Scientific computing with D

[dsimcha]

== Quote from Lars Kyllingstad (public at kyllingen.NOSPAMnet)'s article
> I think D is, or at least could be, the scientific programming language
> of the future. Here's why -- and possibly how:
> A couple of years ago, I took a university class called Numerical
> Physics. After finishing the course, I was left with the impression that
> numerical computing was all about squeezing every last bit of
> performance out of the computer -- that unnecessary operations and
> function calls should be avoided at any cost, even if the resulting code
> is full of nasty hacks and tricks, making it completely illegible and
> utterly unmaintainable. And of course, in many cases this is true.
> Now, however, I have a bit more experience in the field and I know that
> it is not always so. In numerics, as in other areas of programming, it
> is a trade-off between development time and execution time.
> Traditionally, if one has a desperate need for speed (sorry), one uses
> FORTRAN, C or C++. The programs run very fast, but can be hard to
> develop, debug and maintain. For less processor-intensive tasks one uses
> Matlab, Mathematica, etc. which have a lot of built-in functionality and
> make for rapid development, but programs run at a snail's pace.
> With D one has the best of both worlds. I've used both C++ and
> Mathematica for numerics in the past, but now I use D almost
> exclusively. I find it a lot easier (and more fun) to code in than C++,
> and I spend a LOT less time debugging my programs. On the other hand,
> calculations that would take an entire day in Mathematica are finished
> in a matter of minutes using D.
> It's a fact that D programs don't have the performance of C(++) ones,
> but that, I think, is just a matter of time (pun not intended). It's a
> relatively new language, and the compilers are still somewhat immature.
> The one thing I miss the most, however, and which I think is necessary
> for D to "take off" as a scientific language, is a native D scientific
> library.
> Searching dsource, I find that many nice modules and libraries have been
> made already:
>   - MultiArray (Bill, Fawzi)
>   - dstat (dsimcha)
>   - blip (Fawzi)
>   - MathExtra, BLADE (Don)
>   - Scrapple/backmath (BCS)
>   - Scrapple/units (BCS)
>   - bindings to GSL, BLAS, etc.
>   - ...and probably more
> Myself, I've written/ported some routines for numerical differentiation
> and integration, one- and multi-dimensional root-finding and some very
> basic linear algebra, but so far only for personal use. Currently, I'm
> thinking of porting QUADPACK to D.
> I think it would be really nice if many or all of the above mentioned
> things could be collected in a single library, together with all kinds
> of other stuff. Something like the GSL, only written in D. (In my head
> it's called SciD. At first I thought of DSL - D Scientific Library - but
> that acronym is used all over the place.) I haven't the time, nor the
> skills, to write an entire such library myself, but I'd be happy to
> contribute where I can.
> Here are some design goals I find important:
>   - sensible, logical and tidy package hierarchy
>   - access to high-level functionality for rapid development
>   - access to low-level functionality for performance
>   - make use of D's awesome compile-time functionality
>   - reusability, pluggability and extensibility
> (By the last point I mean that if one method doesn't work it should be
> quickly and easily replaceable in code with something else. This is
> achievable through the use of templates and interfaces, and ties in with
> the second point.)
> So, what do you think? Am I making any sense? Am I the only one
> interested in these things?
> All of the above are, of course, my personal opinions. What are yours?
> -Lars

I think you're definitely onto something.  My other problem with Matlab, R, etc.
besides that they're slow is that they're _too_ domain specific.  They're very
good at what they're good at, but the minute you try to do any more general
purpose programming with them the deficiencies become very obvious.  A lot of
engineers I know try to use Matlab as a general purpose language b/c they don't
want to learn anything else.  I think that, in addition to speed, D is a good
language for this kind of stuff because it's general purpose, but has enough
features (operator overloading, templates, garbage collection, etc.) to
reimplement a lot of Matlab, etc. as a plain old library with decent syntax and
ease of use.  This way, when your domain specific language isn't enough for some
subproblem, you have a _real, full-fledged_ general purpose language standing
behind it.