A Framework for Understanding The Portability of C Types

The C standard does not specify the memory layouts of types, opening the door for portability problems that arise from discrepancies in how various platforms choose to represent types in memory. I report on the design and implementation of a framework that provides automated help for finding and understanding these discrepancies. A dynamic analysis gathers memory layout constraints on types. An offline analysis checks the truth of these constraints against two abstract platform descriptions. The host description models the platform on which the program has been developed, tested, and assumed to work as intended. The target description models the platform on which the program is to be (perhaps hypothetically) ported. A warning is issued when a type may be used in a way that violates its associated layout constraints on the target platform but not on the host. I envision the framework being used in two key ways: to find type-portability bugs in C programs, by checking the program’s constraints against platform descriptions that model concrete target platforms of interest, and to understand the type-portability of programs, by checking constraints against many possibly unrealistic descriptions and observing the platform features that cause the constraints to become false. This process helps programmers discover and understand portability boundaries: what problematic assumptions the code makes, the locations where these assumptions arise, and consequently the kinds of platforms on which the program behaves as intended.