From: "Paul Ranson" 

FWIW I think an initial underbar on an identifier is reserved for the
implementation, so bad practice for the user to use.

I'm not sure what Geo's requirement was, but if it boils down to a class
with simple methods then it's a reasonable expectation. In this case the
template magic is trivial.

Paul

"Gregg N"  wrote in message
news:419d388a$1{at}w3.nls.net...
> Geo wrote:
>> "Gregg N"  wrote in message
>> news:419c0ffb{at}w3.nls.net...
>>
>>>     Operands(T a, T b) : a_(a), b_(b) { }
>>
>> what's the a_(a) stuff about?
>>
>> Geo.
>
> A common convention is to stick an underscore at the beginning or end of
> the
> name of a class member variable. In this case, a_ and b_ are the names of
> the two variables that will store the operands. Another convention
> commonly
> used Windows-based code is to use an m_ prefix.
>
> The sequence
>
> : a_(a), b_(b)
>
> is an example of a constructor initializer list. It tells the compiler how
> to provide arguments for the construction of the object's own member
> objects. In this case the member objects are just built-in types, so I
> could
> have done this as
>
> Operands(T a, T b)
> {
>   a_ = a;
>   b_ = b;
> }
>
> but that won't work for member objects whose constructors require
> arguments,
> and is conceptually wrong because it theoretically means you are
> constructing a_ and b_ and then assigning to them. For non-trivial objects
> this would be inefficient.
>
> Gregg
>

--- BBBS/NT v4.01 Flag-5