INTRODUCTION
One of the common tasks every developer needs to do is to locate different bugs in specific applications. When program crashes it's not obvious to trace down the error. In CLEO experiment it was requested to write a C++ traceback module which will keep track of loaded modules and report invalid data requests.

CLEO software is a perfect example of data on demand system. It consists by a set of dynamically loaded modules which perform various tasks. Among them:

  • sinks writes data to a persistent store
  • processors requests/analyse/filters data
  • sources reads data from a persistent store
  • producers produces data on demand

REQUIREMENTS
  • monitor loaded modules and objects they produce during execution time
  • should not add additional overhead to the execution process
  • be informative upon invalid data request and allow users understand the underlying problems

IMPLEMENTATION
The code was designed to register every loaded modules in a separate stack which kept updated during execution time. Every module provides a proxy to load and registers a particular object with descriptive key (type,usage tag,production tag). Once exception being thrown, a stack content run out up to a proxy key which initiates the data request. Such rollback mechanism allowed users to understand the data flow and easily identify the source of the problem.
EXAMPLE
In this example we demonstrate a dynamically generated traceback output of invalid data request initiated by Level4Proc object. 
 Tue Mar 30 12:17:25 2004 Run:  202126 Event:     471 Stop: event
%% WARNING-Processor.Level3Proc: No data of type "Constants" "" "" 
in Record beamenergyshift
  A proxy for this data exists, but it's unable to deliver the data for this 
beamenergyshift record.; skipping crystal decision
%% ERROR-JobControl.ProcessingPaths: 
Starting from Level4Proc we called extract for
[1] type "Level4Decision" usage "" production ""
[2] type "FATable" usage "" production ""
[3] type "FATable" usage "" production ""
[4] type "FATable" usage "" production ""
[5] type "DoitTrackFinder" usage "" production ""
[6] type "ADRSenseWireStore" usage "" production ""
[7] type "DGDetectorStore" usage "DR3" production ""
[8] type "Constants" usage "" production "" <== exception occured
caught a DAException:
"No data of type "Constants" "" "" in Record dralignment
  A proxy for this data exists, but it's unable to deliver the data for this 
dralignment record." ; will continue...
As can be seen, the report message dumps which data objects were requested and instructs which data were missing.

RESULTS
Such implementation significantly speed-up debugging process of the client code. In most cases no additional debugger were required. We benchmark our exception library and measured only 5% of additional overhead to the execution time which was acceptable for our legacy application.

REFERENCES
For more information about CLEO software, please visit this link. A specific talks about CLEO software can be found in Computer in High-Energy Physics (CHEP) conference proceedings.