I chose C as the implementation language for the optimizer. This language is readily available on any platform I have ever heard of, and most of the platforms support the standardized ANSI C.
The C programming language allows for sophisticated data structures like trees etc. that would be either very hard or perhaps infeasible to implement in a language like Fortran. Because of the run-time efficiency needed in the optimizer, only languages with Fortran- or C like performance could be considered. I chose C because, I at the beginning of this project felt it would meet my demands. Had I known at that time, what I know now, I would not for a second have doubted that C++ would be the language of choice. C++ offers C or Fortran run-time efficiency, but especially the specialized memory handling I implemented would have been a lot cleaner done in C++ with eg. templates.
The Fortran code generator was written in Perl, since this
language offers great string-handling capabilities, and completely
frees the programmer from even considering allocation of data.
Performance is of no importance in this tool, so the fact that Perl
is interpreted (rather than compiled) doesn't matter. The Fortran
code should only be generated once for every new optimized system, and
even though I didn't give performance a second of thought implementing
this tool, it easily generates the tens of thousands of lines of
Fortran code needed to solve the 
system in a matter
of seconds on a standard PC.
The regression test tool was written in Fortran. This choice was made mostly because the automatically generated solver is also a Fortran program. It's a very simple (read: Spartan) piece of software, and I saw no need to complicate matters further by implementing this tool in eg. C.