Conclusions

In this tutorial, we have covered a variety of issues in the area of system renovation. Starting with a discussion on the economic need for maintenance and renovation, we have presented our approaches to problems like object identification, system understanding, grammar reengineering, and the creation of renovation factories.

Several observations can be made about the field of software renovation as a whole. A most challenging aspect of software renovation research is the number of different areas in which proficiency is required. These include:

• Historic programming languages and systems, such as COBOL and the mainframe environment.
• Target new programming language technologies, such as CORBA or Java, and the current and future market standards.
• Migration technology, such as the ASF+SDF Meta-Environment.
• Software engineering theory and commercial practice.
• Knowledge transfer, coping with conservatism (``COBOL is the best''), unrealistic expectations of new technology (``Java will solve everything''), and unfamiliarity with migration technology (``What did you say a grammar was?'').

Clearly, this should also have an effect on the curricula for software engineering.

Concerning the migration technology used, we benefited from the use of the ASF+SDF Meta-Environment. Three of its distinctive technical properties turned out to be of great significance for renovation purposes:

• The techniques used are generic: they do not depend on one particular programming language, such as COBOL or PL/I, but they are parameterized with a language definition. As a result, major parts of our renovation techniques can be directly reused for different languages. To give an example, the Year 2000 problem resides probably in systems written in 500 ``standard'' languages [31] plus another 200 proprietary languages [30].

• The techniques used are formal: the underlying formal notions are many-sorted signatures and positive/negative conditional rewriting. Program conversions are, for instance, expressed as rewrite rules. This formalization increases the quality of analyses and conversions [27].

• Syntactic analysis is based on generalized LR parsing (GLR). This enables the construction of modular grammars and parsers for languages with large grammars and many dialects (like COBOL) [12]. More traditional parsing techniques (e.g., LALR(1) as used in parser generators like Yacc and Bison) lead to increased maintenance problems for large grammars: the time needed to add language constructs needed for new dialects becomes prohibitive.

The driving force behind software renovation is the strong need to maintain and renovate parts of the software volcano. This implies that software renovation research has to be carried out in close cooperation with industrial partners that are in the possession of problems that have to be solved. Fortunately, we can be confident that progress in areas like compiler and programming language technology and formal methods, will continue to help to offer the right tools at the right time. At the same time, the analysis of legacy systems provides the empirical foundation for programming language research. It uncovers the effects, both positive and negative, of years of intensive use of a programming language in the real world.

As we have tried to show here, software renovation research also reveals new challenging problems. Techniques for analysis or code generation that were satisfactory from the perspective of a traditional compiler may no longer be satisfactory from the perspective of interactive program understanding. The gigantic scale of renovation projects presents implementation problems (and opportunities) that may inspire research for many years to come.

Finally, the largest challenge we see is to try to bridge the gap between research aimed at building new software and research aimed at maintaining or renovating old software. We strongly believe that an integrated approach to both is the best way to proceed. This implies introducing maintenance and renovation considerations much earlier in the software construction process than is usual today. This also implies designing new languages and programming environments that are more amenable to maintenance and renovation.