Why do we have transitive const, again?

[Timon Gehr]

On 09/25/2011 01:18 PM, bearophile wrote:
> Walter:
>
>> struct LogicalConst(T)
>> {
>>       @property T v() {
>>          if (!set) {
>>             _v = some_expensive_computation();
>>             set = true;
>>          }
>>          return _v;
>>       }
>>
>>     private:
>>       bool set = false;
>>       T _v;
>> }
>
> If this idiom becomes common (thanks to its presence in D docs, etc) it is possible to add logic to the DMD front-end to recognize it (even when inlining of v() doesn't happen), and use this knowledge to perform some const-like optimizations. Is this possible and useful?
>
> Bye,
> bearophile

I think it should be a language feature, as proposed by Jonathan.