10:18 p.m.
Hello,
together with Nakaner, we discussed about vector tiles at the FOSSGIS
2017 conference in Passau, Germany. We came to the conclusion, that
vector tiles have a lot of disadvantages and are far away from
productive use and decided to remove them from OpenRailwayMap.
One major problem is the speed. HTML5 Canvas rendering is still too
slow, that is why OpenRailwayMap never used tiles rendered on client
side in production. WebGL would be the better technology, but has some
disadvantages:
* Writing an own WebGL renderer is too much work for a small project
like this
* There are no WebGL-based renderers that are established, using
standard formats and not from Mapbox. All available solutions are
relatively new and it is not clear how their development will continue
(except from Mapbox Vector Tiles).
* WebGL can be either not supported or slow on older systems and mobile
devices
* Clientside rendering will always be slower than serverside rendering.
The fast map of openstreetmap.org is a cause why some people prefer it
over Google Maps.
The main problem is the data of the vector tiles: Vector tiles are just
fast if you filter and generalize the data, but this reduces the
flexibility and decreases the quality of the data. In our case this is
the main problem that makes it difficult to reduce the amount of data
in the tiles. In low zoom tiles, railway tracks are split up in a lot
of very short ways. Reducing the number of features is only possible by
merging these ways. But merging features is just possible by ignoring
tags. And the decision which tags are more important than other tags
depends on the rendering style.
Also there is a general difficulty with vector tiles: It is not
possible to drop bitmap tile support completely without dropping the
possibility of embedding OpenRailwayMap layers in OsmAnd, QGIS, etc.
That is why we will always need a bitmap tile interface.
The question is how we should continue now:
* Clean up node-tileserver and remove the vector stuff from the code,
moving to node-mapnik for tile rendering in a longer perspective
* Drop node-tileserver completely and move to Mapnik, mod_tile,
renderd, Tirex, ...
In both cases there is the problem that Mapnik (and node-mapnik) do not
support MapCSS styles. Therefore it would be necessary to develop a
converter from MapCSS to Mapnik XML first.
If we would choose the first way we would have to decide if we either
should apply these changes to node-tileserver or create a fork?
Personally I prefer own code over being dependent from other projects,
but on the other hand I do not want to invest much work to reinvent the
wheel.
Regards
Alex