On Wed, Dec 28, 2016 at 2:25 PM, Jim Nasby <Jim.Nasby@bluetreble.com> wrote:
> That's my whole point of why this needs to be settable at a global level: so
> that people with a lot of legacy code can set the OLD behavior at a global
> level, and deal with the old code over time.
This has the same problem being discussed nearby on the case-folding
thread, though: any extension or third-party tool has to either work
with every possible value, or else it has to require one particular
value and therefore not be usable if you need another value for some
other reason.
Now, that's not to say we should never break backward compatibility.
Sometimes we should. I think the problem with PL/pgsql is that many
of the compatibility breaks that people want are likely to lead to
subtle misbehavior rather than outright failure, or are not easy to
spot via a cursory look ("hmm, could that SELECT query ever return
more than one row?"). Also, while everybody agrees that a bunch of
things should be changed and improved, not everybody agrees about
which ones, and sometimes person A desperately wants X changed while
person B desperately wants it changed in the other direction or left
alone. If there were a set of changes that we could make all at once,
call the result plpgsql2 or nplpgsql or whatever, and make everybody
happy, that'd be fabulous, but we don't. So we're left with doing
nothing, or having 2^n language variants controlled by GUCs or
pragmas, neither of which is appealing.
I think it would be a good idea to lock all the people who really care
about PL/pgsql in a room until they agree on what changes should be
made for the next version of the language. If they don't agree
quickly enough, we can resort to the techniques described in
https://en.wikipedia.org/wiki/Papal_election,_1268%E2%80%9371
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company