Re: Should we get rid of custom_variable_classes altogether?

On Mon, Oct 3, 2011 at 12:25 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Mon, Oct 3, 2011 at 11:16 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> Robert Haas <robertmhaas@gmail.com> writes:
>>>> Yeah.  custom_variable_classes is a pretty annoying wart, but if it's
>>>> set to the default value (namely, empty) then it actually does prevent
>>>> people from setting bajillions of completely pointless settings, which
>>>> seems like it has some merit.
>
>>> Well, that argument was essentially why we put it in to begin with.
>>> But I think pretty much everybody agrees that it's more trouble than
>>> it's worth (in fact, weren't you one of the people complaining about
>>> it?)
>
>> Well, yes.  But I was arguing that we should replace the leaky dam
>> with one that's watertight, rather than demolishing the dam.
>
> If we had some idea how to do that, I'd probably agree.  But we don't.
> In any case, custom_variable_classes as currently defined is not the
> basis for a solution to that desire, and removing it won't create an
> impediment to solving the problem properly, should we come up with
> a solution.

Yeah, that's why I'm not complaining too loudly.  :-)

> (This is, however, a good reason for continuing to not document that
> you can create random GUC variables --- we might someday shut that
> off again.)

Or maybe better still would be to explicitly document the fact that
behavior in this area should not be relied upon.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company