Re: Signature Package With Generic Proceedure

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On Tue, 24 Jul 2012 01:02:55 -0700 (PDT), Keean Schupke wrote:

> On Tuesday, 24 July 2012 03:53:20 UTC+1, Randy Brukardt  wrote:
>> "Keean Schupke" wrote in message 
>> ...
>> > This is datatype generic programming, one of the things I thought Ada is 
>> > good at.
>> > Signatures are the only way in Ada of doing this with no runtime cost, and 
>> > the only way
>> > using generics. Signatures belong to a family of constructs across several 
>> > languages with
>> > similar properties. For example:
>> 
>> There's nothing in Ada that says generics have to be implemented with "no 
>> runtime cost". Many Ada 83 implementations, and a few more recent 
>> implementations, share all generic instantiations. The effect is very 
>> similar to tagged type dispatching, and it's most useful when there are a 
>> lot of instances in a program. Janus/Ada certainly does this (it might be 
>> the last to do "universal generic sharing"). Some Ada implementations do 
>> "partial generic sharing", where the parameters involved determine if 
>> sharing is used.
>> 
>> Normally, at this point, I'd say something about premature optimization, but 
>> I realize from your past messages that you're already beyond that stage, so 
>> I won't say any more.
> 
> Using DG is a choice like using OO.

DG is "dangerous goods" or "differential geometry"?

> Languages seem to be very keen to add features to support OO programming.
> The zero run time cost is 'potential' because all types can be statically
> determined at compile time. GNAT does this (maybe because it shares a
> backend with C++)

GNAT does this because in Ada static dispatch is always zero run-time cost.
This gives full advantages of OO over generics with no performance loss.

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de