darcy@druid.net (D'Arcy J.M. Cain) writes:
> OK, there are more problems. If you apply the following patch to the
> regression tests you will crash the backend in a number of places.
Yipes!
I must withdraw my prior opinion that we should shoehorn in a repair to
the INET datatypes for this case. It's clear that we have a wideranging
problem that ought to be fixed more globally. But presumably it's
been there for quite a while, and we didn't know it; therefore it's not
critical enough to hold up the release.
My guess is that maybe this should not be fixed in the individual
datatypes at all; instead the generic function and operator code should
be modified so that if any input value is NULL, then NULL is returned as
the result without ever calling the datatype-specific code.
There might be specific operators for which this is not the right
behavior (although none spring to mind immediately). In that case,
I think the best bet would be to have a per-operator flag, defaulting
to OFF, which could be turned on for those specific operators that are
prepared to cope with null inputs.
Thoughts?
regards, tom lane
