> The guideline we've used for implicit coercions is that in
> the numeric type family, only "up conversions" are implicit, that is
> int2->int4->int8->numeric->float4->float8. "Down conversions" are
> marked as assignment-only. Without some such restriction,
> there's nothing to keep the parser from making very
> surprising choices of operators.
I was unaware of the details, yes, but your explanation makes perfect sense.
When I say it's broken though, I'm talking more specifically about trying to
use round(float, int); is there some reason that this function overload
should not exist in the absence of an implicit float->numeric cast?
Rounding a floating-point to a specific number of decimal places is a common
enough operation after all...
Glen Parker