Re: [boost] Suggestion to add class nonistantiable to utility.hpp

From: Beman Dawes (beman_at_[hidden])
Date: 2000-05-26 09:07:30

At 08:18 PM 5/26/00 +1200, Mark Rodgers wrote:

>> An argument in favor might be that a programmer instantiating fred
is
>> probably misunderstanding it, so there is a benefit to letting the
>> programmer know the usage is amiss.  It might make the intent of
fred
>> clearer to a human reader, too.
>
>OTOH, it _might_ prevent a legitimate use.  Suppose we have
>
>  namespace library
>  {
>      struct some_thing
>      {
>         static void foo() { /*...*/ }
>      };
>  }
>
>Now, suppose some user wants to write a function that could use
>library::some_thing, or could use an instance of their own class 
>which has state, and therefore can and should be instantiated. 
>
>They might write
>
>  namespace user
>  {
>      struct my_thing
>      {
>          void foo() const { /*...* }
>          //...
>      };
>
>      template <typename T>
>      void bar(const T &t) 
>      {  
>         //...
>         t.foo();  
>         //...
>      }
>  }
>
>  user::bar(library::some_thing());
>  user::bar(user::my_thing());
>
>As it stands, this would be legitimate and might even be useful.
>However, make library::some_thing non-instantiable and we would
>prevent this usage.
>
>Now, I'm not going to say this is important or even likely.  I'm
>just asking whether we can be too clever in trying to second guess
>how users might want to use our code, and so cut off options 
>needlessly.  
>
>In short, if no harm can come from instantiating a class, and some
>benefit just might result, do we need to prevent instantiation?

Understood.  I'm not sure either.

Anyone out there have a strong opinion or a killer argument?

--Beman