Partitioning ObjectOriented Code for Inspections

1
Partitioning Object-Oriented Code for Inspections
Neil Walkinshaw, Marc Roper, Murray Wood The
University of Strathclyde, Glasgow neil.walkinsha
w_at_cis.strath.ac.uk
2
Code Inspection

• Conceived by Fagan in 1976
• Involves groups of individuals reading through
  code documents, trying to find faults
• Best approached by following a policy of divide
  and conquer (Lawson and Parnas)
• Object-Oriented systems encourage distribution of
  functionality across multiple classes
• Difficult to inspect a class in isolation
• Small methods, polymorphism and dynamic binding
  make following trails of functionality difficult

3
Motivation

• Hotel booking system used at Strathclyde for
  undergraduate course
• Full System 280 methods spread over 13 classes
• Call graph contains 1406 edges
• What is the best way to divide this system?

4
Approach

• Divide system into use-cases and scenarios
• Having identified a set of methods M from a
  scenario specification (e.g. sequence diagram),
  proceed as follows

• Identify direct paths on call graph between
  methods in M
• Identify other methods that can influence the
  execution of these paths

5
Step 1 Identify direct paths

• Map methods from a sequence diagram to call graph
• Induce hammock graphs on the call graph, using
  these methods as entry/exit points

6
Step 2 Resolve Method Dependencies

• Methods identified in hammock graph(s) have
  dependencies on other methods
• These methods may not belong to the direct paths
  identified in the hammock graphs
• Intra-procedurally slice backwards from call
  sites in hammock graphs
• Any call sites belonging to slices that are not
  already in hammock graph are marked
• Follow all paths originating from newly marked
  call sites

7
Resolve Method Dependencies (1)

• A call that is not on a direct path between two
  methods may still be relevant to their execution
• Here, the call to the Person constructor is not
  part of a direct path.

8
Resolve Method Dependencies (2)

• We take call statements that are the source of
  edges in the hammock graphs
• We create intra-procedural backward slices to
  identify further relevant calls

9
Size of Final Partitions

• Number of methods to be inspected from a scenario
  in Hotel System

10
Future Work

• Investigation of precision-recall of approach
• How do we proceed to read through the source code
  in the reduced call graph?
• How can this be integrated with a formal
  inspection methodology?
• Does it scale to larger systems?
• How do we ensure complete coverage of the system?
• Inter-procedural slicing on reduced call graph