsimha]

build-system/ changes LGTM.

[danielrozenberg]

This experiment breaks releases, I created #27937 to revert the experiment config (none of the other files are being reverted)