PostGIS 2.0.0 has inched a lot closer to completion. This past week, Paul enabled his gserialization work which changed the on disk-format of PostGIS and in return I think we'll have a much better platform to grow on. With this change we now have the 3D index and bounding box bindings in place. Say hello to the &&& operator which is like &&, but is 3D aware and comes with its own companion 3D spatial indexes. This will allow you to do true 3D bounding box searches with any of the new 2.5/3D geometries we have in place for PostGIS 2.0.0. We are still noodling out the semantics of boxes. Read Paul's call for action on the The Box Plan?, if you are interested. PostgreSQL 8.4 is the lowest supported version for PostGIS 2.0.0. It took a bit of squabbling between PSC members to make that decision, but I put my foot down and I think in the end was for the best to allow us to use new features, less platforms to test, and get rid of some unnecessary code.
PostGIS Windows 32-bit Experimental builds fresh off the presses
With all these changes, if you are running an earlier alpha release of PostGIS 2.0.0, you'll need to do a dump restore since the on disk format is now changed.
If you are on windows and want to give some of this all a test drive, you can download one of our PostGIS 2.0.0 Windows experimental builds. We still only have 32-bit builds. We have builds
for PostgreSQL 8.4, PostgreSQL 9.0, and PostgreSQL 9.1 beta 2. The problems we faced in PostgreSQL 9.1 beta 1 were resolved in beta 2 so that most regress tests past except some minor ones involving stupid things like difference in line number marking of errors. Complement your PostgreSQL 9.1 beta 2 meal with a yummy large helping of PostGIS 2.0.0 goodness.
PostGIS team has been a very exciting and fast-paced team to work on; particularly these past couple of months. I feel like a kid in a candy factory trying out all these new great features as soon as they come out of the oven and thinking of the workflows they can improve or make possible.
On the raster front, we have now ability to reproject rasters using transform function similar to what we have for geometry as well as various analytic like histograms, summary stats, and value counts. We also have the ability to export rasters to various image formats. Plus more to come before release. Bborie has been a great team player thus far with getting all these things done, and plans to do more before release such as ability to rasterize geometries. I've still got a ways to go to finish off the Raster section of the PostGIS documentation to hold up my end of the bargain.
From what I have seen and heard thus far from others using other tools and thinking about using PostGIS raster support to replace or supplement, we've already got many raster features unheard of in any other spatial database including Oracle Spatial. There are still some raster processes people commonly do in R that are either faster in R or not yet available in PostGIS, but even that is narrowing and R has the disadvantage of not being quite as seamless when trying to perform operations that involve both rasters and vectors. Same I think can be said about ESRI raster support, though I know much less about that animal. Nothing beats the slickness and brevity for flipping rasters and vector geometries like SQL. So hats off to all those who have worked on PostGIS Raster-- Pierre, Jorge, Bborie, Strk,Mat, David
Sandro Santilli (strk) has been diligently working on Topology support to make our PostGIS topology more inline with SQL-MM standards. Andrea Peri has also been helping out a great deal with testing and providing some patches. I still hope to create a loader for TIGER data (which in itself is a topological model) very close in structure to PostGIS topology. Both systems are composed of edges, faces, and nodes as the basic building blocks. This will serve as both a good learning tool for topology, a good way to stress test the system with real world data, as well as providing useful data that people would want to work with anyway. Not sure if I will get to that before release, so might come as a separate add-on.
- Tiger Geocoder Finally got the loader fully functional on Linux/Unix I think and I have been surprised how many people have already started using it and posting bug reports and patches. We are still working on speed improvements that we are doing via our consulting work and will contribute back in the code base. It still intentionally works fine with PostGIS 1.5 since we are improving on it for clients currently using PostGIS 1.5. However some of the new changes we have made require 8.4 or higher.
- Much improved 3D Support - as mentioned before we have new 3D indexes, operators, functions and types (TINS, POLYHEDRAL SURFACES etc). Though I don't show it , this is the part that I am most excited about. I'm still fleshing out the ST_ASX3D - making enhancements and hope to release a tutorial showing how to create a simple web-based 3D query tool with x3dom and PostGIS.
- Shape file dumper /loaders get some love We now have transformation support in the shape file loader, multi-file import for the gui shapefile loader and some other goodies.
PostGIS 64-bit builds -- when are they coming?
First I must say aside from a few people who have whined, I don't get the sense that people care too much either way with running PostGIS under 64-bit PostgreSQL windows install.
Our interest is mostly out of curiosity if it will work better than it does under 32-bit Windows. Despite most people's snide remarks about running PostgreSQL on windows,
it works fairly well for many of our needs and the PostGIS speed is for the most part on par with the Linux speed. There are still many suites that only work on windows and where having both a Linux box and windows box to maintain is not worth the hassle for the marginal difference in speed.
Generating 64-bit builds for PostGIS is proving to be a bit more problematic than we had hoped. I guess you can say that about most things in life.
We first started with an attempt at VS VC++ build and quickly gave up on that after running into too many road blocks. It also doesn't help that the process
would be much more different than the PostGIS Unix like build structure in place which would mean we would be constantly playing catchup with keeping our VS in synch with the configure scripts etc. used by PostGIS.
Aside from that I don't particularly care much for the VS development environment since it feels a bit too heavy for my liking. I use it for ASP.NET development for its intellisense which I do just as well with Visual Web Developer Express in and never need to compile anything in it. So the whole compilation side and the idea of solutions etc seems pretty alien to me.
This brings us to Msys64 which is the mode we chose to take because it seemed least painful and also allows us to be able to work with the same tool chain that
the rest of the PostGIS team works with. Hats off to Andrew Dunstan. Thanks to him PostgreSQL 9.1 just compiled out of the box with Msys64, so that worked fine and was an easy check-off. The proj library that PostGIS uses for transformation, also more or less compiled using some minor hacks provided on the bug list provided by Paul Ramsey and others. GEOS sadly a key component, we got far into the compile process before it started coughing up blood. Details of our steps are documented in Compiling using MingGW-w64 for both 32-bit and 64-bit. If anyone wants to give it a stab and cares to please do so. At this point we are tired of the whole thing and will retire our efforts for now since we've got more rewarding things to do with our time. GDAL we've had issues with, but didn't try again to see how insurmountable they are. The rest of the items in PostGIS tool chain aside from PostGIS itself, are pretty common variety that we shouldn't run into any obstacles with those.