Criteria

• no dynamic memory allocation
• must buffer errors in node and send them approx. every 5 secs to the TCC
• must provide means to limit number of a given message type
• code needs to be simple and compact - we can't include huge libraries
• very limited or no use of STL: this may mean no streams or strings since it's not clear that these work in the L2 alpha environment
• need at least some context information without continuous calls error system to set it up (it would be nice to know a bit more than just the node ID but we don't have time to call set methods throughout the code)
• errors must be capable of being sorted after they have left the L2 node. All our messages will be sent to a central workstation which must then forward them to the alarm system or log files. While it's possible to do this by parsing a string it's simpler to just keep the ID and severity in machine-readable format and reduces CPU time on the node.

Possible Design

• use error objects which store severity, ID and text as well as methods to produce a string (char * or class)
• have central dispatcher object which has input objects and output device objects
• dispatcher keeps count of messages by ID and attaches count to message object before sending it on
• output devices attached via a filter object which can reject messages over the limit or of too low a severity
• possibly different input objects for L2,online and offline - L2 can be function call others can use stream interface if wanted

Possible Class Diagram