The source for this post is online at 2013-11-05-chk.scrbl.
I have been thoroughly brainwashed into believing that writing test for software is extremely important. rackunit is a pretty typical unit-testing framework. For instance, suppose that we have a function that does a basic math calculation:
You can write tests for this by writing things like:
(check-equal? (f 1 1) 2) (check-equal? (f 1 1/10) 11)
I find that the vast majority of the time all my tests are calls to check-equal?. Of course, there are other kinds of tests you can write too. For instance, suppose that you want to test if a boolean expression evaluates to false:
This is slightly more convenient than using check-equal? with #f as one side. An even more convenient thing is when you want to see if a boolean expression isn’t false:
It’s nice to not have to specify what the actual answer is by just using check-not-false.
Another place where it feels verbose is with functions that return multiple values. For instance, quotient/remainder returns two values. If you wanted to test it, here are some different ways:
(let-values ([(x y) (quotient/remainder 10 3)]) (check-equal? x 3) (check-equal? y 1)) (call-with-values (λ () (quotient/remainder 10 3)) (λ vs (check-equal? vs (list 3 1))))
I find both of these to be extremely awkward. Of course, because it’s Racket we can take either approach and make it a macro.
Finally, dealing with errors can be pretty painful in Rackunit. For instance, if you write the test a test where either the actual or expected results throws an error, then the entire test suite dies. You can’t even protect against this by using check-not-equal?. Each one of these expression will crash the whole test suite:
(check-equal? (f 1 0) 11) (check-equal? (f 1 2) (/ 1 0)) (check-not-equal? (f 1 0) 11)
It is very awkward to work around this problem.
Similarly, if you want to make sure there is an error, then it is fairly verbose, because you are required to thunk the expression that errors. You have to write a predicate that determines if the thrown exception is the "right" one, although there is a built-in short-cut for matching the message against a regular expression.
(check-exn exn:fail? (λ () (f 1 0))) (check-exn (λ (x) (regexp-match #rx"di.ision" (exn-message x))) (λ () (f 1 0))) (check-exn #rx"division" (λ () (f 1 0)))
I’ve written rackunit/chk to deal with all these things and simplify writing lots of tests. It is inspired by tests/eli-tester, but it has a different syntax and compiles to Rackunit, rather than its own testing system.
The macro is called chk and supports all the uses case from the prior examples as follows:
(chk (f 1 1) 2 (f 1 1/10) 11 #:f 1 2 #:f #:f 1 1 #:f #:t (and #t (or #f #f) #t) #:t (and 1 (or #f 2)) (quotient/remainder 10 3) (values 3 1) #:f (f 1 0) 11 #:f (f 1 2) (/ 1 0) #:exn (f 1 0) exn:fail? #:exn (f 1 0) #rx"di.ision" #:exn (f 1 0) "division" (f 1 0) (error '/ "division by zero"))
The chk macro accepts any number of tests, without additional parenthesis grouping them. If there are two elements, then it is implicitly a check-equal?-like test, but with protection from errors, support for multiple values, and (as the last example shows) it can ensure that two erroneous expression have the same error. You can negate the meaning of any test with #:f. You can test that a single expression is true-like with #:t. Finally, you can explicitly check for a certain kind of exception with #:exn.
Most of the implementation of this is very simple, but the coolest part is the way I use a define-splicing-syntax-class to specify the syntax of tests. Each different variant of a test provides two attributes: unit and fail-unit. This allows #:f to have a simple implementation that just switches the two of them:
The other cool part of the implementation is the way (chk a) is an abbreviation for (chk #:t a). Normally if you wanted to do that, you would expand the first into the second, but this has to work during the parsing of a syntax class, so you can’t produce some output to re-expand. Instead, I use a with-syntax to immediately parse the expanded version and then grab its attributes:
[pattern (~seq a:expr) #:with (c:test) (syntax/loc #'a (#:t a)) #:attr unit #'c.unit #:attr fail-unit #'c.fail-unit]
This was a fun macro to write and I’ve already found it very useful.
But first let’s remember what we did today!
It’s wonderful to write test cases.
rackunit is great for writing test cases and wherever it is lacking macros can make up the difference.