Hacker News Reader

Comment by yyx | Hacker News Reader

Is there working demo?

Sidewalk Labs (an Alphabet Inc.'s company) has some demos at https://s2.sidewalklabs.com/

The related blog post is at https://medium.com/sidewalk-talk/s2-cells-and-space-filling-...

Here's a blog post I wrote comparing geofencing algorithms one of which leverages s2. https://medium.com/@buckhx/unwinding-uber-s-most-efficient-s...

In practice, we use s2 for in memory location indexing.

Pokemon Go and Foursquare are S2 user

If I'm reading the signs correctly, your working demo is at maps.google.com

Remember a couple weeks (months?) ago when google changed their maps from being a flat Cartesian style to having a more polar spherical look to them as you actually moved away from the surface of the earth? This is the library responsible for that.

S2 has been used inside google for 10+ years.

Honestly, it's one of the best abstractions I've ever used.

Are you sure? I just checked for myself, but (on the google maps version I am served) the south pole is still shows as a black hole when zoomed out far enough that the earth is rendered as a sphere. Also the elongated shape of the tiles near the poles looks an awful lot like a mercator projection.

Pretty sure? I'm just going off a cursory examining of the github repo linked on the website from the OP and a conversation I had in passing with a friend who works on the google maps team.

Also, I'm not a mathematician and everything I know about computer science (and most of math for that matter) I've taught myself or learned on the job. So... maybe I'm using the wrong terminology?

I'm 98% certain the library is the same one currently used on google maps web app.

Coordinate system != map projection.

As far as I know, the map tiles google serves are still in mercator projection -- it's just that mercator was always designed to be projected onto a sphere, not a plane, so now it looks better.

What S2 is, is a better way of representing sets of coordinates within GIS systems, such that searching for features that are close to each other, or within a complex polygon, can be performed much quicker than with other coordinate reference systems ie. floating point latitude/longitude pairs. They're storing points on a 3D sphere (unit vectors) as easily subdivisible 64-bit integers which have a logical order which is much faster to index / iterate over nearby areas.

Makes sense for applications like google maps; "find local businesses near point X,Y" etc.

Yeah I don't actually know if I was using the right terminology. Most of what I wrote in that comment stemmed from me asking a friend 'Hey, so that thing you work on -- looks less flat and more round. What's up with that?'

If I remember correctly, he also mentioned something about the same project being responsible for better stitching and less overlaps/voids in the map.

Thanks for explaining though, helped me understand things a bit better.

The mercator projection is designed to project a ~sphere onto a plane. The distorted tiles are now being redistorted to drape the plane back onto the sphere.

I find that more annoying than useful.

Reply on news.ycombinator.com