Real world database schemas contain a lot of duplicated data.
Some column names like `id`, `created_at` etc can easily be repeated
hundreds of times. Same for SqlTypeMetada, most database will contain
only a limited number of possible combinations.
This result in a lot of wasted memory.
The idea here is to make these data sctructures immutable, use a registry
to substitute similar instances with pre-existing ones.
These new methods are used from the Active Record model layer to
determine which relations are viable to back a model. These new methods
allow us to change `conn.tables` in the future to only return tables and
no views. Same for `conn.table_exists?`.
The goal is to provide the following introspection methods on the
connection:
* `tables`
* `table_exists?`
* `views`
* `view_exists?`
* `data_sources` (views + tables)
* `data_source_exists?` (views + tables)
Making this change revealed several subtle bugs related to models with
no primary key, and anonymous classes. These have been fixed as well,
with regression tests added.
ActiveRecord json/xml serialization should use as base
serializable_hash, provided by ActiveModel. Add some more coverage
around options :only and :except for both json and xml serialization.
Previously we would just silently write the attribute. This can lead to
subtle bugs (for example, see the change in AutosaveAssociation where a
through association would wrongly gain an attribute.
Also, ensuring that we never gain any new attributes after
initialization will allow me to reduce our dependence on method_missing.
