Version 76 (modified by jbronn, 12 years ago) (diff)

re-factored out database and model api sections


The GIS branch intends to be a world-class geographic web framework. Our goal is to make it as easy as possible to build GIS web applications and harness the power of spatially enabled data.


  • Place your questions here.
  • Q: When dealing with points (say, degrees) from, do they need to be converted to be useful on the back-end data, assuming -that- data is in degrees? Is it enough to have the same datum and origin? (Reading the intro above is likely to answer the question.)
    • My (JDunck) reading indicates yes. Given the same coordinate system (i.e. datum, origin, and axes), degrees are useful without conversion.
  • Q: Can this implementation work with MySQL spatial-extensions. If not, it's planned?
    • No. It is (now) planned, see Phase 3 below. From the last time I (jbronn) checked, MySQL's spatial capabilities have improved. However, we're going to focus our efforts on PostGIS until things are worked out a bit more. As a spatial database PostGIS it is more standards compliant (OpenGIS consortium), more widely used, and has more features (e.g. coordinate transformation, geometry_columns and spatial_ref_sys tables). It is definitely something I would want to implement in the future since I do like MySQL.
  • Q: Is this going to be a WMS Server/WMS Client/Both? OWSLib is just a WMS Client from what I have seen (from ruckc)
    • WMS Server first, client capability a possibility in certain situations (i.e. you want to cache data from another WMS server). Yes, OWSLib is a client, but it contains code for validating the correct parameters to send to a WMS server, thus it can be adapted into a Django view that validates whether the proper WMS parameters were given. Mapnik has an ogcserver module that can parse the correct parameters for WMS 1.1.1 and 1.3.0, however since it is licensed under the LGPL it cannot be easily incorporated into this branch (unlike OWSLib).
  • Q: Per this discussion in django-developers, "I can do spatial queries if they are directly between two models with polygon fields but I can't seem to get at the spatial queries through foreign keys."
    • Geographic queries require GeoManager, even if the model does not have a geographic field itself (in the case of a foreign key to a geo-field). The reply in the discussion gives detail into why.


Phase 1

  • Create Geometry-enabled fields and manager. Status: complete as of r4788.
  • Allow for Geometry-enabled queries. Status: complete as of r4788.

Phase 2

  • Pending
    • Distance queries, calculations, and related utilities.
    • Utilities for importing vector and raster data (SHP files first) directly into Django models -- will be done with the forthcoming LayerMapping class.
    • Add as much from the PostGIS API as possible.
    • Support for a mapping framework (e.g. Google Maps/Earth, Yahoo Maps, MS Live, etc.)
      • Admin fields and forms (WKT field currently as of r4884, but we want widgets to view and manipulate geographic objects).
  • Complete
    • PostGIS indexing capability.
    • As of r5008, added a GEOS wrapper object for geometry-enabled fields that call directly on GEOS routines (e.g. z.get_poly_geos().area). See Extra Instance Methods section)

Phase 3

  • Support MySQL databases.
  • Geocoding framework.

Design Issues

  • Client JS/Flash framework, i.e., do we want to support OpenLayers, the Google Maps API, the Yahoo API?
    • So far, Google Maps looks the most promising for being supported first (people are familiar with it, and it's more stable than open layers).
    • Yahoo! has a really slick flash interface, I'd like to support this eventually.
    • OpenLayers supports WMS/WFS/tiles as well as Google, MSVE, and Yahoo layers. It is very flexible and open(!!). (Rob Coup)
  • Mapping Framework (generating custom tiles, layers, labels, etc.)
    • Mapnik is modern, but very early on in development and completely lacks documentation. However, the code is elegant and clean, and it was designed for integration with Python -- we're leaning towards this right now.
    • Mapserver has been around for a while, strong backing in the community (e.g. native support in QGIS). Even with documentation, the code looks less inviting than Mapnik (all in C); also has archaic text-based configuration files (pre-dating markup languages).
  • GEOS
    • Update: As of r5008, GeoDjango has its own GEOS interface (via ctypes)
    • GEOS is no longer maintained by Sean Gillies. See Sean Gillies, Geometries for Python (blog post explaining rationale for abandoning GEOS support); see also Sean's message on the GEOS-Devel Mailing List (Mar. 5, 2007).
    • Might consider either using PCL or implement a ctypes wrapper for the routines that we need -- can't really port PCL code here because it is GPL (Django is licensed under BSD).
  • WMS Server
    • I'm not satisfied with any of the current WMS/WFS implementations. One implemented in Django would be desirable, e.g., django.contrib.gis.wms. Thoughts anyone? (OWSLib looks good, see below)


