nan value inconsistency

rec:=record
string id;
real val;
end;
ds:=dataset([

,

],rec);

ds2:=project(ds,transform(recordof(ds),self.val:=left.val/power(left.val,0.5),self:=left));
output(ds2);

The val for -2 outputs -nan when used in a dataset.

Alternately,
//real val2:=-2/power(-2,0.5);
//output(val2);
But when the expression is used as a definition it gives compile time error.
Consistency is missing and a macro like isNaN is not present to check for nan.