Martin Pool's blog

Mail me RSS Bazaar distcc Canonical is hiring! flickr

The Geneva Convention On The Treatment of Object Aliasing

The Geneva Convention On The Treatment of Object Aliasing, Hogg, Lea, Wills, deChampeaux and Holt.

Aliasing has been a problem in both formal verification and practical programming for a number of years. To the formalist, it can be annoyingly difficult to prove the simple Hoare formula {x = true} y := false {x = true}. If x and y refer to the same boolean variable, i.e., x and y are aliased, then the formula will not be valid, and proving that aliasing cannot occur is not always straightforward. To the practicing programmer, aliases can result in mysterious bugs as variables change their values seemingly on their own. A classic example is the matrix multiply routine mult(left, right, result) which puts the product of its first two parameters into the third. This works perfectly well until the day some unsuspecting programmer writes the very reasonable statement mult(a, b, a). If the implementor of the routine did not consider the possibility that an argument may be aliased with the result, disaster is inevitable.

Over the years, solutions or workarounds have been found for aliasing problems in traditional languages, and the matter is seemingly under control. Unfortunately, as described below these solutions tend to be too conservative to be useful in object-oriented programs.

The object paradigm has been sold partly on the basis of the strong encapsulation that it provides. This is a misleading claim. A single object may be encapsulated, but single objects are not interesting. An object must be part of a system to be useful, and a system of objects is not necessarily encapsulated.

posted Sat 15 Nov 2003 in /software/theory | link

Copyright (C) 1999-2007 Martin Pool.