> > Is there a reason why hstore or json is not an option? That may work a lot > > better than this approach. > > I don't want to move away from SQL common features so I can have a > test environment using SQLite and deploy on PostgreSQL. This approach > looks elegant and simple to me. Using a new table per attribute type > or even per attribute for the values, I can easily index them for > quick lookups.
Couple of things to keep in mind: * SQLLite and PostgreSQL are not 100% compatable. It's cheap and easy to set up PostgreSQL for testing/development, so it seems silly (to me) to test on something that might behave differently than the production environment.
* Any setup where the application can execute DDL is a potential security concern. If the code can manipulate tables then a bug in the code can allow an accidental or malicious user to quickly and easily destroy data.
This is a good point I probably should have mentioned specifically. Making my solution secure means restricting access to the catalog and catalog to DDL features to a very small subset of trusted users (ideally not through a standard application connection). In this case, one might as well restrict it to those who are already db superusers unless there is a clear need to broaden it.
BTW, a specific case in point... DDL can't be parameterized since it doesn't have a query plan. This means you are doing string concatenation to create your ddl queries. If you aren't careful someone can add an attribute like:
'; DROP TABLE things; --
In practice this means a great deal of manual review and restriction on which users can access this feature.