[Kst] How about kmathlab ?

Brisset, Nicolas Nicolas.Brisset at eurocopter.com
Mon Oct 3 18:55:30 CEST 2005

Dear kst users and developers,

There is something I have been pondering over for quite some time, and
considering the vitality of kst I figured I might ask the question here:
how in your opinion could the KDE community develop a viable alternative
to Matlab ?

Mathworks (the editor of Matlab) is the Microsoft of scientific
engineering: same practices, same trap, same outlandish prices, same
relationship with customers (purely commercial). I know a lot of
companies who are very angry at Mathworks, even though they continue to
pay their "tax" in the absence of decent competition. There are a few
packages that try to rival Matlab, but they lack a lot of maturity to
really allow users to change over. 
Scilab is probably the most comprehensive and advanced of those, but
just start it and play with it for a while, you'll understand what I
mean by "not mature". It has some nice functionality, but before people
with real-world needs start considering scilab as a viable alternative,
it needs to add:
- decent looks and a modern GUI
- the possibility to develop user interfaces with more than tcl/tk
- good graphics functionality
- a real block diagram editor
- more (interactive ?) toolboxes
- etc...
I have been following scilab progress for about 5 years now, and I am
really disappointed. They have quite a lot of means, and progress is
really slow. I suspect this is linked with a rather bad infrastructure,
although that has been improving in the last months with the creation of
the Scilab consortium. The code is mostly F77, the interfaces are old
X11, or tcl/tk and recently a bit of gtk (but mostly looking like the
old interface). Well, you get the idea...

So, now I'd like to throw a few ideas into the arena. What would a good
kmathlab package need ? What could serve as a basis to develop it ? Here
is my list, probably too vague and incomplete but at least it's a start:
- a core for numerical computation (think gsl or lapack, but there are
probably   many others) with a command-line interface
- a 100% Matlab-compatible parser (extremely important to keep the cost
of migration low)
- a way to develop nice user interfaces with callback script functions
(think kommander ?)
- good graphics (think kst + 3D extension :-) !)
- a block diagram editor (think kivio) coupled with solvers (I'm sure
that exists, even though I don't know them well) to allow simulation of
dynamic systems
- an interface with a computer algebra system (maxima or yacas for
instance) so that parameters in the equations/state matices/transfer
functions can remain formal and be instantiated only at the very end
(this remark is probably a bit obscure, but I could develop it if there
is interest)
- ...
Of course, reusing some parts of scilab or at least coordinating with
them would also be a good idea.

To sum up, it seems the open source (and specially KDE) world has a lot
of components that could help in building a nice scientific engineering
component, and I am sure that a lot of people would benefit from it and
contribute. Imagine if Matlab users paid 1% of their Matlab licences to
KDE to develop this :-) We just need to find a way to convince them, and
people to do the development !

Sorry for this off-topic post, but I am very interested in this question
and somehow surprised that there does not seem to be much activity to
try and fill the Matlab gap. Fell free to comment/flame/ask
questions/start coding/transfer this message to more appropriate lists
or people :-)


More information about the Kst mailing list