Re: Review: Special functions, Quaternions, Octonions

From: Hubert HOLIN (Hubert.Holin_at_[hidden])
Date: 2001-05-19 17:18:42

Paris (U.E.), le 20/05/2001

--- In boost_at_y..., Gary Powell <Gary.Powell_at_s...> wrote:
> > >   #include <boost/filename>
> > > 
> > > should be
> > >    #include "boost/filename"
> > 
> > 	Well, this is the way it was, but the change to <...> was 
> > requested by Jens Maurer during the early phases...
> > 
> My understanding is that <> searches the complier's library first, and
> "" searches the users. This suggestion was made to me by B.S. so I'm
> following it, even though my compliers seem to find the right include file
> any way I do it.

        This is my understanding as well. I'll abide by the final 
recomandations, whatever they are...

> > >   class octonion:
> > >   default constructors, and constructors taking one argument, with the
> > rest
> > > default, should be declared "explicit"
> > 
> > 	I disagree in the case of the default constructor. Why would you 
> > want them in the other cases you mention?
> > 
>    In general I have found that its better to not have the compiler generate
> implicit conversions, hence the request that any single argument or default
> argument constructor be declared explicit. But I would be glad to be shown
> an example where this would cause pain for the general user.

        In this case I believe the implicit conversions will not cause 
unexpected behaviour: if it occured it is because one was actually 
expecting a quaternion (or octonion, resp.). This is an upwards 
conversion, with no loss of information. But I agree "explicit" is in 
general safer. Again, I'll abide with the final recomandation.

> > 	I'd rather keep my distance with "swap" until we clearly know 
> > where to put it.
>   Fine by me. Since it appears that a default swap would do the correct
> thing anyway.
> 
> > 	Unfortunately no, template specializations may be essentially 
> > unrelated.
>  Dang!
> 
> > >   At some future time, you might consider using expression templates for
> > > those operators for which it makes sense. (I'm thinking addition and
> > > subtraction)
> > 
> > 	Could you please give more details on that?
> > 
>    It will take more than one email, but it essential takes
>    octonion a, b, c, d;
> 
>    a = b + c + d;
> 
>    and generates code which removes the temporaries 
> so that instead of doing
> 
>   t1(b)
>   t1 += c;
>   t2(t1)
>   t2 += d;
>   a = t2;
> 
>   that using an expression template the code would be more like
> 
>   a.a = b.a + c.a + d.a;
>   ....
> 
>   see PETE at http://www.acl.lanl.gov/pete for a general discussion of
> expression templates.

        I'll read the documentation, but I may not be able to use it 
meaningfully before the end of the review period. This does not seem to 
be something minor to do. Perhaps for an ulterior evolution?

> > >   Also you may want to consider adding your own manipulators for setting
> > the
> > > '(', ')' and ',' delimiters for io. There are some good articles over on
> > the
> > > CUJ expert forum site that tell you how to do this.
> > 
> > 	Any URL welcome...
> > 
> Try Matt Austern's article at
> 
>   http://www.cuj.com/experts
> 
> The February 2001 has a discussion on how to do it.

        I'll read it. I would really like to improve the io (especially 
for octonions...). Not sure I can do anything sensible before the end 
of the review period, but I'll try.

> > 	If boost delineates namespaces, I will gladly go along with them.
> Some libraries have them, and IMO, most should.

        I'me fully open to any suggestion.

> > 	Well, once I feel more confortable with swap...
> Probably won't happen. I'm still not comfortable with it.
> 
>   Thanks again.
>   -gary-
> 
> 
> gary.powell_at_s...

                You're welcome!

                        Hubert Holin
                        Hubert.Holin_at_[hidden]