postgis/MIGRATION

PostGIS 2.0 Migration Guide
===========================

PostGIS 2.0 introduces changes that may affect backward compatibility
for some applications. This document lists those changes.


geometry_columns
----------------

In PostGIS 2.0, the ``geometry_columns`` metadata has been changed from
a table to a view. 

  * For applications that manage ``geometry_columns``
    using the standard AddGeometryColumn() and other methods, there should
    be no change. 

  * Applications that have inserted directly into ``geometry_columns`` 
    will need to either move to the standard functions, or change
    their create table syntax to include type/srid/dimsionality
    information. 

    For example, to specify a 3D feature in WGS84::

      CREATE TABLE mytable (
        id SERIAL PRIMARY KEY,
        g Geometry(PointZ,4326)
      );

  * Applications reading metadata from ``geometry_columns`` should expect
    that the geometry type strings will be mixed case and include the
    dimensionality modifiers. For example: PointZM, MultiLineStringZ.
    
  * If you have views built on tables that were built
    the old constraint way or that use processing
    functions to return new geometries, they will not register correctly 
    in geometry_columns.  They will all appear as Geometry with srid=0.
    
  	In order to fix this, you have to cast the geometries in the view.
  	So for example, If you had a view that looked like this:
  	CREATE OR REPLACE VIEW vwparcels AS
  		SELECT gid, parcel_id, 
  			ST_Transform(geom, 26986) As geom_26986, -- a transformed geometry
  			geom, --one that used constraints before
  			ST_Centroid(geom) As cent_geom -- one that used processing function
  		FROM parcels;
  		
  	You will need to drop and recreate the view. 
  	Sadly CREATE OR REPLACE will not work since the casting
  	makes geometries be treated as a different type as far as CREATE OR REPLACE 
  	is concerned.
  	
  	NOTE for this, it might help to lookup in geometry_columns 
  	    to see the raw table srid/type
  	Alternatively you can convert some of your table columns to typmod, 
  	then you would only need to CAST when doing processing.  This however still
  	requires you rebuild your dependent views
  	DROP VIEW vwparcels;
  	CREATE OR REPLACE VIEW vwparcels AS
  		SELECT gid, parcel_id, 
  		   ST_Transform(geom, 26986)::geometry(MultiPolygon,26986) 
  		      As geom_26986,
  			geom::geometry(MultiPolygon, 2249) As geom,
  			ST_Centroid(geom)::geometry(Point,2249) As cent_geom
  		FROM parcels;

    
Restoring data from prior versions -- GOTCHAS
-----------------------------------------------
The populate_geometry_columns() function in PostGIS 1.5 and below, 
used the deprecated functions ndims() and srid() for defining constraints.  
As a result these tables will not
restore into a fresh PostGIS 2.0 database unless you

1) Drop these contraints before you restore the data

or 

2) Install the legacy functions srid() and ndims() found in the legacy.sql file
before you try to restore these tables.

Function is not unique after restore
-------------------------------------
After you restore old data into a fresh PostGIS 2.0 database, 
YOU MUST install the uninstall_legacy.sql
file after the restore.  If you don't you will run into 
"function is not unique" errors in your applications.
The other issue with having the old functions is the old functions you restore
will be bound to the old postgis library which is incompatible with the
new postgis geometry structures.

If you still require legacy functions, you can install the legacy.sql file 
after the uninstall_legacy.sql.

GOTCHA with uninstall_legacy.sql
--------------------------------
The uninstall_legacy.sql will not be able to uninstall functions and will 
output errors detailing objects using these functions that are currently 
being used in views and sql functions.

If you run it in PgAdmin, the whole script will rollback not 
uninstalling anything. So you should run it in PSQL mode.
If you get errors in uninstall_legacy.sql -- convert those views and sql
functions to use the newer equivalent functions 
and then rerun the script again.

BENCHMARK NOTES
------------------------------
Querying a whole geometry_columns table of 200-240 someodd records 
(tables/views all constraint based)( On windows 9.1beta2)
Speed for this set of data I'm okay with though a bit slower
I think the main issue why fancy is slower is that to get the fancy name
I need to do a double call to get dims since fancy_name needs dims as well
This I expect to be much faster using typ_mod
SELECT * FROM geometry_columns_v; -- 2063 ms, 1063 ms (using fancy name)
SELECT * FROM geometry_columns_v2; -- 656 ms (this is not using fancy name) 

SELECT * FROM geometry_columns; -- original static table 31ms - 150 ms

-- Doing single selects of a table
-- constraint based one -- 76ms, 46ms
SELECT * FROM geometry_columns_v where f_table_name = 'streets';

-- One I converted to typmod -- 76 ms, 48 ms (first call was slower)
SELECT * FROM geometry_columns_v where f_table_name = 'buildings_1995';

--Filter query for schema --
-- these are constraint based (7 rows) -- 76 ms, 240 ms, 68 ms
SELECT * FROM geometry_columns_v where f_table_schema = 'tiger';

-- CONVERTING TO Typmod as I feared is slow for tables with data.  
   I think it follows the same model 
-- as converting varchar(100) to text or varchar(100) to varchar(300) etc.
-- This maybe can be remedied somehow in 9.1 since it hasn't been released
-- (haven't tried doing this exercise in 8.4, 9.0)
-- This took  36,018 ms for table of 350,572 records.  
I assume it will be linear with number of records.

ALTER TABLE buildings_1995 
    ALTER COLUMN the_geom TYPE geometry(MultiPolygon,2249);