Merlin Moncure <mmoncure@gmail.com> writes:
> odd that stable function is inlined but immutable isn't!
Well, it knows that the expansion to to_date() would only be stable not
immutable (because to_date depends on some GUC settings), so doing the
expansion could change the behavior, eg by preventing constant-folding.
Although usually wrapping a stable function in an immutable one is a
recipe for disaster, we don't forbid it because there are cases where it
makes sense --- for instance, you might know that the function really is
immutable *in your usage*, and want to use it as an index function or
some such. But the SQL-function wrapper adds a lot of overhead. I
think a plpgsql wrapper would be better here, if you need to cheat about
the mutability.
regards, tom lane