The JOT Blog

My name is Alexandre Bergel and I’m an assistant professor at the University of Chile; for JOT, I’ll be blogging about TOOLS EUROPE 2011. This is the fourth TOOLS EUROPE I’ve attended. This 49th edition took place in the lovely city of Zurich, in the German-speaking part of Switzerland. This blog entry relates the conference as I lived it.

Opening Session

The conference is opened by Judith Bishop, Antonio Vallecillo (the program co-chairs) and Bertrand Meyer. Observation (and evidence!) suggests that this year’s event has more participants than the previous year. The amphitheater is full, and not many attendees had their laptop in their hand, which is a good sign! Judith and Antonio give us some stats, which you can find elsewhere, but what I found noteworthy was that the acceptance rate was 28%; by comparison, ECOOP’11 has an acceptance rate of 28% and OOPSLA’11 36% (!).

This year’s event contains three novelties: Catherine Dubois will give a tutorial about Coq on Wednesday; a best paper prize will be awarded; and a selection of the papers will appear in a special issue of JOT. I like surprises, especially when they are this good. A tutorial by an expert, at no additional cost, is indeed something one should not miss.  The award is also an inexpensive mechanism to advertise the conference: winners are always happy to mention it. The special issue is also a good thing. Unfortunately, JOT is (still) not an ISI journal, which means that is not considered by national science agencies in South America. Oscar, the editor-in-chief, is aware of the importance of being in ISI; I cannot resist to raise this with him each time we meet. Apparently JOT has a nice backlog, meaning that the journal camera ready version of papers is due in December.

Bertrand then tells us something about next year. TOOLS EUROPE 2012 will be the 50th event and will be held in Prague at the end of May. The conference will take place just before ICSE, which will be held in Zurich. This is earlier than usual, but Prague is terrific and I hope the conference as a result attracts lots of great papers.

As much as we like surprises, it’s now time to move on to the even better stuff: the talks!

Keynote – Oscar Nierstrasz

Oscar’s keynote is called “Synchronizing Models and Code”. He considered naming it “Mind the Gap”. However, Oscar is Canadian and lives in Bern, where there is no underground 🙂

Oscar divides his keynote into four parts.

Part 1: “programming is modeling”. This first part is about trying to understand what is modeling in the object-oriented paradigm. Oscar asks what is the “OO Paradigm”? What people traditionally think about OOP is “reuse”, “design”, “encapsulation”, “objects+classes+inheritance”, “programs = objects+messages”, or “everything is an object”. Oscar argues that none of these buzzwords gives an adequate definition, and that the essence of object-orientation lies in the possibility for one to define their own domain specific language. “Design your own paradigm” seems to reflect the truth behind OOP. This reflects a slight change from his previous blog on that topic (Ten Things I Hate About Object-Oriented Programming): Oscar now agrees Object-Orientation is indeed a paradigm. I buy Oscar’s claim that the essence of OOP cannot be easily captured, however I am wondering whether the ability to easily express new paradigms and domain specific abstractions is really what crystalizes OOP. Scheme programmers have a powerful macro mechanism that seems to provide equal capability.

Part 2: The second part of the talk is subtitled “mind the gap”. Here, Oscar is arguing that there is a gap between model and code. This can be easily seen from activity by the reverse engineering community, which spends most of its time making sense of code by providing adequate representations. The modeling community is actively working in the other direction (as can be seen by some of the papers at ICMT’11, which is co-located with TOOLS EUROPE 2011). Oscar presents his argument by clarifying that there are actually many gaps. For example, there is a gap between static and dynamic views: what you see in the source code is classes/inheritance; what you see in the executing software is objects sending messages. I like Oscar’s punch. It’s astonishing to see that code debuggers, code profilers and test coverage tools still base their analysis on classes, methods and stack frames. We profile and debug Java applications the same way as COBOL and C. How can one effectively identify the culprit of a bottleneck or a failure by solely focussing on classes and methods? There is indeed something odd here and Oscar puts his finger on it.

Part 3 of the talk is called “bridging the gaps”. After so many open questions, the natural one to dwell on is: how can be we bridge the gaps between model and code? Oscar’s group has extensive experience in reverse engineering and component-based software engineering. Moose, a platform for software and data analysis (moosetechnology.org) is a solution for bridging these gaps. Oscar spends some time talking about Moose’s features, including an extensible metamodel, metrics and visualisations. Polymetric views are particularly interesting. They’ve had a strong impact on my personal work since most of my research results are visually presented using a polymetric view. To demonstrate some of these ideas, Oscar talks about Mondrian for a while; it’s an agile visualization engine that visualizes any arbitrary data from short scripts written in Smalltalk/Pharo. I am myself the maintainer of Mondrian. I encourage the reader to try it out, people find it mind blowing 🙂

Oscar finishes this part of the talk by summarising a number of language and runtime extensions, including Bifrost (for live feature analysis), Pinocchio (an extensible VM for dynamic languages), Classboxes (which I defined in my PhD), which are the precursor to Changeboxes, which enable forking and merging application versions at runtime.

