A Research of Reverse Engineering Tool which generates a non-procedural specification from COBOL programs

软件的维护工作逐年加重。开发/维护成本比，例如COBOL程序，据说是1：3。软件维护的主要困难来自这样一个事实，即一个程序员应该理解另一个程序员编写的程序。本文提出了一种COBOL程序的逆向工程技术，并在此基础上开发了一个COBOL逆向工程软件CORE/M-，CORE/M从执行文件处理的COBOL程序中生成抽象规范。CORE/M采用了非过程性的需求规格说明EOS规格说明来表达对COBOL程序理解的结果。EOS规约由一组等式组成，称之为等式关系，EOS规约可以自动地转换为可视化模块规约，表示实体的属性项之间的关系以及实体的输出和更新需求。在每一个相等关系中，每一项都用“”修改。" ...更多信息 CORE/M不使用变量名来理解程序的含义，因为它们的命名取决于程序员的个人原则。CORE/M的分析完全依赖于程序结构和数据结构之间的对应关系。CORE/M首先对COBOL程序进行语法分析，并将其开发成块结构。接下来，CORE/M决定在每个块中在什么条件下处理哪些实体。可以认为，一个结构良好的程序块中的所有语句都是满足相同条件的相同实体的计算。因此，CORE/M基本上用相同的实体标识符和相同的文件修饰符修改这样一条语句中的每一项。当一个特定的处理通常由一个初始化和一个求和体组成时，它们被相关地转换成一个更抽象的函数，如SUM，MAX和1st。CORE/M的上述分析共分11步，CORE/M实际上转换了由234行组成的COBOL示例程序，并生成了它的EOS规范，它与普通EOS设计人员编写的EOS规范具有相同的含义。未来的问题是CORE/M应该扩展到理解更复杂的COBOL程序，并生成更高级别的规范，如业务数据模型。少

The maintenance of software becomes more serious year after year. The development/maintenance cost ratio, for example of COBOL programs, is said to be 1 : 3. Major difficulty of software maintenance arises from the fact that a programr should understand the program written by another programr. I propose a technique of reverse engineering of COBOL programs, and develop a "COBOL Reverse Engineer-CORE/M-" based on this technique.CORE/M generates the abstract specification from COBOL programs performing the file processing. CORE/M adopted the non-procedural requirement specification called EOS specification to express the result of understanding of COBOL programs. EOS specification consists of a set of equations, called equality relation, and can be automatically translateed to visual module specification by EOS.EOS specification shows the relationships among the attribute items of entities and their output and update requirements. In each equality relation, each item is modified with "." … More by the entity having it as an attribute, and suffixed with "-" by the file storing it.CORE/M does not use the name of variables to understand program meaning because of their naming depend upon programr's personal principles. CORE/M's analysis depends purely upon correspondence between the program structure and the data structure. First, CORE/M parses the COBOL program, and develops it into a block structure. Next, CORE/M decides which entities are processed under what conditions in each block. It can be thought that all the statements in a block of a well-structured program are the calculations of the same entity fulfillling the same condition. Thus, CORE/M basically modifies every item in such a statement with the same entity identifier and the same file modifier. When a specific processing typically consisting of an initialization and a summation body are found, they are correlatively translated into a more abstract function such as SUM,MAX,and 1st. Above CORE/M's analysis consists of eleven steps in total.CORE/M actually converted the sample COBOL program consisting of 234 lines, and generates its EOS specification, which bears the same meanig of that written by an ordinary EOS designer. Future problems are that CORE/M should be extended to understand more complex COBOL program and to generate higher level specification like business data model. Less