Re: PATH: default fortran naming scheme when using F2C

[David Bateman]

Michael Goffioul wrote:
> Providing a pre-compiled octave is not really a problem. Even in the
> form of
> a self-executable installer (with the help of NSIS). But I guess it
> would make
> more sense to provide such a package for a public release, and not a
> particular
> CVS image. Anyway, this is something I can provide, if needed.

That I definitely agree with. You can't support a non release version of
a package as its impossible to know whats fixed and what isn't..

> 
> There's also the question of what to put into such a package in terms of
> external
> libraries (FFTW, SparseSuite...), M-packages (octave-forge, which does not
> compile well under MSVC) and so on. 

Packages in octave-forge can and should be patched as needed. However
worry about the octave core first and the rest will follow. It might be
interesting to see how the package manager itself performs under an MSVC
compiled version of octave.

One concerned I'd have with this setup, is that how do we make a version
of octave capable of building oct- or mex-files? We can't distribute
MSVC ourselves and so we have to have instructions of how to go about
getting a setup capable of building oct- and mex-files easily accessible.

> For the moment, I'm compiling octave
> against FFTW, GPLK, a hacked version of SparseSuite and readline, because
> it suits my needs. But you might think about other configurations.

The only thing I feel missing from that is ATLAS, which really is
needed. We'd also need to package with a gnuplot binary to allow
plotting to work correctly out of the box, So there might be so path
issues to address. In short the packaging might be a little more than
just taking octave.exe and making a NSIS installer.

Cheers
David