Actually, sorting on sparse matrices might work, while bug #54143 is covering up this fact:

octave:15> X = sparse ([1 1], [3 4], [4 3], 1, 8)
X =

Compressed Column Sparse (rows = 1, cols = 8, nnz = 2 [25%])

  (1, 3) ->  4
  (1, 4) ->  3

octave:16> full(X)
ans =

   0   0   4   3   0   0   0   0

octave:17> sort(X)
ans =

Compressed Column Sparse (rows = 1, cols = 8, nnz = 2 [25%])

  (1, 7) ->  3
  (1, 8) ->  4

octave:18> full(sort(X))
ans =

   0   0   0   0   0   0   3   4

octave:21> issorted(X)
ans = 0
octave:22> issorted(sort(X))
ans = 1

If that is the case, then the attached changeset is all that is needed.

There is another use of the phrase "support sparse" in this file:

diff --git a/libinterp/corefcn/data.cc b/libinterp/corefcn/data.cc
--- a/libinterp/corefcn/data.cc
+++ b/libinterp/corefcn/data.cc
@@ -6702,8 +6702,6 @@ ordered lists.
 // Sort the rows of the matrix @var{a} according to the order
 // specified by @var{mode}, which can either be 'ascend' or 'descend'
 // and return the index vector corresponding to the sort order.
-//
-// This function does not yet support sparse matrices.

 // FIXME: Is this function used anymore?  12/14/2015
 DEFUN (__sort_rows_idx__, args, ,

I was going to remove that, but in this case that statement still holds.  Furthermore, there is a FIXME question about whether this routine is even used anymore, so maybe investigating and removing that routine is the better route.  It appears that the script file sortrows() uses the above routine, and for the sparse case it implements a row-by-row sort via a script sort_rows_idx_generic().  Apparently, the sorting method of the sort_rows_idx_generic() accepts sparse matrices.  For sparse matrices, the generic script-based sort probably doesn't have the cost that full matrices do.  I wonder if the better solution here is to move what is in sortrows.m into the built-in routine _sort_rows_idx_ and call that routine "sortrows".

(file #44394)