x(idx)=y(idx) for empty idx

[John W. Eaton]

On  3-Mar-2004, Paul Kienzle <address@hidden> wrote:

| Consider:
| 
|       y = rand(100);
|       idx = y>1;
|       y(idx) = y(idx)-1;
| 
| With warn_fortran_indexing=1, this prints:
| 
|       warning: A(idx) = []: expecting A to be row or column vector or scalar
| 
| As a result we have a bunch of code like this:
| 
|       idx = condition;
|       if any ( idx (:))
|           A(idx) = f(A(idx));
|       end
| 
| instead of:
| 
|       idx = condition;
|       A(idx) = f(A(idx));

Please try the following patch.  It eliminates the warning you see
above and also fixes a bug (if the warning flag was set, deleting
elements didn't work at all).

You will still see a warning about "single index used for matrix" in
the 2-d case.  For N-d arrays, there is no warning.

Yes, we should be consistent here.  I hope that all the different
indexing and indexed assignment functions will eventually be combined
so we only have one of each.  But first, I think the N-d versions
could use some more work.

Thanks,

jwe


2004-03-05  John W. Eaton  <address@hidden>

        * Array.cc (Array<T>::maybe_delete_elements_2):
        Omit Fortran-indexing warning.


Index: liboctave/Array.cc
===================================================================
RCS file: /usr/local/cvsroot/octave/liboctave/Array.cc,v
retrieving revision 1.108
diff -u -r1.108 Array.cc
--- liboctave/Array.cc  5 Mar 2004 16:19:11 -0000       1.108
+++ liboctave/Array.cc  5 Mar 2004 16:34:03 -0000
@@ -1270,14 +1270,6 @@
       n = nr * nc;
       nr = 1;
       nc = n;
-
-      if (liboctave_wfi_flag)
-       {
-         (*current_liboctave_warning_handler)
-           ("A(idx) = []: expecting A to be row or column vector or scalar");
-
-         return;
-       }
     }
 
   if (idx_arg.is_colon_equiv (n, 1))