We now arrive at Part 4: “lessons learned”. What Oscar’s group has learned is that “less is more”, i.e., “keep it simple”. All the experience gained by Oscar’s group is that simple programming languages can often be more successful than complex languages. “Keep it simple” is apparently a favorite lesson learnt of many keynote these days (e.g., Craig Chambers at ECOOP’11, David Ungar at  ECOOP’09).

What else has Oscar’s group learned? Visualizations often use simple metrics defined on simple structures. Oscar also argues that currently there are too many different tools to support software engineering that are not connected. I indeed agree with Oscar. I often found opportunities by connecting different mechanisms and tools (e.g., profiling and visualization).

Oscar’s conclusion is that the things we talk about in software engineering are about managing change. We spend too much time recreating links between artifacts. Instead, we should maintain the links between them. To do this, we should manipulate models, not code. How can I disagree? Oscar’s talk is a success and leads to more than 30 minutes of questions. I had the impression that most of them are about software visualization.

Session 1

After coffee (much needed), we moved to the first technical session, which was actually a shared session with ICMT. As such, the presentations were chosen to be of interest to both TOOLS EUROPE and ICMT attendees. This is always a nice part of TOOLS, the cross-pollination of presentations from different co-located events.

Jochen Kuester presents the first paper on “Test Suite Quality for Model Transformation Chains”, focusing on issues to do with testing model transformation chains (and not just single transformations, which are easier) based on specifying models used as input and output. The authors have lots of experience in dealing with workflow models in particular. What is interesting here is how the authors put an emphasis on the quality of the test suite, including element coverage. This seems to be a good move towards a proper testing discipline for model transformations.

The second talk of the session is given by Marco Trudel, and  is titled “Automated Translation of Java Source Code to Eiffel”. Their work is driven by (i) the need to use Java libraries within Eiffel and (ii) compile Java applications to native code. Translating Java code into Eiffel has a number of challenges: “continue” and “return” Java instruction, exceptions, Java Native Interface (JNI). Only 2.2 overhead performance (based on running Mauve, the java.io unit tests). There are lots of questions after the talk, many of which seem to be abstractions of the question “why translate Java into Eiffel in the first place?”, even though the presenter addressed this at the start.

The third talk is by Markus Lepper, and it is an ICMT paper but from a group outside of the typical ICMT community (more from the compiler construction community). The paper is “Optimizing of Visitor Performance by Reflection-Based Analysis”. This is a bit outside of my personal research interests, but I understood the following: “Everything is a model. Every computation is a model transformation”.

The fourth and final talk of the session is by Fredrik Seehusen entitled “An evaluation of the Graphical Modeling Framework (GMF) based on the Development of the CORAS tool”. GMF is an Eclipse-based framework for development of graphical language editors. The paper evaluates GMF as a result of building a tool called CORAS, and some interesting empirical observations are made, even though the presenter noted that there were potentially simpler approaches to producing the GMF editor (e.g., using a tool like EuGENia) that would lead to different observations. Nevertheless, this is an interesting presentation showing clear benefits, but I admit this conveys an impression of deja-vu for the Smalltalker that I am!

Session 2

The second session of the day focuses purely on TOOLS EUROPE presentations. The first talk is called “KlaperSuite: an Integrated Model-Driven Environment for Non-Functional Requirements Analysis of Component-Based Systems”. The problem on which the authors are focussing on is how to control the cost of enhancing QoS. Their approach is to use an automatic QoS prediction at design-time. It is based on an automatic model transformation of several input models. This is an interesting presentation of a complicated transformation with real applications.

The second talk is titled “Unifying Subjectivity”, by Langone, Ressia and Nierstrasz. “Subjective” means “based on or include by personal feelings”. Subjectivity is related to ContextL, perspective (from Us), roles (from gBeta). The problem addressed by Langone et al. is that all these related work are based on a common point of view. One of their message is that “context is subjective”.

Session 3

The first day ends with three presentations. First up is “Lifted Java: A minimal Calculus for Translation Polymorphism”. Compared to FeatherweightJava, Lifted Java has base classes, role classes, and a playedBy relation. It also has lifting and lowering to both shrink or widen a class during execution. This is an excellent talk on an excellent paper; it turns out that it has been awarded the best paper prize. I observe that today there are a number of papers about context and sujectivity. This is a clear indication that this is a hot topic, and not only for TOOLS!

There are two more talks in the session, “Location Types for Safe Distributed Object-Oriented Programming” and “Static Dominance Inference”, but I am off discussing ideas with some of the previous presenters. Apologies to the last two speakers; there’s so much good research I can’t keep up!