Nauer - 程序设计是一个构造理论的过程

source

 

Case 1 concerns a compiler. It has been developed by a group A for a language L and worked very well on computer X. Now another group B has the task to write a compiler for a language L + M, a modest extension of L, for computer Y. Group B decides that the compiler for L developed by group A will be a good starting point for their design, and get a contract with group A that they will get support in the form of full documentation, including annotated program texts and much additional written design discussion, and also personal advice. The arrangement was effective and group B managed to develop the compiler they wanted. In the present context the significant issue is the importance of the personal advice from group A in the matters that concerned how to implement the extensions M to the language. During the design phase group B made suggestions for the

manner in which the extensions should be accommodated and submitted them to group A for review. In several major cases it turned out that the solutions suggested by group B were found by group A to make no use of the facilities that were not only inherent in the structure of the existing compiler but were discussed at length in its documentation, and to be based instead on additions to that structure in the form of patches that effectively destroyed its power and simplicity. The members of group A were able to spot these cases instantly and could propose simple an effective solutions, framed entirely within the existing structure. This is an example of how the full program text and additional documentation is insufficient in conveying to even the highly motivated group B the deeper insight into the design, that theory which is immediately present to the members of group A.

 

In the years following these events the compiler developed by group B was taken over by other programmers of the same organization, without guidance from group A. Information obtained by a member of group A about the

compiler resulting from the further modification of it after about 10 years made it clear that at that later stage the original powerful structure was still visible, but made entirely ineffective by amorphous additions of many different kinds. Thus, again, the program text and its documentation has proved insufficient as a carrier of some of the most important design ideas.

 

The conclusion seems inescapable that at least with certain kinds of large programs, the continued adaption, modification, and correction of errors in them, is essentially dependent on a certain kind of knowledge possessed by a group of programmers who are closely and continuously connected with them.

 

In terms of Ryle's notion of theory, what has to be built by the programmer is a theory of how certain affairs of the world will be handled by, or supported by, a computer program. On the Theory Building view of programming the theory built by the programmers has primacy over such other products as program texts, user documentation, and additional documentation such as specifications.