Here are some things that I think it would be wise to pursue going forward:
- Complete SQL/MM support. We already have a good deal of it, to provide compatibility with SDE (should SDE wish to provide spatial SQL access to PostGIS). We will need to complete curve support to get substantially closer.
- Clean up function names, deprecate old OGC function names and move to using XX_ prefixes to define where the functions come from. ST_ for SQL/MM, SE_ for SDE-specific, PGIS_ for PostGIS specific, etc.
- Automagic indexes. Wrap all the ST_ functions in index magic so that Contains(A,B) kicks in the index automatically.
- Automagic geographic support. More controversially, recognize when simple spatial ops are being called on lat/lon features and apply appropriate translations to do them in a planar system, or on a sphere/spheroid. The danger here is that in pleasing the crowd, we will quietly confuse a small number of people for whom the automagic assumptions are wrong.
- Update documentation. PostGIS is now far in advance of its last serious documentation update. The reference material is all correct, but the tutorial level stuff needs to be updated. All the examples work, but they are not the "right" way to do things anymore. Particularly important if we start deprecating old function names.


1 comments:
Automatic projection to perform spatial operations is going to open a can of worms imho... this can be done only in a small area, then you also have to find out a good projection for that area (EPSG database has definition area for projections in fact).
Doing things directly in lat/lon is hard, but not impossible.
Distance and buffer related operations are in fact hard, I guess you could use spheric geometry there. I asked a university professor about the formulas needed to compute point/line distance in spheric geometry, and I'll hopefully get an answer soon. Straight point to point distance is easy, and well as computing the end point of a line given starting pont and bearing. Once you get those three, you can implement buffer and dwithin afaik (did not check in detail thought).
Intersections could be performed directly in lat/lon space, with a catch when working along the anti-meridian of Greenwich (what do you do when a geometry crosses that one... hmmm... not sure).
Post a Comment