> TileMill has shifted to an open open source contributor model and moved to its own organization, tilemill-project.
A shift which has come a year or so after the project was abandoned by Mapbox in favor of Mapbox Studio (the official, supported successor to TileMill).
TileMill alone is still useful for creating one's own raster tile set, as long as the application remains in working order.
However, last I checked it gave me tremendous issues running on Linux as many dependencies had become outdated.
Although, I haven't checked back in a while since I moved to Mapbox Studio.
MapBox hired the maintainer of maputnik though, so the future is unclear.
(Not disagreeing with your point.)
I was a lot of fun to work with, CartoCSS allowed me to be immediately productive with it. In the end though it was too expensive to serve tiles from AWS.
It is no longer the case, modern computers and browsers have acceptable performance when using vectorial maps even for very complex maps. The need for firing your own tile server is very unusual.
The most glaring difference is that label placement in common vector renderers is generally "try once, fail if no room", whereas in Mapnik (standard raster renderer), for example, you can supply a list of positions to try in order of preference.
There's a lot to like about client-side rendering (which is why I wrote https://github.com/systemed/tilemaker/ to generate vector tiles) but it's not up to feature parity with raster rendering yet, and for some of us those features are important.
In fact most of the time I can get way with just a vector layer over a 3rd party tile layer using Leaflet and some fancy fireworks showing labels on click/hover events.
I'm really glad I don't need to render tiles in the server side anymore but I remember the time when I had not choice but manage my own tile server simply because many clients could not handle something simple like a choropleth map with all the voting districts in my state.
Also, map data is getting progressively richer and more expensive to render, particularly when adding 3D data, so I'm not convinced that offloading everything to the client is still a 90% use case.
