So a view isn't a real table. Reading Momjian's book, I get the impression of a view as more of a macro, a series of descriptions to the pg engine on how to expand a query. I don't if this is true or not.
I've been drinking from this well that has me using views to denormalize normalized table. Hey d00d, selecting from denormalized tables is so much easier (than thinking.)
If I create a view with a bazillion columns, some computed (e.g. (substr(a, 1, 4) || substr(b, 1, 4)) as passwd), and then I don't use those computed colums in a specific select, are those column values computed anyway?
Apart from use in a computed column used in a where statement, does selection from a view cause any copying to be performed?
Is a faster/better/cheaper to use a view of 100 columns in 10 queries where each query uses only 10 of the columns, or is it more performant (cough) to break that view up into smaller views?