Re: Will woody ever become stable?

[David Schleef <ds@schleef.org>]

On Wed, Jan 16, 2002 at 10:52:38PM -0500, Joey Hess wrote:
> 
> Well it has many example files that lacks permission to sell. See the
> bug report. Worse though, it turned out one of its source files had the
> same problimatic license. Upstream appears to be cooperating (sorta),
> but there are delays getting a release from them and they seem to have a
> weird policy of not licencing beta code freely or something.
> 
> David, is your replacement really a good enough replacement? It's odd
> that Thimo hasn't commented it at all in the bug report.

Many of the math function implementations were replaced with
calls to libc, such as erf(), erfc(), etc.

Two functions (incomplete beta (ibeta) and incomplete gamma
(igamma)) were replaced with calls to libgsl.  It's not the best
solution, since it adds a dependency, however, it removes math
function implementations which are not as extensively tested as
libgsl.  And, as I observed while testing, the gnuplot and libgsl
implementations don't always agree to reasonable precision.  The
gsl parts are #ifdef'ed for convenience.

There were two functions that still need work, the inverse
gaussian function and inverse error function.  Inverse gaussian
can be replaced with some permutation of sqrt(-log(y)), but for
some reason, the existing gnuplot function doesn't act like
this, IIRC.  Inverse error function should be around somewhere,
but I just haven't looked for it.

Uh... I just thought of something with using libgsl -- libgsl is
GPL, so gnuplot might not be able to link with it.  Gnuplot is
BSD-ish with a changes-distributed-as-patches restriction.



dave...