How to use a class as key type in AAs?

[JAnderson]

Bill Baxter wrote:
> JAnderson wrote:
>> Koroskin Denis wrote:
>>> On Sun, 06 Jul 2008 22:26:46 +0400, JAnderson <ask at me.com> wrote:
>>>
>>>> Jarrett Billingsley wrote:
>>>>> "Moritz Warning" <moritzwarning at web.de> wrote in message 
>>>>> news:g4r043$28f$1 at digitalmars.com...
>>>>>> I've read the docs and implemented size_t getHash, int 
>>>>>> opEquals(Object o)
>>>>>> and int opCmp(Object o); but it still doesn't work.
>>>>>>
>>>>>> The following code prints out: 2 2 1
>>>>>>
>>>>>> Btw.: I tested (new Foo(1)) == (new Foo(1)) and it gives true,
>>>>>> (new Foo(1)) == (new Foo(2)) gives false.
>>>>>>
>>>>>>
>>>>>> void main()
>>>>>> {
>>>>>> class Foo
>>>>>> {
>>>>>> uint i;
>>>>>> this(uint i ) { this.i = i; }
>>>>>>
>>>>>> size_t getHash()
>>>>>> {
>>>>>> return this.i;
>>>>>> }
>>>>>>
>>>>>> int opEquals(Object o)
>>>>>> {
>>>>>> auto t = cast(Foo) o;
>>>>>> if(t is null) return false;
>>>>>> return (this.i == t.i);
>>>>>> }
>>>>>>
>>>>>> int opCmp(Object o)
>>>>>> {
>>>>>> return opEquals(o);
>>>>>> }
>>>>>> }
>>>>>  You don't have opCmp implemented correctly.  opEquals returns true 
>>>>> if they are equal, while opCmp returns 0.
>>>>>  It should be
>>>>>  int opCmp(Object o)
>>>>> {
>>>>>     auto t = cast(Foo)o;
>>>>>     if(t is null)
>>>>>         return 1;
>>>>>      return i < t.i ? -1 : i > t.i ? 1 : 0;
>>>>> }
>>>>
>>>> Or just
>>>>
>>>> int opCmp(Object o)
>>>> {
>>>>      auto t = cast(Foo)o;
>>>>      if(t is null)
>>>>          return 1;
>>>>
>>>>      return i - t.i;
>>>> }
>>>>
>>>> -Joel
>>>
>>> Be careful, it is fast but gives wrong result when comparing int.min 
>>> and int.max!
>>> I personally think that opCmp should return -1, 0 or 1 *only*. That's 
>>> both safer and allows more advanced tricks.
>>
>> True, but at least in my line of work using values in the min/max 
>> range is rare and you can always do something different if your using 
>> byte or unsigned.  Having to branch during this sort of operation can 
>> be very expensive since it can live in the innermost of loops.
> 
> Probably in most any line of work if you're getting within a factor of 
> two of the max int then you are asking for trouble.
> 
> But I think the bigger issue is that precisely because you will never 
> hit int.max or int.min, these values are often used as special indicator 
> values.  Like to mean "not found" or "not initialized" or somesuch.

This is true.  Its difficult to imagine many cases with int where people 
want to use the entire range of signed int for a container.  Maybe if 
they are storing a mask or something.  I guess a healthy dose of asserts 
would be handy.

> 
> --bb