Boost logo

Boost :

From: John Maddock (john_at_[hidden])
Date: 2007-02-09 05:18:47


Matthias Schabel wrote:
>> A review does indeed need input from various people with different
>> areas of expertise, but I don't realy see how to avoid this.
>> Hopefully they will just limit their comments to what they know
>> about ;-)
>
> My concern is that we're effectively building a Boost version of the
> GSL (GNU Scientific Library) from the bottom up rather
> than the top down. If we aren't very careful about certain design
> decisions made at the outset, there may come a point where
> it is discovered that some poor and difficult to fix choices were
> made. So there are two levels that need consideration : 1)
> the low-level implementation (accuracy/performance) of the various
> functions and 2) the overall architecture including error
> handling and higher-level implementation issues.
>
>> There are also some serious practical problems about a
>> retrospective division - not least the documentation, where we have
>> worked
>> quite hard to provide cross linking, and now over 250 pages as a
>> pdf! Not to mention the code, cross function use, and common error
>> handling etc.
>
> I understand the difficulties involved and have great respect for the
> effort you and John have put into this work. I certainly
> wouldn't expect that you completely revamp the documentation,
> especially given its extensiveness. In any case, I want to
> make it clear that I will not let my personal opinions on the library
> impact any decisions I make in the Review Manager capacity.
> At the same time, I hope it is acceptable to provide input and
> opinions as a normal Boost member...

Absolutely.

OK, so how best to proceed?

My gut feeling is that the "tools" part is the least well developed, and as
you rightly say there are sections that could be later developed into fuller
libraries in their own right (I've already had feedback from people
interested in doing just this incidentally).

How about if we re-label the tools section as an internal implementation
detail - which is basically what it is - it can then serve as a placeholder
or "seeder-library" for future work. A bit like what fusion was to spirit,
or the few current special functions we have are to quaternions/octonions.

Which leaves the question of how best to review the two main thrusts of the
library: the special functions and distributions. Would it work if we
roughly divided the review time into two halves with each half notionally
devoted to one section or the other? There's bound to be some overlap but
it might help to focus our collective minds better :-)

Of course if one half is accepted but not the other we've in a lot of bother
'cos they're too entwined to separate, but I'm not going to think about that
right now!

How does this sound? If we go this way then how much time should each
section need do you think?

Thanks, John.


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk