CS 4448 - Spring 1998
Object-Oriented Programming and Design
Talk 10.2.1
by
Phil Hugger

Unreachable Procedures in OOP

Introduction

• Unreachable procedures unnecessarily bloat an executable.
• Requiring more disk space
• Changes the cache/paging behavior - more swaps
• Programmers do not write code they don't intend to use.
• Object-oriented writing may contain unreachable code, since not all behavior patterns might be used.
• Statically linked languages are primarily affected
• All procedures in an object module are linked if any of them are referenced.
• To minimize library sizes, programmers split modules into smaller groups.
• Library splitting is not always possible
• Destroys organizational structure
• Not suitable for OOP
• Dead code elimination
• Useless/unreachable code is removed
• Detection of dead code requires entire program to be present.
• Global procedures have a scope larger than a single file, so the compiler cannot detect usage.

Three main sources of unreachable procedures

• OOP programming style
• Behavior of objects is more important than implementation
• Class design - all possible interfaces and manipulations are defined, but only a few may be used.
• Inheritance - Program might not use inherited methods
• Longer inheritance chains = higher probability of unused code
• Virtual functions - derived classes may redefine methods, the original definitions are inherited but unused.
• Design of libraries
• Many common methods for fundamental data types are defined, but not all of them may be used
• Debugging
• In a released program, debugging routines are never called and are unreachable

Detection of unreachable code

• Build a directed call graph
• Nodes are procedures
• Edges are statically present procedure calls
• A conservative algorithm cannot detect dynamically unreachable code.

Code analysis

• Large programs were selected, since small ones give misleading results
• C/Fortran programs without system libraries have 0-5% unreachable code, C++ OOP programs have up to 26% unreachable code.
• In programs with system libraries, percentages are similar but C++ is consistently higher.
• Unreachable procedure analysis - two groups
• Never called (no predecessors) - 45-82%, average 64%
• Have predecessors, no predecessor is reachable from program.
• Libraries/File splitting
• Splitting library files decreases unused methods
• Unreachable user routines will still include library routines
• Splitting is inconvenient or impossible at times
• Shared variables/procedures (static) between routines mean that they cannot be separated