  • PCL (Python Cartographic Library), now part of GIS Python, has done a lot of good work already. Let's apply the DRY principle. Strong opportunities for collaboration with regards to:
    • Mapping framework
    • WMS/WMF Framework -- OWSLib looks excellent for this (BSD licensed and has unit tests!)
    • Utilities
    • Database representation ideas
    • GEOS support, Sean Gilles (lead developer of PCL) looking for help maintaining Python/SWIG interface to GEOS. If SWIG interface no longer maintained, might have to move to PCL for up-to-date GEOS library support.
  • CoordinatesField.
    • Jannis Leidel has already come up with a way to manipulate points in the admin interface, BSD licensed.
  • geopy
    • Brian Beck has written a good foundation for geocoding and distance calculations, BSD licensed.


Geographic Models

Here is an example of how the model API currently works (assume this example is in geo_app/

from django.contrib.gis.db import models

class District(models.Model, models.GeoMixin):
    name = models.CharField(maxlength=35)
    num  = models.IntegerField()
    poly = models.PolygonField()

    objects = models.GeoManager()

class School(models.Model, models.GeoMixin):
    name  = models.CharField(maxlength=35)
    point = models.PointField()

    objects = models.GeoManager()

Notes: The GeoMixin class allows for extra instance methods. By default, a GiST index will be created for the School PointFields fields. This behavior can be turned off by using models.PointField(index=False).

Using syncdb

Use the to invoke syncdb like you normally would:

$ python sqlall geo_app
CREATE TABLE "geo_app_school" (
    "id" serial NOT NULL PRIMARY KEY,
    "name" varchar(35) NOT NULL
CREATE TABLE "geo_app_district" (
    "id" serial NOT NULL PRIMARY KEY,
    "name" varchar(35) NOT NULL,
    "num" integer NOT NULL
SELECT AddGeometryColumn('geo_app_school', 'point', 4326, 'POINT', 2);
CREATE INDEX "geo_app_school_point_id" ON "geo_app_school" USING GIST ( "point" GIST_GEOMETRY_OPS );
SELECT AddGeometryColumn('geo_app_district', 'poly', 4326, 'POLYGON', 2);
CREATE INDEX "geo_app_district_poly_id" ON "geo_app_district" USING GIST ( "poly" GIST_GEOMETRY_OPS );
$ python syncdb geo_app

Note: The geometry columns are created outside of the CREATE TABLE statements by the AddGeometryColumn. This is done according to the OpenGIS specfication. See Open GIS Consortium, Inc., OpenGIS Simple Feature Specification For SQL, Document 99-049 (May 5, 1999), at Ch. 2.3.8 (Geometry Values and Spatial Reference Systems, pg. 39).

Spatial Queries

After a geographic model has been created, the PostGIS additions to the API may be used. Geographic queries are done normally by using filter() and exclude() on geometry-enabled models using geographic lookup types (see the Database API below for lookup types). In the following example, the bbcontains lookup type is used which is the same as the PostGIS && operator. It looks to see if the bounding box of the polygon contains the specific point. The next example uses the PostGIS Contains() function, which calls GEOS library to test if the polygon actually contains the specific point, not just the bounding box.

