A Static Analysis to Detect Re-Entrancy in Object Oriented Programs.

Abstract

We are interested in object-oriented programming methodologies that enable static verification of object-invariants. Reasoning soundly and effectively about the consistency of objects is still one of the main stumbling blocks to pushing object-oriented program verification into the mainstream. More precisely, any sound methodology must be able to guarantee that the invariant of the receiver object holds at all method call sites. Establishing this proof obligation is tedious, and instead programmers would like to reason informally as follows: methods should be able to assume that the object invariant holds on entry, as long as all constructors establish it, and all methods guarantee that the receiver invariant holds on exit. This reasoning is only correct under certain conditions. In this paper we present sufficient conditions that make reasoning as above sound and show how these conditions can be checked separately, allowing us to divide the verification problem into two well-defined parts: 1) reasoning about object consistency of the receiver within a single method, and 2) reasoning about the absence of inconsistent re-entrant calls. In particular, when reasoning about the object consistency of the receiver within a method, our approach does not require proving invariants on other objects whose methods are called. We present a novel whole program analysis to determine the absence of inconsistent re-entrant calls. It warns developers when re-entrant calls are made on objects whose invariants may not hold. The analysis augments a points-to analysis to compute potential call chains in order to detect re-entrant calls.