[Map] Improve web performances by not calling "createInfoWindow" if unnecessary

[Kocal]

Q	A
Bug fix?	no
New feature?	yes
Issues	Fix #...
License	MIT

This afternoon we spoke a bit with @simondaigre, and he told me about InfoWindow performance issues when you have too many of them (and depending your device).

If we have 1.000 InfoWindow, hidden by default and only visible after clicking on a marker, why should be only create the InfoWindow at that time?

This PR introduce lazy-creation of InfoWindow, it should improve web-performances on the user's browsers.

I've used the following PHP code:

       $map = (new Map())
            ->center(new Point(0, 0))
            ->zoom(2)
            ->fitBoundsToMarkers()
            ->options(new GoogleOptions(
                mapId: 'XXX',
            ))
        ;

        for ($i = 0; $i < 1_000; $i++) {
            $map->addMarker(
                new Marker(
                    position: new Point(random_int(-90, 90), random_int(-180, 180)),
                    title: 'Marker ' . $i,
                    infoWindow: new InfoWindow(
                        content: 'Info window ' . $i,
                        opened: $i % 499 === 0
                    )
                )
            );
        }

Benchmarks

I've did some benchmarks for both Google and Leaflet:

• with 1.000 markers,
• on a Macbook M1 Pro, CPU Throttle x6

Google

Before	After

connect() takes 936ms instead of ~1,80s, we won ~900ms.

Leaflet

Before	After

connect() takes ~1s instead of ~1,20s, we won ~200ms

---

The next steps for performance optimizations are:

• Lazyloading markers and infoWindows, in order to keep a good TTFB
• Clusters!

[smnandre]

Does not seem a good idea to run every time.

[Kocal]

What do you mean?

[smnandre]

"createInfoWindow" will be called on every "click" ?

[Kocal]

Ah, yes you're right.

Hum, I may have better ideas to rewrite the PR, but I don't have time right now, let's put it in draft.

[smnandre]

Not sure this should be exposed like this, seems to complexify the contract, while not beeing this "explanatory"

[smnandre]

Confusing to me... this mixes runtime values and definition values

[smnandre]

I feel like the current implementation could be improved, but i don't think we could/should avoid exposing this internal code.. and the loop seems a bit odd.

Why not having separate create / open methods ?