Bugs + Fixes

Soon it will be one year since the introduction of Release 7. Since the transition of existing data banks to this new release implied the complete internal restructuring of all intersection data, we were initially somewhat concerned as to what could happen, if --despite of all our efforts and tests-- we missed some serious bug in this automatic transformation. As it turned out, our fears were unnecessary and the transition to this new release went just as smoothly as for the previous releases.

Also in the day-to-day operation, Release 7 has proved to be a solid performer, not causing many problems for our support team. But, of course, this does not mean that there were no bugs at all. Fortunately for us (and you, too!), most of these problems were limited to rather special applications of the newly introduced features. All of the reported bugs were corrected and updates were made available. In November, a partial or complete update was sent out to all users, so that all installations should now be at update level 7.3p/7.4 or higher.

Here is a summary of the reported problems which were corrected in more recent software updates:

• For a data bank not containing node labels, the MORENODE option of macro MODIFYDB did not work properly. The problem can be avoided by adding node labels with option addlabel before increasing the number of nodes with morenode.
• In module 3.23, ``Matrix convolutions'', the use of a full masking matrix with indexing mode ``intermediate zone to destination (kq)'' could lead to invalid results under certain conditions. The problem occurs only for matrices above a certain size (which depends on the licence size of the installation).
• A bug was corrected in module 5.31, which occurred under certain conditions when the additional options transit assignment was used to retain minimum or maximum attribute substrategies. The bug, which resulted in a memory protection fault at the first destination, occurred only for larger networks and also depended on the matrices used. The problem can be circumvented by storing at least one additional options result matrix (additional attribute or active demand) and initializing its values to zero during the assignment preparation.
• A very subtle bug was found and corrected in module 5.11, again related to the additional options transit assignment. Under very particular circumstances, a matrix used during a transit assignment could appear to be read or written in its transposed form, i.e. inversing origins and destinations. As it turned out, this bug only occurred when a new assignment was done on a scenario on which an additional options assignment had previously been carried out and the additional attributes or the active demand were saved in a matrix. If this same matrix was subsequently used in the new assignment as demand or trip component matrix, the problem mentioned above would happen. Note that this particular situation occurs in DEMADJT.MAC -- a macro on the SHARE diskette, which implements the gradient method for adjusting a transit demand matrix using counts. This bug does not cause the macro to crash, but it leads to non-optimal step lengths and results in a deteriorated convergence. The diskette SHARE 1.4 contains a modified version of DEMADJT.MAC which works correctly even with the uncorrected module 5.11.

If any of the above bugs are causing problems in your application, please contact INRO to receive the latest update kit for your type of installation.