Question: Issue using matrix sparse storage

Consider the following code, which just generates two "identical" matrices, differing only in their requested storage type, and then does some simple manipulations.

restart;
#
# Define matrix using sparse storage
#
   testM:= Matrix( 40,40,
                           (i,j)->`if`(j>=i,1,0),fill=0,
                           storage=sparse
                        ):
#
# Define identical(?) matrix with rectangular storage
#
   nm:= Matrix( 40,40,
                        testM,
                        storage=rectangular
                     ):
#
# Define procedure to return some matrix properties
#
   matData:= proc( myMat::Matrix)
                            return op(3, myMat)[2], # check storage type
                                      myMat[5, 1..-1], # get 5-th row
                                      add(myMat[5, 1..-1]); # add elements in 5-th row
                    end proc:
#
# Get properies of the two matrices - should be identical
# but check result of adding elements in the 5-th row
#
    matData(testM);
    matData(nm);

The matData procedure ought to produce the same results for the two matrices, with the exception of the storrage type. But the 'add()' command does not. The 'myMat[5, 1..-1]' command produces the same vector, the 5-th row - but stick an add() wrapper around it and all hell breaks loose.

Is this a bug or am I missing something?

Suggestions such as avoiding sparse data storage are not really acceptable: the above is a much simplified version of my original problem where I was using graph theory to play with a "cost function" and (with G a graph) the command,

WeightMatrix(MinimalSpanningTree(G))

returned a sparse-storage matrix - and I didn't notice. There appears to be no option on the WeightMatrix() command to control the storage tyoe of the returned matrix. Result was that all subsequent code based on slicing/dicing/and particularly 'add()ing' sub-blocks of this weight matrix fell apart

Don't get me wrong: I can sort of accept that the weight matrix of minimal spanning tree would (hopefully) be mainly zeros so sparse-storage might be a good default option but I don't see why the results of a command such as

add(myMat[5, 1..-1])

should vary depending on the internal storage used for the matrix, particularly when I have no control over the storage type being adopted