BTW, is there still no online map that would use some better projection (i.e. anything but [Web-]Mercator)? I mean, there is Google Earth, of course, but it has too much visual effects added, to use it as a go-to tool, as I use OSM or Google Maps. But mercator makes large scale distance and area comparison absolutely unintelligible, and I would rather much prefer to be able to use Kavrayskiy VII/Natural Earth or something like that with OSM. No way it could be too computationally expensive in 2020, right? (I mean, once again, we do have Google Earth.)
I don't think there is an area-preserving projection that can work for all the scales online map services are designed for.
If you are creating your own global-extent map at a limited range of scales, most of the OSM tooling for example PostGIS lets you specify a custom projection like http://shadedrelief.com/ee_proj/
If I'm creating my own map I can kinda draw whatever the fuck I want. I need a web-map (preferably OSM-based) that I can use as a web-map, but with a projection less fucked up than Mercator, of which I can name at least 10 in addition to 2 I already named. Also, web-mercator generally is not a problem on small scales, so there isn't really a problem of alternating between projections depending on scale. (Which is not really necessary too, since other projections are generally more or less ok even on small scale. You can totally use globe view on google maps on city-scale and barely notice any difference at all.)
Esri's ArcGIS Online supports over 100 projections, and they're always working on more.
I had the privilege of chatting with one of the developers there who works on their projection engine (which is unmatched in the industry, as far as I know). To contribute to it, one effectively has to be at the graduate level in mathematics, low-level programming (especially numerical methods), and geography at the same time.
I did a project where we produced maps of car dealerships. We clustered close together dealerships. The distance function that I used for measuring "closeness" evolved from a simple Euclidean distance (sqrt(∆x²+∆y²)) to one that used box-shaped neighborhoods (max(∆x,∆y)) to a variation on this last one that took Mercator distortions into account because Alaska. This was one of the most fun projects I worked on.
There was one subtle bug that I had in the web interface where at the US level, clicking on the wrong place in the spash screen loaded, instead of the clustered dealerships at that level, instead added a pin to the map for every single dealership in the country. It took about 10 minutes to render on 2007 hardware but was really pretty when it was done. I no longer have the screenshot unfortunately.
