Re: [OctDev] complex error function

[Steven G. Johnson]

Jordi Gutiérrez Hermoso wrote:
> On 21 November 2012 17:47, Steven G. Johnson <address@hidden> wrote:
> > 3) Other than Octave and possibly SciPy, it is not clear who would use it in
> > library form.
>
> Oh, btw, off the top of my head: R, Scilab, Julia, GNU libc, GNU
> Scientific Library, The Madlab (I'm sure you've already contacted
> them?), and perhaps CASes like Sage, Maxima, Axiom, Yacas and if you
> can stomach the attitude of the non-free ones, Mathematica and Maple,
> assuming your algorithm is the best there is.

I'm not saying that no one will want to *use* my code.  I'm suggesting 
that most users will probably prefer to grab their own copies rather 
than introduce a dependency on an external library.

GSL seems to prefer its own implementations of special functions (and 
other algorithms, for that matter); it rarely links external libraries, 
even when the implementations in external libraries are far superior to 
GSL's.

The Julia source, when I checked it a week or two ago, distributes its 
own copies of nearly all external code, presumably so as to limit 
dependencies.

glibc, if they were to accept a cerf contribution, would presumably want 
it as part of glibc (like all of their other special functions) and not 
as an external library.

Non-free software almost always packages their own copies of 
dependencies rather than relying on the user to install dependencies 
separately.

...

Again, it is perfectly easy for me to make an autoconf tarball that 
installs a shared library.  But I am reluctant to do so unless I get 
more indication that users (other than Octave) would want to use it in 
library form as opposed to C&P.

> Do you still want to go the C&P route for Octave?

I still think this is the most reasonable route for users.