Re: cascading column drop to index predicates

Andreas Pflug <pgadmin(at)pse-consulting(dot)de> writes:
> In that sample mentioned the index might be used mostly with a,b
> columns. Dropping the index silently might damage the application
> because it relies on an (a,b) index to be present. IMHO only Indexes
> that span that single column should be dropped without CASCADE.

That argument makes no sense to me at all. If you drop the *column*
a or b, and do not thereby break your application, how is the
disappearance of the index on it going to break anything? The index
is meaningless without something to index.

I think the question at hand is whether the same logic applies to
partial indexes: if the index's condition is no longer meaningful, is
the index meaningful? I think we can handle both cases the same way.
But clearly an index condition isn't quite the same thing as an index
column, so maybe someone can make a good argument for treating them
differently.

regards, tom lane