Re: OON: Re: oon-digest V1 #67

From: Michael Soliman (michel@alannah.dki.tu-darmstadt.de)
Date: Fri Apr 14 2000 - 14:02:01 EST


Moin Moin.

"Mukund S. Kalisi" wrote:

> Hi all,
> It seems to me that we need to push for a compiler extension mechanism for
> C++. Consider what we'd like to have.
>
> 1. Compiler is given only the information that is absolutely necessary, with no
> duplicate requirement.
> 2. Input in natural language. If * operator can be overloaded, no reason why
> we can't define other symbols which will affect only compilation and not
> performance. Notations that each library uses. This should be really
> straight forward.
> 3. Optimization information. Expression implementation. We'll identify areas
> where major inefficiencies arise. In Matrix stuff seems like expression
> implementation by expanding each of the operator etc. is extremely inefficient.
> We'll give the compiler this info.

Cool. This could mean to defined functional-relations between functions or classes
rather than giving only the keywords const, volatile and so on. If the compiler
knew strict relations between functions ( , e.g., a is the inverse of b. ) it
could do much more.

>
> 4. All info pertaining to one object should be at one place. Including the
> information that the compiler needs to have.

Schnuerr.

>
> 5. No reason to stick to ascii based source code still. Its too clumsy.

Just like in Mathematica ( or Maple, or whatever ). I for myself share the
opinion, that the pure asc-times should be over (, they can be embedded in a
postscript or tech-like style. ).

        Tschuess,
            Michael.

--
_____________________________________________________________________________
Michael Soliman.
Email: michel@soliman.de
       michel@alannah.dki.tu-darmstadt.de
       michel@sunny.dki.tu-darmstadt.de
Adresse: Hauptstrasse 11a
         64401 Rodau
         GERMANY
Telephon: +(49)616660178
Fax:      +(49)6166920683
 ------------------------------------------------------------------------

--------------------- Object Oriented Numerics List -------------------------- * To subscribe/unsubscribe: use the handy web form at http://oonumerics.org/oon/ * If this doesn't work, please send a note to owner-oon-list@oonumerics.org



This archive was generated by hypermail 2b29 : Wed Feb 20 2002 - 03:20:12 EST