On Friday, 17 April 2015 at 13:27:19 UTC, Biotronic wrote:
I've been writing a lot of generic code lately that has to deal
with various kinds of numbers, and have near been driven nuts
by the fact there is no uniform way to get a zero or one.
Consider:
void foo(T)(T a) {}
foo!T(0);
Does this work with all built-in numeric types? Yes.
Does it work with T=BigInt or Complex!float? No.
Now, those are a limited set of possibilities, and one could
easily enough create a template such that
foo!BigInt(zero!BigInt);
would work. But why can't I instead, for every numeric type,
simply write
foo(BigInt.zero);
foo(float.one);
foo(Complex!float.zero);
foo(Rational!BigInt.one);
foo(Meters.zero);
?
This would also work for strong typedefs and units of
measurement, where simply assigning 0 to a variable might not
work (because it lacks the correct unit).
It's a very simple change, both in the compiler and Phobos, and
I could have a pull request ready tomorrow.
--
Simen
This can be implemented via a library without requiring any
changes to the language. It would look like Zero!T or One!T
instead. You create a value template for it.
When you are writing less generic code, you can commit to zero or
one in certain types via the prefixes. 1L, 1, 1.0, 1.0f. There's
nothing for short or byte, but you can do short(1), byte(1). You
can also write T(1) or T(0) to get a numeric type T with the
value 0 or 1. That might be better than a template, I haven't
tried it.