Tool of Thought

APL for the Practical Man

On Control Structures

August 1, 2022

The latest episode of Array Cast is on control structures. As usual, it is an excellent episode, covering history and all the options in the various array languages. However, the panelists raised a number of interesting questions which were not fully examined, let alone answered - understandable, given how much material was covered and the time constraints. One question was why use control structures at all?

One of the main arguments made in favor of using control structures was expediency . While this argument was meant to be proscriptive, it is definitely descriptive: it is obvious from inspecting large APL code bases that make heavy use of control structures. But expediency is just another word for a shortcut. We don't have time to do it right, so we take the fast and easy way out. As tempting as this always is, it is almost never the right approach.

Expediency piles on the technical debt. When hitting a bug, instead of refactoring the code, and perhaps extracting a new function or two, we just if-then-else or trap around the problem. Functions lose their focus, they get longer and longer, variables proliferate, control structures get nested, and nested, and nested again.

Note that the issue here is not searching for an array-based solution, which, as noted in the podcast, may be difficult (or even impossible) and may not be worth the time. The issue is general refactoring, important in any language or paradigm. When there is a function that appears to require lots of control structure statements, it is often possible to rearrange things and factor out the control structures. A new function or two may need to be extracted to cleanly eliminate some control structures. This effort, and make no mistake, it is indeed effort, is well worth it.

By eliminating control structures, one gains a deeper understanding of the problem, and the resulting code is shorter, more direct, and easier to debug, maintain and enhance. And by extracting more and more functions a domain-specfic English vocabulary arises and separates itself from the APL primitives. The code can become self-documenting. In the end we may still need some control structures, but they are separated and are above our APL primitive code. We don't have a :If keyword followed by a welter of primitives, a function train, and an embedded assignment. Functions are written at the same level of abstraction. Functions are mostly written in English, or mostly written in APL. As was noted in the podcast, it has been suggested that control structures are sort of outside the language of APL. If it is a good idea that we code at the same level of abstraction, it follows that control structures should not be mixed with the heavy use of primitives.

Most importantly, control structures provide an easy way to avoid naming things. Refactoring code to remove control structures requires naming new functions. Naming things is important. Naming is also hard under the best of circumstances. When looking at a large rambling function is can be impossible to find any section that is nameable - without rewriting everything. When a program is a single large function full of control structures, a fix or enhancement can be made with little thought to where it goes, precisely because there is no particular place for anything. There is only one name, the name of the big function. It's like a large messy room with no closet, chest of drawers, or desk. Anything can just get thrown in the room with no concern for where it is. This is chaos.

In a program with many small well-named functions, the place for a change must be carefully considered. There may be no good place for the change, and a new function may be required. Or the change may go in a particular function, expanding or changing its duties and making its name inappropriate. A change can easily require multiple function name changes and the introduction of new functions. This takes work. This program is like a large neat room with appropriate storage; everything has a place, and everything in its place. The dirty shirt has a place in the hamper, the book has a place on the desk. If the room acquires a lot of books, and the desk becomes unmanageable, a bookshelf is installed. This is order. Bringing order out of chaos requires naming. Keeping things in order requires renaming.

In addition to the question of why use control structures at all, the panelists pondered if it is possible to write large scale applications without them. These two questions are related, and if the answer to the first is primarily "expediency", the answer to the second is definitely "yes".