PEP0238 lament

[Guido van Rossum]

Duncan Booth <duncan at NOSPAMrcp.co.uk> writes:

> I agree that it would make Python a different language. Neither better nor 
> worse.
> 
> You arguably gain in meeting some peoples expectations of the division 
> operator, but you lose consistency in that previously (as far I can think), 
> the operators +, -, *, /, % when given two arguments of the same numeric 
> type always return a result of that same type. In future we have to be 
> prepared to explain to beginners that division can give a result of a 
> different type: in fact you cannot even predict the type of the result when 
> you know the types of the arguments, you have to know their values as well.

In a farther future, when PEP 228 is fully implemented, Python may
only have *one* (built-in) numeric type, "number".  Then all
operations will defined entirely on the mathematical value, and the
different "types" as we know now are merely details of storage
efficiency -- except that numbers will also have an indication of
"exactness" or "approximateness", to acknowledge the fact that the
number system may choose to give approximate results for some
operations.  We may end up deciding that rationals have too many
problems, and replace binary floating point with decimal floating
point to address a different set of confusions, and so maybe division
of two exact numbers may not always yield an exact result -- just like
the square root of 2 is conventionally calculated only to a limited
precision.  Or we may choose to represent rationals exactly -- in any
case, 1/2==0 will be false, just as PEP 238 proposes.

Such a (hypothetical!) world is just as consistent, yet it does not
guarantee that the type of the result of an operation is the same as
that of the operands, if by type you mean the storage type (int, long,
float etc.).

Also look at PEP 237, which proposes to unify ints and longs.  Would
you really prefer your "consistency" and see 1000000 * 1000000 raise
OverflowError because the result cannot be represented as a 32-bit
int, or would you prefer to drop the "consistency" in favor of getting
a result at all?

--Guido van Rossum (home page: http://www.python.org/~guido/)