of parameters, to be selected in a non-mutually-exclusive manner. The selection of the parameters is so complicated that it is appropriate to put a compiler on the front end of the software distribution mechanism. Perhaps we are talking more about compilers than we realize. Concerning catalogues: in England a catalogue of building materials is a very ad hoc catalogue, you have left hand flanges to go with left hand gates, etc. I think the catalogue is likely to be ad hoc in that nature, rather than like an electronics catalogue where the components are more interchangeable.
The second issue is the question of writing this compiler. Our file management generator effectively would generate a large number of different file management systems, very considerably in excess of the 300 that McIlroy mentioned. There was no question of testing all of these. We produced an ad hoc solution to this problem, but until more research is done on this problem I don’t think McIlroy’s suggestion is realistic.
Graham: I will speak of an adjunct to this idea. In Multics we used a subset of PL/1, although PL/1 is quite inadequate, in that the primitive operations of the language are not really suited for system design. In Multics you do a lot of directory management, simple operations like adding and deleting entries, but in a complicated directory. With a higher level language with these operations as primitives one could easily write a new system. By simulating the primitives one could test the performance of the system before actually building it. If one had McIlroy’s catalogue stored in the system, with the timings of a lot of routines, then the simulation backing up this higher-level language could in fact refer to the catalogue and use the actual timings for a new machine that this company offered and get realistic timings. Another point, I wish to rebut is McIlroy’s suggestion that this is not for universities; I think it is, There are very difficult problems in this area, such as parametrizing more sophisticated routines, in particular those in the compiler area. These are fit for universities.
Bemer: I agree that the catalogue method is not a suitable one. We don’t have the descriptors to go searching. There is nothing so poorly described as data formats, there are no standards, and no sign that they are being developed. Before we have these we won’t have the components.
McIlroy: It is for that reason that I suggest the Sears-Roebuck type now. On-line searching may not be the right answer yet.