The 18th International Conference on Software Engineering (ICSE -18) was held March 27 through March 29, 1996, at the Technical University of Berlin in Berlin, Germany. A number of tutorials, workshops and technical committee meetings also took place in or near Berlin in conjunction with the main conference, thus the week of March 25-31 was dubbed "International Software Engineering Week." ICSE is considered the premier international conference in software engineering. The primary conference sponsors were the IEEE Computer Society's Technical Committee on Software Engineering (TCSE), ACM Special Interest Group Software (Sigsoft), and the German Informatics Society, "Gesellschaft fur Informatik". The selection of Berlin as the conference site was made to encourage attendance from Eastern Europe. The attendee listing from the main conference contains nearly 900 names, more than half from European countries. 350 people attended the fourteen tutorials and a similar number participated in the nine associated workshops.

The conference featured three keynote speakers: Tom DeMarco on "The Role of Software Development Methodologies: Past, Current and Future"; Dr. Anthony Hoare on "The Role of Formal Techniques: Past, Current and Future"; and Dr. Victor Basili on "The Role of Experimentation: Past, Current and Future." (It appears that the keynote titles were assigned, or at least the result of some collaboration.)

The first keynoter on Wednesday morning was Tom DeMarco, of The Atlantic Systems Guild. He described the 1980s as the methodology decade, and proposed that the 90s is the process decade. He identified four paradoxes of software improvement. 1) Every advance made makes the remaining work harder. We've solved the easy problems -- the accidents; what's left are the hard ones -- the essence. 2) Focusing on process increases risk aversion. 3) Reuse is happening, but not where software engineers expected it. The tools and utilities, such as word processors, spreadsheets, and GUI interface builders, that everyone relies on, are what are being reused, and what account for the greatest increase in productivity among software developers. 4) People don't notice or respond to slow changes as well as to rapid ones, therefore software value has incrementally gone down. DeMarco believes the next great discipline for software will be Risk Management, because dealing with the remaining essence involves nontechnical issues -- conflict management and resolution, mediation and negotiations. He identified the two great current debates among software professionals as: Are there any more silver bullets? Is the Capability Maturity Model (CMM) useful?

Thursday's keynote, on the topic of formal techniques, was given by Dr. Anthony Hoare, from the University of Oxford. His alternate title was "How did Software get so Reliable without Proof?" According to Hoare, projects begin to fail in requirements definition. The hard question that does not get asked, or get answered adequately, is how the new system will fit into the existing environment. Efforts at formalizing proofs of correctness, therefore, should be aimed at requirements, because it is easier to find errors in a line of reasoning than in a line of code. As to why software works without formal proof, his answer was that software engineers employ techniques similar to other engineering disciplines, specifically, over-engineering and building in redundancy, which is also known as defensive programming.

Dr. Victor Basili, of the University of Maryland, gave the final keynote on the topic of experimentation. He also made comparisons between software and other scientific disciplines, but noted that software differs because it is about development rather than production. Software does not yet have a good observational knowledge base, yet an understanding of the variables is what will enable us to incorporate risk management into software development efforts. In response to a question about how to encourage Ph.D. students to replicate work, if degrees are awarded on the basis of originality, Basili responded that the act of replication is also a research problem, and we can learn from replication, too, in the form of new insights.

The conference concluded with a final panel session that provided a lively discussion from the audience as well as among the panelists. The provocative title, "If software is so bad, why does it sell so well?" was also restated as "Software Crisis: where is it?" The panelists included Wladyslaw Turski, chair, Manfred Broy, Lee Osterweil, David Parnas, Tony Hoare, and Tom DeMarco. The implication of the topic is that software as an industry is doing well, despite the shortcomings recognized by software engineers (and others). One explanation for the success of software in spite of its quality, from Parnas, is that there is no alternative. Others included: low expectations of software consumers, built-in redundancy, work-arounds, and that we're good at maintenance (bug-fixing). This was echoed by DeMarco, who claimed we've trained our customers to expect a maintenance phase, so that V2.0 and V3.0 of any product are remarkably better than V1. Hoare contended that software sells because the hardware is so good.

The rest of the conference included sixteen paper sessions, containing 52 technical papers in many subdisciplines of software engineering. Sessions on topics related to testing were the most numerous, followed by design issues and methods. Also represented were requirements, measurement, quality, maintenance, and configuration management. Other sessions could be classified as process-related, and several focused on specific techniques. Components showed up in several contexts, from reuse to object-oriented applications. In addition to the paper presentations, there was an industrial experiences track, in which software developers from industry were invited to present examples and case studies of software engineering techniques in practice, and a series of mini-tutorials, which presented particular software technologies, such as methods or tools.

Issues and Trends

Although ICSE is aimed at both researchers and practitioners, participation from academia, in both attendance and papers submitted, far outweighs the industrial representation. This generated much discussion, and is being addressed by the ICSE'97 organizing committee. The differences are partly due to the emphasis that is put on publishing and professional society activities in university settings compared to within organizations that are concerned with product development and profitability.

There was surprisingly little discussion among the participants about the World Wide Web as a phenomenon of interest. It was, however, apparent that this community uses the Web extensively, and is beginning to rely on it as a basic tool. Many included WWW addresses in their presentations, which implies that information about current research activities is being made available on the Web. Access to this information is no longer limited to conference attendees, nor is the level of detail limited to what can be distilled into published papers.

A discussion of the implications of using of the Web as a distribution medium for software came up in the last panel session. For some, this illustrates how the quality of initial releases of new software products is decreasing -- everyone becomes a beta tester, using Version 0.X. For others, this is a sly way to not only get feedback from potential customers, but also get them hooked. Others debated the impact of distributing free software -- how can users complain about the quality of a product, when they haven't paid for it, and how much responsibility can developers be expected to take for supporting a product they're giving away?

Information on the next ICSE, to be held in Boston in May 1997, can be found at http://www.ics.uci.edu/pub/icse97/. Abstracts from the proceedings of ICSE are also on the Web, at www.computer.org/conferen/proceed/icse96/abstract.htm.