[Qgis-developer] The future of OpenStreetMap plugin

[Martin Dobias]

Hi

recently I have started to think about the future of the OpenStreetMap
plugin we currently ship with QGIS. The plugin (~2K lines of c++ and
~6K lines of python) supports downloading, viewing, editing and
uploading OSM data. My concern is that it has received little to no
updates from the time we have imported it to QGIS tree back in
mid-2009. Given that it is not being maintained and the issue tracker
reveals several major bugs, I have started to wonder how useful it
actually is in its current form and what could we do about it.

There are few other things apart from lack of maintenance that need fixing:
- it has a special provider because data are stored in custom sqlite3 DB
- data access logic is spread through either provider and plugin
- working with tags (aka attributes) is VERY inefficient
- there's no API that could be used for scripting / analyses
- it mixes topology with simple features
- using custom hardcoded old symbology renderer
- some hacks here and there (e.g. messaging from long-running XML
import in provider, "all tags" attribute)

Currently my thoughts are that we should simply remove the plugin
because fixing its problems would need huge amount of time. In
meanwhile I have started to play with an idea of fixing the
architecture with the following ideas:
1. import from XML to database should only import the topological
structure and tags (and relations), without post-processing that would
create simple features representations.
2. instead of having a special provider, we could reuse an existing
provider for viewing OSM layers - SpatiaLite seems to be the most
viable candidate. Once data is imported from XML into a database, it
should be possible to create one or more views (SpatiaLite tables) of
the database (e.g. all polygons or linestrings with "highway" tag)
3. OSM-related code should be in analysis library. providing
c++/python API. In this way, other plugins/scripts could do automatic
processing without having to deal directly with OSM XML or some
database calls. Extraction of data should require just few lines of
code.

I have a working proof of concept (<1K lines of code) and it looks
good - it's fairly simple and fast. For example, import of Essex
county data (2135K nodes, 203K ways, XML size 420MB), including
creation of SpatiaLite layers took less than 3 minutes, the subsequent
map viewing was as fast as any other SpatiaLite data. There's just one
catch (there always is, right?) - it does not support editing and I am
not even considering adding it myself. OSM editor would be a big
project and it would take ages to get at least close to the editing
experience of specialized editors such as Potlatch or JOSM. And
frankly, is there anyone using QGIS OSM plugin for editing?

What are your thoughts and wishes on QGIS - OpenStreetMap integration?

Regards
Martin