U++ Core vs D programming language

[Brad Roberts]

Georg Wrede wrote:
> Jason House wrote:
>> Poking around the U++ web site, I spotted this page:
>> http://www.ultimatepp.org/www$uppweb$vsd$en-us.html
>>
>> It says "C++ is still well ahead of D (by 70%) if not being hold back
>> by standard library design and average implementation..."
> 
> http://www.ultimatepp.org/www$uppweb$overview$en-us.html
> says:
> 
>> Everything belongs somewhere
>>
>> In Ultimate++, most objects are bound to some logical scope. As a
>> result, you will not see many new operators in code using Ultimate++
>> and almost no delete operators outside the implementation of
>> containers.
>>
>> That of course does not mean you are not allowed to use pointers, but
>> it is good practice to use pointers just to point to things, never to
>> manage heap resources. This also avoids all confusion regarding
>> ownership of the underlying object, time of its deletion etc. If you
>> need to manage data sets of variable size or polymorphic type, you
>> should prefer using one of Ultimate++ containers.
>>
>> Speaking about it, there are no shared smart pointers (like
>> boost::shared_ptr) in Ultimate++ used to manage heap resources at
>> interface level. They are not needed and considered bad practice.

That's certainly _a_ philosophy.  If it happens to work well enough for
your app, then it's a very valid way of doing things.  But often that's
much more easily said than done.

> And here comes the point:
> 
>> In C++, this approach proves to be equally good or better than
>> garbage collected languages like Java or C#. While those languages
>> are able to provide automatic management of heap resources, U++
>> approach provides very deterministic automatic management of all
>> resources.
> 
> What should I tell people who quote the last paragraph?

I'd challenge the original premise.  The assumption is that every
application can be managed by static allocations on the stack and that
it's sufficiently easy to do so that no other technique need be used.
If that's true, then ok, every other technique is useless and almost
certainly slower.  I have a very hard time believing that.

Next, if you can't use pure scope based allocations, then there are
different tradeoffs between RAII, refcounting, and various garbage
collection schemes.  The exact runtime performances vary depending on
usage and exact implementation.  It's impossible to make absolute X is
better than Y withouut loads of qualifications (that make them
non-absolute :).

Later,
Brad