data input/output for programs using GSL

[Tribulations Parallèles <paratribulations at free dot fr>]

Hi everybody,

I just discover GSL. I am not an expert in numerical analysis.
I just wonder how reading and writing data (vectors and matrices) in files.
In the chapter 9 of the documentation, some functions are presented, but we 
learn here that "it may not be portable between different architectures". 
Moreover, for example we must know the size of the vector, before using 
gsl_vector_fread. To my mind, these two aspects are rather annoying.
Why not having functions that read in files and place the scalars, vectors, 
and matrices in a convenient structure named entities:

struct gsl_vector_with_name
{
    char * name;
    gsl_vector foo; // Or a pointer, as you want
};

struct gsl_matrix_with_name
{
    char * name;
    gsl_matrix foo;
};

struct entities
{
    gsl_vector_with_name * vectors;
    gsl_matrix_with_name * matrices;
};

So after file reading, vectors pointer could be empty (so still equal to NULL 
if initialized to NULL) if there were only matrices in the file. I have only 
treated the case of matrices and vectors here.
The fied name would be the name of the vector or matrix in the file, as saved 
by the user. The same name that we encounter in Matlab/Octave/Scilab after we 
type 'load data.mat' (given by command 'who' in the workspace). There would 
be two types of storage:
- binary: a format as Matlab .mat one (understood by Octave, and perhaps 
Scilab, I don't remember).
- ascii: an explicit storage, readable by the user. Rather Matlab oriented:

a=[5 4 7]; % Line-vector
b=[3
4
7]; % Column-vector
c=[5 6 7
3 4 5]; % Matrix

So it would be possible to use the same data files in Matlab/Octave/Scilab and 
GSL, it would be rather convenient.

What do you think of that? It is just an idea, perhaps I'm wrong. How do you 
manage I/O in your programs?

Julien