On Wed, Sep 17, 2014 at 11:08 AM, Peter Eisentraut <peter_e@gmx.net> wrote:
> I also wrote PostGIS dependent libraries, not PostGIS itself. If you
> are comparing RHEL 5 and 6, as you wrote elsewhere, then some of those
> will most likely be different. (Heck, glibc could be different. Is
> glibc never allowed to fix insufficiencies in its floating-point
> implementation, for example?)
The operator class author has a responsibility to make sure that
doesn't happen. If he or she should fail, then it's a bug, and
possibly a failure of imagination on their part. This is the only way
of thinking about it that makes sense. If you want to use a library
feature in your opclass B-Tree support function 1, then you'd better
be damned sure that it implies immutability insofar as that's
possible. Sure, it's also possible that your users could be the victim
on an unfortunate upstream bug that you couldn't reasonably predict,
but when is that not true?
In general, I am totally unconvinced by this line of argument. It
implies that everyone has to be an expert on everything just to use
Postgres.
--
Peter Geoghegan