This removes both the pl/pgsql version and the shared library version
of the `tile_for_point` SQL function. This function was only used in some
old migrations, and is not required for production usage.
Removing this function simplifies the installation and configuration for
new developers.
These SQL functions are separate from the `tile_for_point` ruby/C function
which is part of the quad_tile gem. This function is still used when
creating and updating database records.
Fixes#3110.
In both the case of primary keys, and also foreign key references,
there's no need to set a default value.
This doesn't have a big impact in routine situations, but can be
very confusing when debugging corner cases.
This is seed data that should reasonably be loaded into every database,
even those that are otherwise empty (e.g. no geo data). Using the seeds
process means that it will usually be loaded by default, and is easier
to find for existing rails developers.
This repo contains code for three functions to be loaded into PostgreSQL as a shared library:
1. `maptile_for_point`, which is used only by the `/changes` API call. This API call is little-used and IMHO should be deprecated and removed. However, even now it's hardly on the hot path for most development activities.
2. `tile_for_point`, which is used only in migrations. At this point, it seems unlikely that anyone will be doing a migration on existing data which would call this function (most developers will be running migrations on an empty database, to set it up).
3. `xid_to_int4`, which is only used for replication using Osmosis and isn't used in the Rails code at all. Hopefully this will be replaced Real Soon Now, but until then it's a quite advanced feature that most developers won't need.
Therefore, this patch proposes to replace the above three shared library functions with SQL implementations of the first two. These are _much_ slower - by a factor of about 30x, however this makes no difference when they're run on a completely empty database. In return, we're able to drop a dependency on the PostgreSQL server development package, and clean a few lines out of the installation instructions.
It's still possible to make and install the shared library functions, and I've included instructions about how to do that - although it shouldn't be necessary for the vast majority of `openstreetmap-website` developers.
Instead, we define models in the migrations themselves. This ensures that
column names etc match the state of the database during the migration,
not the current live version of the model.
