From bug-request at octave dot org Tue Jan 18 14:03:16 2005
Subject: Bug concatenation of NDArray's
From: "John W. Eaton" <jwe at bevo dot che dot wisc dot edu>
To: David dot Bateman at motorola dot com
Cc: bug at octave dot org
Date: Tue, 18 Jan 2005 15:06:20 -0500

On 17-Jan-2005, David Bateman <David dot Bateman at motorola dot com> wrote:

| There is a bug in the concatenation of NDArray's that can easily be seen 
| with the example
| 
| A = cat(3,[1,2;3,4],[1,2;3,4]);
| B = [A,A]
| 
| That will seg-fault. The problem is in the function
| 
| template <class T> Array<T>&
| Array<T>::insert (const Array<T>& a, const Array<int>& ra_idx)
| 
| where the calculation of the number of elements to skip between copied 
| blocks is incorrect. The problem only appears for concatenation of 
| NDArray's where the dimension along which they are concatentated is not 
| the last one.
| 
| Attached there is a fix that should be applied to all branches.

I applied this change.

| However, the fix assumes that
| only one element of ra_idx is non-zero and that only one element (the 
| same as non-zero ra_idx element) of dims() and a.dims() differs. This is 
| in fact the case everywhere that this function is currently. However, it 
| doesn't cover the general case of insertion of one arbitrary matrix into 
| another. Is this a problem?

Yes, it seems that it could be.  Even if this is incorrect usage, I
think it should be detected so we can issue an error message instead
of (potentially) crashing.

jwe



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------