From: pgsql-general-owner@postgresql.org [mailto:pgsql-general-owner@postgresql.org] On Behalf Of Guyren Howe
Why schema-on-demand? Can you explain what you mean by that?
Something that is attractive, for beginners or perhaps when prototyping is that you don't have to declare a table. You can just insert tuples into a predicate whose name you provide and they go in and you've defined a relation just by using it.
The problem is which side of the fence you’re writing code. If you write code on the app side, it’s a hassle and every table is a cost. If you write on the SQL side, you can create temp tables with little effort.
So the answer to that one is: in a relational language (like Andl) it’s so easy to create a relation (table) you just do it.
Much of my point in raising this discussion is that there are features that the NoSQL folks are implementing that are useful in some cases. Things they are doing like eventually consistent distributed stores are really required at sufficient scale, but there are other great ideas. SQL's storage model is not the only way, nor should it be. We shouldn't have to abandon the relational model to get such features, but we *do* have to abandon SQL to get them. And good riddance.
Agreed.
I would like to have relational stores providing such features before some monstrosity like Mongo or CouchDB becomes so entrenched we'll never be rid of it.
They do provide some genuinely useful non-relational features, but yes. Friendly relational is what I’m working on.