> You guys call this "simplification"? You're out of your minds.
> This proposal is ridiculously complicated, and yet it still fails
> even to consider adjusting non-numeric parameters. And what about
> things that require more than a trivial arithmetic expression to
> compute? It's not hard at all to imagine wanting log, sqrt, etc.
>
> We do not need to put computational capability into GUC. Any
> computations needed to determine a parameter setting should be done
> by a wizard.
+1 (save the comment speculating about anyone's location relative to
their mind ;)
Additionally, obvious as it may be, there's nothing stopping anyone
from developing a tool to generate the configuration file from a
more "interesting" source. Whether that's XML or some DSL that
supports computations, doesn't matter. I would think if such a tool
showed dramatic merit it would provoke another discussion about core
integration, but ISTM that leaving it dead simple is best.
[mm, hook into the postgres startup script, shouldn't be that hard to
administer..]
