Don wrote:
Currently, template versions of opXXX_r appear to be useless.
Consider:
struct A
{
int opSub(T)(T x) { return 1; }
int opSub_r(T)(T x) { return -1; }
}
A a;
int x = a - a; // is this a.opSub(a) or a.opSub_r(a) ???
This is annoying and rather silly. I can't imagine a case where you
would template opXXX_r, without also templating opXXX. And by
definition, a.opSub(a) is identical to a.opSub_r(a). Templated opSub_r
has a broken design in both D1 and D2.
We could fix it by adding a rule to operator overloading:
[New rule] 1. If a and b have the same type, the expression is written
as a.opfunc(b).
[Existing rules] 2. If any a.opfunc or b.opfunc_r functions exist, then
overloading is applied across all of them and the best match is used.
...
Patching this is pretty easy. For example, in DMD2.031, opover.c, line
263, simply add:
if ((t1 == t2)) {
// x OP x can only be x.opOP(x), not x.opOP_r(x)
s_r = NULL;
}
Good point. This also reminds me that the entire operator overloading
feature must be thrown away and redesigned.
Andrei
