Not a liveblog about a conference this time, but hey, still the talks were interesting!
From 22 to 24 April I attended FSEN 2015. For me, the main reason to attend the conference was presenting my work on benchmarking parity games. The conference was extremely well-organised at IPM in Tehran, in particular the students part of the local organisation deserve a special word of thanks for the effort they made in making the conference into an unforgettable experience! By the way, did I already say that the people in Iran are very friendly, and that it is a really nice country to visit? Well, now you know, anyway!
The program of the conference was nicely balanced, with topics such as software architecture, runtime verification, static analysis, software testing, and formal methods. There are a few talks that I would like to discuss in a bit more detail (disclaimer: the choice for these particular talks does not say anything about the quality of the talks).
Automated Integration of Service-oriented Software Systems by Paola Inverardi
As the title suggests, in her talk Paola concentrates on the ways we integrate software. Today, when developing software, having to integrate both black box components (those of which we do not have the source code), and white box components (for which do have access to the code) is a fact of life. The goal is to build a software system satisfying our requirements.
Given the nature of today ‘s systems, getting 100% correct systems is impossible (especially since we do not really know what we get when using blackbox components) therefore it makes sense to concentrate on what we can do. The statement Paola makes is the following:
Final system behavior and dependability depends on the integration code, integrating white box and black box components.
In other words, the code that we write to integrate systems is what we can control, and therefore it is the key to building dependable systems from components. However, in order to achieve confidence in the system we need:
- behavioural models,
- the process lifecycle, and
For this, she proposes to view software science as an experimental science. That is, we need to move from a creationist world in which formal models come with the code, to an experimental world in which the knowledge of software is limited to what can be observed.
In her talk Paola discussed two experimental approaches that help to move towards this goal:
- Learning interaction protocols using a WSRL description of an interface (based on the tool Strawberry).
- Synthesising integration code using, e.g., integration patterns.
Important but hard question to ask when using these approaches: when is your model complete enough with respect to a given goal? In partial when using quantitative and/or partial models we need more research!
QuickCheck: Testing the Hard Stuff and Staying Sane by John Hughes
The second keynote of the conference was by John Hughes. This was one of the most entertaining talks I have seen in the past years, while still covering a lot of content.
QuickCheck is a tool for automated test case generation, and when a test fails a minimal counter example is generated. It used Erlang as a declarative language for modeling and specification. Using a single model, many bugs can be found using random generation of test cases. John explained the theory behind the tool using a wonderful interactive example of a circular buffer, implemented in C, which of course contained all sorts of subtle bugs that were found using QuickCheck.
Now of course, you are wondering, does it work when applied to real world software. John showed it does by providing some industrial examples that he worked on with his company QuviQ.
When you put the car together, you can ‘t even boot it, let alone drive it.
- Testing cars agains the AUTOSAR standard, which is a standard for software components in cars. This project is all about the integration of software from different vendors in a single car. They discovered around 200 problems using QuickCheck, half of which turned out the problems in the standard. The severity of the integration problems is nicely illustrated by the quote “When you put the car together, you can ‘t even boot it, let alone drive it.”
- Analysing software problems that were hard to reproduce using traditional testing approaches at Klarna, race conditions in Erlang ‘s Dets database we discovered. For this case study it was necessary to extend test case generation to generate parallel unit tests.
The take home message of this talk is: Don’t write tests; Generate them!
Formal Methods Friday
This post is already getting quite lengthy, but I would still like to say something about the Friday of the conference. As the title of this section suggests, the Friday was mainly on formal methods. The level of talks during this day was particularly high (although I admit, that might have to do with my preference). Two extremely interesting talks were the following (I do recommend checking out the papers!):
- A behavioural theory for a pi-calculus with preorders by Jean-Marie Madiot. An excellent paper introducing an axiomatisation for a pi-calculus in the context of preorders. He discusses interesting problems with respect to congruence that pop up in the process.
- A theory of integrating tamper evidence with stabilization by Ali Ebnenasir. In this talk Ali takes a look at system correctness, but not in a traditional sense. Instead, he wonders what happens in unexpected circumstances; for instance, what happens if an adversary tampers with the system? He goes into the notions of stabilisation (a legitimate state is eventually reached by system actions only) and introduces tamper-evident stabilisation, which is weaker than stabilisation, and allows for some tampering by the adversary.