Re: Copying rows in a two dimensional array.

[Adam Beneschan <adam@irvine.com>]

On Feb 2, 2:23 pm, Jerry <lancebo...@qwest.net> wrote:
> On Feb 2, 1:52 am, Jean-Pierre Rosen <ro...@adalog.fr> wrote:
>
>
>
>
>
> > Jerry a écrit :> I've never understood why Ada does not allow slicing in
> > > multidimensional arrays. What are the safety issues involved? And how
> > > is it safe to force the programmer into ad hoc methods?
>
> > One-dimensional slices are simple and efficient. Multidimensional slices
> > are a can of worms.

> Well, yes, I was thinking of rectangular slices. No doubt the (cost of
> implementation) / usefulness is high (and usage difficult) for non-
> rectangular parts, but that is far less common than rectangular parts.
> Python, Matlab/Octave, Igor Pro... all pull it off without too much
> hassle (although Python asks you to imagine addressing the array by
> the "cracks" between elements, as I recall--probably a disease of C-
> style counting).

I don't know anything about any of those languages.  What I know about
Ada is that when multi-dimensional arrays are passed as parameters to
procedures, they're normally passed by reference (if they're large
enough).  Some arrays are required to be passed by reference; in other
cases, it's unspecified, but I'd expect any array to be passed by
reference unless it's a packed array of 32 Booleans or something.  So
say you have a procedure:

   type Matrix is array (Natural range <>, Natural range <>) of
Float;
   procedure Operation_On_Matrix (M : in out Matrix) is ...

You later declare a matrix:

   X : Matrix (1..10, 1..10);

and want to call the operation on a rectangular slice.  If the "slice"
consists of an entire row, or one or more consecutive rows:

   Operation_On_Matrix (X (3..4, 1..10));

this could be done as efficiently as if all of X were passed, since
the procedure would see it as an array of 20 consecutive floats (in a
typical implementation).  However, a slice consisting of one or more
columns:

   Operation_On_Matrix (X (1..10, 5..6));

or some smaller rectangle:

   Operation_On_Matrix (X (4..6, 4..6));

would be tricky, since the procedure now has to be told that there are
gaps between the "rows" of the array that it's seeing as a parameter.
This means making Operation_On_Matrix less efficient, since it has to
be given more information about each Matrix that comes in; and the
inefficiency has to be put there even if no caller ever used a 2-D
rectangular slice, ever.  Maybe the efficiency hit isn't all that
large---I don't know.

It wouldn't really work to fix the language to say that rectangular
slices are allowed only if they include the entire index range in all
dimensions but the first, since that would fail for array types
declared with the Fortran convention.

I realize that the original question was about using assignment to
copy slices.  I suppose that, theoretically, the language could be
changed to allow multi-dimensional slices in assignments but not as
subprogram parameters.  Yuck.  I wouldn't want to add that kind of
inconsistency to the language.

Anyway, I don't know how serious these issues are, but this seems to
me to be a possible reason why adding this feature isn't as simple as
it sounds.

                                  -- Adam