Favor readers over writers. —
Yaron Minsky, JaneStreet, 2010 at NEU/CCS
This ordering is occasionally wrong. For example, we could avoid IEEE floating point numbers nearly all of the time. To make this precise, the Racket sqrt function could return a rational number close to the IEEE float result. We don’t do such silly things, however, because we have decided to value speed over precision in this context. Strive to write code that is correct; maintainable; and fast. The ordering of these adjectives is critical: correct is more important than maintainable; maintainable is more important than fast; and fast is important to include, because nobody wants to live with slow programs.
This section explains these three points as far as the Racket code base is concerned. The rest of this guide is to spell out suggestions that should help you make correct, maintainable, and fast contributions to the Racket code base.
I have bug reports, therefore I exist. – Matthias, watching Matthew, Robby, Shriram and others create the original code base
It is the way we choose to fight our bugs that determines our character, not their presence or absence. – Robby, in response
PLT aims to release good code and to eliminate mistakes as quickly as possible. All software has mistakes; complete correctness is a perfectionist goal. If mistakes are unknown, the software isn’t being used. The goal is, however, to ensure some basic level of correctness before a feature is released and to ensure that the same mistake isn’t introduced again.
We ensure this basic level of correctness with large test suites. Our test suites contain tests at all levels. In addition to unit tests, you will find test suites that use a “random testing” strategy and tools, others use fuzz testing, yet others are end-to-end “systems level” tests, and DrRacket comes with an automatic GUI player that explores its functionality.
For details on testing in the context of the Racket code base, see Testing.
If we wish to create maintainable code, we must ensure that our code is comprehensible. Code is comprehensible when you can understand its external purpose; when you can guess from its external purpose at its organization; when the organization and the code live up to consistent criteria of style; and when the occasional complex part comes with internal documentation.
Released code must have documentation. Conversely a change to the external behavior of code must induce a simultaneous change to its documentation. Here “simultaneous” means that the two changes are in the same ’push’ to the code base, not necessarily in the same ’commit’. Also see Branch and Commit for more on Git actions.
For style rules on documenting code, refer to the style guide in the Scribble manual. Ideally documentation comes in two parts, possibly located in the same document: a “Guide” section, which explains the purpose and suggests use cases, and a traditional “Reference” section, which presents the minutiae. The documentation for HtDP/2e teachpacks is an example where the two parts are collocated. Also consider adding examples for each function and construct in your “Reference” section. Finally, ensure you have all the correct for-label requires and make use of other useful cross-references.
Having said that, the production of a system like Racket occasionally requires experimentation. Once we understand these new pieces of functionality, though, it is imperative to discard the “failure branches” of an experiment and to turn the successful part into a maintainable package. You may even consider converting your code to Typed Racket eventually.
Without adherence to basic elements of style, code comprehension becomes impossible. The rest of this document is mostly about these elements of style, including some suggestions on minimal internal documentation.
Making code fast is an endless task. Making code reasonably fast is the goal.
As with correctness, performance demands some “testing.” At a minimum, exercise your code on some reasonably realistic inputs and some larger ones. Add a file to the test suite that runs large inputs regularly. For example, a regular test suite for a Universe display deals with a 50 x 50 display window; one of its stress tests checks whether Universe event handlers and drawing routines can cope with laptop size displays or even a 30in display. Or, if you were to write a library for a queue data structure, a regular test suite ensures that it deals correctly with enqueue and dequeue for small queues, including empty ones; a stress test suite for the same library would run the queue operations on a variety of queue sizes, including very large queues of say tens of thousands elements.
Stress tests don’t normally have an expected output, so they never pass. The practice of writing stress tests exposes implementation flaws or provides comparative data to be used when choosing between two APIs. Just writing them and keeping them around reminds us that things can go bad and we can detect when performance degrades through some other door. Most importantly, a stress test may reveal that your code isn’t implementing an algorithm with the expected O(.) running time. Finding out that much alone is useful. If you can’t think of an improvement, just document the weakness in the external library and move on.
And as you read on, keep in mind that we are not perfectionists. We produce reasonable software.