Page:AIM-453.djvu/46



Of course, then everyone who uses PRINT-NUMBER must supply the radix. This is mildly annoying, because most of the time one wants decimal printing, and one tires of writing "10." all the time. One might write another program for most people to use:

This example is simple, but a real PRINT procedure in a real LISP system may be controlled by dozens of parameters like RADIX: format parameters for printing floating-point numbers, which file to print to, file-dependent format parameters such as line width and page length, file-dependent processing routines (e.g. scrolling for display terminals), abbreviation format parameters for S-expressions, etc. All these extra parameters to PRINT are really determined by the larger context in which PRINT is used, but this context is usually not determined by the immediate caller of PRINT. A program which generates and prints successive prime numbers should not have to deal with the complexities of output files; in particular, one does not want to have to rewrite the program just to direct the output to a line printer instead of a disk file. Context decisions are usually made at a much higher level (perhaps interactively by the user). Therefore the solution of using procedures like PRINT-10 is not acceptable; such procedures only serve as abbreviations, binding the many parameters to constants at too low a decision level.

Another idea is to pass the extra parameters for print control through the intermediate levels of the program. But this violates the modularity of the intermediate modules, which generally have no interest in PRINT'S screwy parameters. On the other hand, an occasional intermediate module will be interested in dealing with a few of the parameters (but probably not all of them!). We would like a mechanism for dealing with only the parameters of interest, without having to deal with all of them all of the time.

Side effects can do the job. We can make all the parameters globally available variables (in the top-level environment), initialized to reasonable default values, and invite all interested parties to perform SETQ as necessary. This technique has disadvantages. If every program just changes the parameters at will, then each program must re-set all the parameters (even the ones not of interest) for its own uses of PRINT. This is even worse than just passing PRINT all the parameters!

We can require a convention whereby the parameters normally have their initial default values, and any program which modifies a parameter