From bug-octave-request  Thu Aug  5 13:11:18 1993
Subject: octave-0.74 && suggestions
From: hornik at neuro dot tuwien dot ac dot at (Kurt Hornik)
To: bug-octave
Date: Thu, 5 Aug 93 20:10 MET DST

John,
here are a (minor) bug report and some remarks/suggestions.

On our linux systems (we always compile ourselves with FSQP added; the
versions are octave-0.74, linux-0.99pl9, gcc-2.3.3, libc-4.3.3) we
still have typical readline problems.  Hitting the TAB key does not
always do what it should and sometimes makes further keyboard input
impossible (eg. when in nested input levels).  Typically, regaining
control is possible with C-c, but sometimes this causes Octave to stop
itself after the next input, typically with SIGSEGV.

Anyway, here are some remarks/questions/....

At our site, we have some special functions (complete and incomplete
beta and gamma && derived stuff like erf, bincoeff and some statistical
distributions) implemented in Octave by "porting" them from the C code
in "NUMERICAL RECIPES in C" by Press et al.  The ports are almost
trivial and work really well and could be included in the standard
release, but I am not sure about the legal side.  Do you know anything
about this?

We also have some other simple stuff that might be of use to others.
Have you already thought about starting a contrib subdirectory at your
site?  If so, any rules about how to put code there and compatibility
with MATLAB?  (If this sounds like too much work, we might volunteer
to help ...)

In our applications, we very often come about situations where we
would like to solve an equation of the form "f(x)=y" without having to
explicitly define an auxiliary function g(x) as f(x)-y and then use
fsolve.  Of course, this can be accomplished using global variables,
but perhaps you could add support for this (e.g., if one invokes 
	x = fsolve("f", y, x0),
one gets the solution of f(x)-y=0 starting from x0.)  Or is there a
trivial way of doing this?

Another thing is the following.  Suppose we have functions of the form 
	y = f(x, w)
and think of this as a parametric class of model functions, then
obtain some w0 by fitting the model to data.  This gives us 
	g(x) = f(x, w0)
as the best model function.  Now suppose I want to use this g for
something else, e.g. as the first argument to fsolve.  One possibility
I see is to define g explicitly using some global variables.  Is there
any easier way of doing this?  Clearly, something like
	g = FOO(f, LIST, VALUES)
would be great, where (more generally) LIST contains the arguments to
f which should be considered as fixed for g and VALUES contains their
values.  Do you think something like this might be doable? 

Finally, what do you think about starting an Octave newsgroup?

Best,
-kurt