Re: FTP functions

[Xavier Delacour]

On Mon, Mar 24, 2008 at 9:52 PM, John W. Eaton <address@hidden> wrote:
> On 24-Mar-2008, Xavier Delacour wrote:
>
>
> | Other than being a lot more work, I don't see how this solution is
>  | really any different from what is in octave-forge. The package as is
>  | provides the full interface, and is given in about 250 lines (most of
>  | which passes through to ftplib.
>
>  The SWIG solution may be OK.  I have not tried using it.  Looking at
>  the sources, I don't don't see any doc strings defined for these
>  functions.  Am I missing something?  Is there a way to insert the
>  Texinfo doc strings for the DEFUN macros in the interface definition
>  file you use for generating the wrappers?

That is missing, yes. I am working on it, both automatic generation
and manual insertion of texinfo docs.

Manual insertion would look something like
%docstring func_name {
usual texinfo docs, as given to DEFUN
}

There is already similar support for various SWIG targets (eg Python).

> How thoroughly are
>  arguments checked, and how are those checks performed?

By using "typemaps", which match a sequence of input arguments by
their type and optionally names, and insert the argument checking and
translation code. A simple example of a custom typemap:

%typemap(in,noblock=1) (const double* A,int m,int n) (Matrix tmp) {
  if (!$input.is_matrix_type()) {
    error("A must be a matrix");
    SWIG_fail;
  }
  tmp=$input.matrix_value();
  $1=tmp.data();
  $2=tmp.rows();
  $3=tmp.columns();
}

double sum_2d(const double* A,int m,int n) {
  double s=0;
  for (int j=0,k=m*n;j<k;++j)
    s+=A[j];
  return s;
}

This would generate a wrapper for sum_2d that looks like this:

static octave_value_list _wrap_sum_2d (const octave_value_list& args,
int nargout) {
  double *arg1 = (double *) 0 ;
  int arg2 ;
  int arg3 ;
  double result;
  Matrix tmp1 ;
  octave_value_list _out;
  octave_value_list *_outp=&_out;
  octave_value _outv;

  if (!SWIG_check_num_args("sum_2d",args.length(),1,1,0)) {
    SWIG_fail;
  }
  if (!args(0).is_matrix_type()) {
    error("A must be a matrix");
    SWIG_fail;
  }
  tmp1=args(0).matrix_value();
  arg1=tmp1.data();
  arg2=tmp1.rows();
  arg3=tmp1.columns();
  result = (double)sum_2d((double const *)arg1,arg2,arg3);
  _outv = SWIG_From_double((double)(result));
  if (_outv.is_defined()) _outp = SWIG_Octave_AppendOutput(_outp, _outv);
fail:
  return _out;
}

SWIG_check_num_args, SWIG_From_double, and SWIG_Octave_AppendOutput
are helpers that do the obvious things (call error if bad num args,
return octave_value from double, call append on _out).

All primitive types and various STL types have typemaps predefined.
Usually there are only a small number of typemap defs required to wrap
an entire library, and it's no problem to have rigorous argument
checking (including for overload dispatching).

There is extensive documentation:
http://www.swig.org/Doc1.3/Contents.html#Contents
http://www.swig.org/Doc1.3/Typemaps.html#Typemaps

>
>
>  | The package
>  | also includes a few test cases.
>
>  How about using a smaller file for your test?  Also, why did you
>  choose the gcc-objc sources?

Arbitrarily. I'll change it to something else, and something that's smaller.

Xavier