>>> from geo_app.models import District, School
>>> qs1 = District.objects.filter(poly__bbcontains='POINT(-95.362293 29.756539)') 
>>> qs2 = District.objects.filter(poly__contains='POINT(-95.362293 29.756539)') 

Both spatial queries and normal queries using filter() may be used in the same query. For example, the following query set will only show school districts that have 'Houston' in their name and contain the given point within their polygon boundary:

>>> qs = District.objects.filter(name__contains='Houston').filter(poly__contains='POINT(-95.362293 29.756539)')

Or combine both the bounding box routines (less accurate, fast) with the GEOS routines (most accurate, slower) to get a query that is both fast and accurate:

>>> qs = District.objects.filter(poly__bbcontains='POINT(-95.362293 29.756539)').filter(poly__contains='POINT(-95.362293 29.756539)')


Installation of the GeoDjango module will also require the installation of existing open source geographic libraries and a spatial database (currently only PostGIS). This section will describe the installation process for these libraries. Initially, these instructions will pertain only to a Linux platform (particularly Debian or Ubuntu). Mac & Windows support will be considered later; however, these instructions will most likely work through the Mac shell. Don't hold your breath for Windows support. Community support for prerequisites is better than previously believed, Windows support will come much earlier than expected.

Python & PostgreSQL

  • Python
    • Required: Python 2.4 is required because of heavy use of 2.4 decorator syntax (e.g. @property). The ctypes module needs to be installed as well.
    • Recommended: Python 2.5 is recommended because the ctypes module comes included. Python 2.5.1 is the current latest.
  • PostgreSQL
    • Recommended: PostgreSQL 8.X
    • We are currently using v8.1 of PostgreSQL, and know of no problems with 8.2
    • On Ubuntu Feisty, you'll need the apt packages postgresql-server-dev-8.x (the development headers are needed for PostGIS compilation) and postgresql-8.x.


  • GeoDjango exists in the gis branch from SVN:
    $ svn co django_gis
    $ ln -s django_gis /path/to/site-packages/django


  • Latest GEOS version is 3.0.0RC4
  • Update: As of r5008, you do not need to enable the Python bindings because GeoDjango has its own GEOS ctypes wrapper.
    • ctypes comes standard with Python 2.5. If you run Python 2.4, ctypes may be downloaded here
  • Configure, make, and install.
    $ ./configure
    $ make
    # make install


  • Latest PROJ.4 version is 4.5.0
  • First, download the PROJ datum shifting files. These will come in handy for coordinate transformations when other programs (like Mapserver or Mapnik) are not able to cope with EPSG transformations (I learned the hard way). Untar/unzip these in the nad subdirectory of the PROJ source. For example, if PROJ was unzipped in a directory named proj, then untar these files in proj/nad. Do this before you do the configure/make/install dance.
  • Next, configure, make and install.
    $ ./configure
    $ make
    # make install 


  • Latest PostGIS version is 1.2.1
  • First build & install PostGIS.
    $ ./configure --with-geos --with-proj
    $ make
    # make install
  • Next, create a role and database for your application, and allow it to access PostGIS functionality:
    # su - postgres
    $ psql
    postgres=# CREATE ROLE <user> LOGIN;
    postgres=# \q
    $ createdb -O <user> <db_name>
    $ createlang plpgsql <db_name>
    $ psql -d <db_name> -f /usr/local/share/lwpostgis.sql
    $ psql -d <db_name> -f /usr/local/share/spatial_ref_sys.sql
    $ psql <db_name>
    <db_name>=# GRANT SELECT, UPDATE, INSERT, DELETE ON geometry_columns TO <user>;
    <db_name>=# GRANT SELECT ON spatial_ref_sys TO <user>;

  • Finally, update your to reflect the name and user for the spatially enabled database. So far, we only plan to support the psycopg2 backend, thus: DATABASE_ENGINE='postgresql_psycopg2'.


  • Optional, but highly useful for coordinate transformations and reading/writing both vector (e.g. SHP) and raster (e.g. TIFF) geographic data.
    • For example, the following command will convert your SHP file into WGS84 (standard lat/lon). Then you can import directly into your database using shp2pgsql (utility from PostGIS):
      ogr2ogr -t_srs WGS84 output.shp input.shp
  • Latest GDAL version is 1.4.1. Configure with GEOS then make and install:
    $ ./configure --with-geos
    $ make
    # make install
  • As of r5397 there's a ctypes layer for GDAL/OGR, no python bindings needed.
Back to Top