Most software and hardware for switching systems is proprietary. This includes central processing units, compilers and operating systems that have been developed in-house. For instance, Lucent Technologies, uses UNIX and C targeted towards their own proprietary processor. They have to maintain their own UNIX and their own C compiler. The same phenomenon occurs at Ericsson: they have developed their own central processor and their own operating systems, languages and compilers. The difference between the two situations is that Ericsson uses tools that are not widely used for other processors. As a result, software renovation tools are not available in large amounts for their software. The Ericsson case is therefore an ideal test case for our approach.
As described in the previous section, the first step in generating a renovation factory is the development of a relevant project grammar. In this case, it concerns the proprietary language SSL (Switching System Language). A program in SSL is in fact a collection of programs in 20 different languages that are combined into a single file.
The only complete and reliable source for the SSL grammar is the source code of the SSL compiler. Several steps are needed to extract this grammar. First, the compiler source code is reverse engineered so that the grammar part is identified. Then the essential grammar information is extracted from this part of the compiler. This resulted in the extraction of more than 3000 lexical and context-free production rules in Backus-Naur Form (BNF). Unfortunately, this grammar is not yet usable for reengineering since it is too heavily oriented towards compilation. For instance, the compilation-oriented grammar removes comments, whereas during renovation, the comments should be seen as part of the source since they have to be retained in the final result of the renovation.
As a preparatory step for the reengineering of this intermediate SSL grammar, we have generated a renovation factory for the language BNF itself and have added a number of components that are useful for the reengineering of grammars, such as a transformation of BNF into SDF (the syntax definition language of the ASF+SDF Meta-Environment). This BNF-factory has been used to retarget the extracted SSL grammars.
Using the BNF-factory, we could transform the grammar of one of the sublanguages of SSL in 7 minutes processing time (on a standard workstation) from its initial BNF version into an SDF version that is usable for renovation purposes. Next, we were able to parse about 1 MLOC in this sublanguage in about 70 minutes processing time. Finally, a complete renovation factory for this sublanguage could be generated. When completed with the desired renovation components, it can be used for the factory-like renovation of programs in this particular SSL subset.
Currently, the complete SSL grammar is being assembled by combining the 20 embedded subgrammars. As a next step, we will generate a complete SSL-factory. When that SSL-factory is ready, we have in fact generated from the compiler source code a production line for the rapid development of component-based tools that facilitate the automated solution of software reengineering problems.
