The CPP Manifesto
In this manifesto, we advocate for the creation of a new international conference in the area of formal methods and programming languages, called Certified Programs and Proofs (CPP). Certification here means formal, mechanized verification of some sort, preferably with the production of independently checkable certificates. CPP would target any research promoting formal development of certified software and proofs, that is:
- The development of certified or certifying programs
- The development of certified mathematical theories
- The development of new languages and tools for certified programming
- New program logics, type systems, and semantics for certified code
- New automated or interactive tools and provers for certification
- Results assessed by an original open source formal development
- Original teaching material based on a proof assistant
Software today is still developed without precise specification. A developer often starts the programming task with a rather informal specification. After careful engineering, the developer delivers a program that may not fully satisfy the specification. Extensive testing and debugging may shrink the gap between the two, but there is no assurance that the program accurately follows the specification. Such inaccuracy may not always be significant, but when a developer links a large number of such modules together, these "noises" may multiply, leading to a system that nobody can understand and manage. System software built this way often contains hard-to-find "zero-day vulnerabilities" that become easy targets for Stuxnet-like attacks. CPP aims to promote the development of new languages and tools for building certified programs and for making programming precise.
Certified software consists of an executable program plus a formal proof that the software is free of bugs with respect to a particular dependability claim. With certified software, the dependability of a software system is measured by the actual formal claim that it is able to certify. Because the claim comes with a mechanized proof, the dependability can be checked independently and automatically in an extremely reliable way. The formal dependability claim can range from making almost no guarantee, to simple type safety property, or all the way to deep liveness, security, and correctness properties. It provides a great metric for comparing different techniques and making steady progress in constructing dependable software.
The conventional wisdom is that certified software will never be practical because any real software must also rely on the underlying runtime system which is too low-level and complex to be verifiable. In recent years, however, there have been many advances in the theory and engineering of mechanized proof systems applied to verification of low-level code, including proof-carrying code, certified assembly programming, local reasoning and separation logic, certified linking of heterogeneous components, certified protocols, certified garbage collectors, certified or certifying compilation, and certified OS-kernels. CPP intends to be a driving force that would facilitate the rapid development of this exciting new area, and be a natural international forum for such work.
The recent development in several areas of modern mathematics requires mathematical proofs containing enormous computation that cannot be verified by mathematicians in an entire lifetime. Such development has puzzled the mathematical community and prompted some of our colleagues in mathematics and computer science to start developing a new paradigm, formal mathematics, which requires proofs to be verified by a reliable theorem prover. As particular examples, such an effort has been made for the four-color theorem and has started for the sphere packing problem and the classification of finite groups. We believe that this emerging paradigm is the beginning of a new era. No essential existing theorem in computer science has yet been considered worth a similar effort, but it could well happen in the very near future. For example, existing results in security would often benefit from a formal development allowing us to exhibit the essential hypotheses under which the result really holds. CPP would again be a natural international forum for this kind of work, either in mathematics or in computer science, and would participate strongly in the emergence of this paradigm.
On the other hand, there is a recent trend in computer science to formally prove new results in highly technical subjects such as computational logic, at least in part. In whichever scientific area, formal proofs have three major advantages: no assumption can be missing, as is sometimes the case; the result cannot be disputed by a wrong counterexample, as sometimes happens; and more importantly, a formal development often results in a better understanding of the proof or program, and hence results in easier and better implementation. This new trend is becoming strong in computer science work, but is not recognized yet as it should be by traditional conferences. CPP would be a natural forum promoting this trend.
There are not many proof assistants around. There should be more, because progress benefits from competition. On the other hand, there is much theoretical work that could be implemented in the form of a proof assistant, but this does not really happen. One reason is that it is hard to publish a development work, especially when this requires a long-term effort as is the case for a proof assistant. It is even harder to publish work about libraries which, we all know, are fundamental for the success of a proof assistant. CPP would pay particular attention in publishing, publicizing, and promoting this kind of work.
Finally, CPP also aims to be a publication arena for innovative teaching experiences, in computer science or mathematics, using proof assistants in an essential way. These experiences could be submitted in an innovative format to be defined.