Re: [OctDev] complex error function

[Steven G. Johnson]

Daniel J Sebald wrote:
> Eh, I'm not sure about that.  When it comes to scientific programming, 
> software developers go to sources they can trust. 

This is orthogonal to whether it is packaged as a shared library or as 
individual source code files...

And most special functions are distributed as source code.  (Even 
"libraries" like SLATEC and CEPHES are really just collections of source 
files; they don't even come with Makefiles.)  There are exceptions, of 
course, like GSL, but they are the exception and not the rule.

I've distributed scientific libraries for many years now, and have used 
autoconf-ed packages extensively (and have even contributed fairly major 
patches to autoconf).  This is a necessity for large, complex packages, 
but the vast majority of individual users I've dealt with seem daunted 
by this, and would prefer just a file they can grab (which is only 
reasonable for small self-contained subroutines like the one here).

(This is especially true for Windows users!)

I've also distributed several large scientific simulation packages with 
many dependencies.  By and large, users get scared by these 
dependencies; they don't know how to build libraries, they don't know 
how to tell the compiler/configure/Makefile where to find them, and they 
don't know how to tell the runtime linker where to find them.  Extensive 
documentation helps, but long installation instructions scare people off 
too.

The only saving grace is that, once your package becomes popular enough 
to be packaged by a distro, you can just tell people to "apt-get" (or 
whatever) it and all its dependencies.  But this takes time.