Sparse matrices and real() function

[Rik]

10/18/13

All,

I just tracked down a segfault in Octave to incorrect behavior of real()
for sparse matrices.

Check out the following:

debug> x = sparse (2i)
x = Compressed Column Sparse (rows = 1, cols = 1, nnz = 1 [100%])

  (1, 1) ->  0 + 2i
debug> typeinfo (x)
ans = sparse complex matrix
debug> xr = real (x)
xr = Compressed Column Sparse (rows = 1, cols = 1, nnz = 1 [100%])

  (1, 1) -> 0
debug> typeinfo (xr)
ans = sparse matrix
debug> nnz (xr)
ans =  1

The problem is that the internal format seems to have become corrupted
after the call to real which changed a non-zero entry to a zero entry. 
nnz() should return 0 after the call to real().  I get the right result if
I force reconstruction of the sparse matrix with

xr2 = sparse (full (xr))

For the time being, I put in just that sort of change in assert.m to stop
the segfault and allow the test suite to run.  But someone with more Sparse
matrix experience should address this.

--Rik