I'm new to using TileMill and it still seems like a popular cross-platform package, though I keep running into common bugs and errors with it (on Mac OSX). On Mapbox's page, it says that Mapbox Studio is meant to "fully replace and improve upon TileMill in functionality".
When learning Mapbox, the service still refers to using TileMill in its tutorials. So am I right in assuming that Mapbox Studio is the way going forward, it's just that they haven't updated their documentation yet?
TileMill is designed for producing raster tiles. It's powerful and easy to use with CartoCSS and I found it ran well with no bugs (PC version). It's also pretty well documented, with lots of examples and the tiles produced are easy to integrate with Leaflet.js web sites.
Vector tiles potentially offer many advantages, in terms of styling flexibility and new interactivity. But at the moment there are many restrictions in terms of the number of layers, and the amount of geometry you can upload to Mapbox through Mapbox Studio. There is also a real lack of examples at this early stage.
Mapbox's business model is based on them hosting your map data, and there are understandable business reasons for their switch towards vector layers. TileMill layers can be exported as .mbtiles and hosted elsewhere. I do hope TileMill is maintained and updated as it's a great piece of software that Studio will do well to match.
I am a long time user of TileMill which is a great tool. After spending two weeks testing MapBox Studio and I can say it was a huge disappointment.
1) It is so buggy that it is practically unusable, both on Windows and Mac versions (and especially on Windows), the program literally crashes every 2 min.
2) The interface is less intuitive and practical than TileMill. It tooks me a while just to understand how to import a simple shape. With Postgis, you will have to re-enter the entire connection parameters for EACH table you wish to add, this is crazy.
3) You can export your shapefiles to a vector tiles db, but then you have to upload it on mapbox server if you wish to style the tiles. This is the biggest concern at the moment as the available space on your mapbox account is ridiculously low and so expensive ($499/month for 30Gb, are you serious ??). I made a try to export my project in vector tiles, the size was about 300Gb, so am I supposed to pay $5000/month just to work with my own data ?
4) Vector tiles are not safe. As they contain vector information, anyone can easily process them to rebuild your original shapefile. Not a big deal if you are using open data like OpenStreetMap but a way too risky if you work with valuable proprietary data.
So I will continue to use TileMill. Unfortunately the program has not been updated for 2 years now, and likely won't be anymore as it is not compatible with MapBox current business model.
You may be interested in Kosmtik, which is an alpha project with similar aims to TileMill and under active development.
This is an interesting question which I've been asking myself every month or two for the last year or so. It's not straightforward. Some things to consider:
Mapbox Studio only does vector tiles, TileMill only does raster tiles (and static images). There is a raster tile endpoint for Mapbox Studio now: https://www.mapbox.com/api-documentation/#static
Support for vector tiles in existing GIS packages is still pretty basic. And of course you have to figure out how you want to style them, if you're not using Mapbox-GL-JS.
With Mapbox Studio, you can't really have a private data source with public maps - the data itself will essentially be exposed to the public. That's bad if your data is somehow private, and it's bad if you can't afford to host it all on Mapbox's servers.
TileMill used Millstone to fetch web assets, so you could do: polygon-fill-file: url('http://...'). Mapbox Studio is strictly local only, and they have no plans to change that.
It still seems to be pretty tricky to host your own vector tiles. PGRestAPI does it (apparently, I haven't tried), and I think somewhere in the Mapbox repos is the bit you need, but it doesn't seem that simple to me.
Maybe I'm doing it wrong, but I find it very fiddly to build data sources with MS. The Tilemill approach definitely had its drawbacks, but it was very easy to add and remove layers. The PostGIS support feels clumsier (maybe just the lack of the bookmarking system...)
Overall, it seems harder to work independently from mapbox.com. Obviously that's a totally legitimate business decision.
Mapbox's Dane Springmeyer did a great talk at FOSS4G explaining both vector tiles and mapbox studio.
I was struggling with the tiled png paradigm. My project was pretty big and I ran into multi-week rendering times. I had to learn how to make my database faster. The size of my tiles were in the gigabytes. How was I to get the data to my customers? Would I expect them to reserve 4 gigs for map data on their phones? I had to cut off zoom levels, because the file size/render times was growing exponentially.
The key idea to resolving this problem and vector tiles is that style and data should be separated-- more like a web page where you deliver text and a way to style the text, not deliver a pixel-by-pixel rendered page.
Mapbox Studio has the two-pronged style/data divide. The program can change your shape file/postigs/etc into a mbtile vector data. You can upload that to mapbox or store locally. Since Mapbox already hosts alot of data and your own data will be much smaller than rendered png tiles, you very likely will be able to host it all on the free (100 mb) account. Mapbox provides hillshade, contours, osm data-- all the stuff that was taken up so much time and storage for my project. No need to reinvent the wheel.
Vector tiles are new so the methods to get your tiles onto webpages/iOS/android probably aren't as easy as linking to a leaflet script and you'll run into some frustrating bugs. For example, Mapbox Studio loves to repeatedly label a linestring running through several tiles, because each tile represents an independent fiefdom (There is a hacky approach through the use of buffers to counter-act this, but it doesn't seem to work for me.) Furthermore, Studio has some serious bugs; it doesn't work well with Postgis on my computer (I found converting to shapefile before uploading to Studio got around this.) Overall, it is an very usable app that has saved time compared to the tilemill approach.
Even though Studio is tied with Mapbox online, I haven't found a way in which non-paying customers were denied functionality. You can't export png tiles, but that's outside of the scope of the app.
So to answer your question, Yes, try switching to Studio. Give it a bit of time to get used to the new approach.
