Echoing Alex's comment - responding to this was on my todo list, but
I've been fighting a head cold for the last few days. On top of that -
we're in the feature development phase, so non-critical bug fixes
aren't taking the highest priority at the moment.

That said: the patch itself looks like a fine implementation of a
given behaviour.

What isn't clear is that the behavior implemented is correct. Malcolm
flagged this in his original comments - NaN and Inf are CPU and
database-dependent values. Postgres can't support Inf, but can support
NaN; I don't know what the situation is with other databases. Inf and
NaN are both Decimal values - unconditionally raising a validation
error doesn't seem like the right approach to me.

So - the patch is fine, as long as we accept the design it implements.
Now you need to convince us that the design is correct.

Yours,
Russ Magee %-)