By now, hopefully, you've read the Intro to Coalton and written some non-trivial Coalton code. Now the question is, "how do I know it works?"
The fact that your code type-checks and compiles should assure you that it's free from a large class of bugs, but that doesn't necessarily mean that it does what you expect. If you're worried about whether you've got the logic right, it's time to write a test suite!
You can find an example project containing the code from this document in examples/coalton-testing-example-project
For a project foo
, you should already have a system definition file foo.asd
located in
the project's root directory, which looks something like:
(defsystem "foo"
:depends-on ("coalton" "named-readtables" ...other dependencies?)
:components ((:file "main")
...other components?))
To that defsystem
form, add the following line:
:in-order-to ((test-op (test-op "foo/test")))
(Replace foo
with your actual system name, obviously.)
This tells ASDF, "when you want to run the test-op
operation on this system, run
test-op
on this other system "foo/test"
."
That's great, but as yet the "foo/test"
system doesn't exist. Add to the .asd
file a
new defsystem
form like:
(defsystem "foo/test"
:depends-on ("foo" "coalton/testing" "fiasco")
:pathname "test/"
:serial t
:components ((:file "test"))
:perform (test-op (o s)
(symbol-call '#:foo/test '#:run-tests)))
(Again, replace foo
with your actual system name.)
Note that the name of your testing subsystem must start with your full system's name, with
some suffix after a /
. This tells ASDF that it's a subsystem, and so to search for its
definition in foo.asd
, rather than trying to find a file foo/test.asd
.
The :pathname "test/"
says that the components of the subsystem will be in the test/
subdirectory, which we'll create in a moment.
:components ((:file "test"))
says that this subsystem contains a single file,
test.lisp
. We'll create it in a moment.
:perform (test-op (o s) (symbol-call '#:foo/test '#:run-tests))
tells ASDF, "when you
want to run the test-op
operation on this subsystem, look up a symbol named RUN-TESTS
in the package FOO/TEST
, and call that function." This means that we'll need to define a
package named FOO/TEST
, and it will need to export a symbol FOO/TEST:RUN-TESTS
which
names a function that runs our test suite.
The :perform
clause actually defines a function to handle the operation, and o
and s
are arguments bound to the operation (in this case an instance of asdf:test-op
) and the
system (in this case, an instance of asdf:system
which will be equal to
(asdf:find-system "foo/test")
.) We don't actually need these arguments, so we'll just
leave them unused.
Create a subdirectory test
within your project's root directory. Inside it, create a
file test.lisp
to hold your tests. (If you used a different :pathname
or :file
in
your system definition, use those names instead.)
In test.lisp
, put the following forms:
(defpackage #:foo/test
(:use #:coalton #:coalton-prelude #:coalton-testing)
(:export #:run-tests))
(in-package #:foo/test)
(named-readtables:in-readtable coalton:coalton)
(fiasco:define-test-package #:foo/fiasco-test-package)
(coalton-fiasco-init #:foo/fiasco-test-package)
(cl:defun run-tests ()
(fiasco:run-package-tests
:packages '(#:foo/fiasco-test-package)
:interactive cl:t))
The defpackage
, in-package
and named-readtables:in-readtable
forms should all be
familiar to you. Add any clauses to your defpackage
that you need to reference your
system, like (:local-nicknames (#:f #:foo))
or (:use #:foo)
. Also add any
:local-nicknames
for coalton-library
packages you want.
We've added one new package to our :use
list, #:coalton-testing
. It defines the
utilities you need for defining tests and making assertions within them.
Coalton uses Fiasco for its testing
framework. Fiasco groups tests into packages, so we need to define a package for our
tests, but we don't actually want to write our code in that package: it's intended for
Common Lisp, so it :use
s #:cl
and #:fiasco
, but we'd much prefer to write our tests
in Coalton. So we define a Fiasco test package named FOO/FIASCO-TEST-PACKAGE
, and use
coalton-fiasco-init
to arrange for defining Coalton tests which Fiasco can find in that
package.
We told ASDF to run our tests by calling the run-tests
function, so we define
run-tests
. It just tells Fiasco to run our test suite.
Now that all that's done, we can finally get to the good part. Define a test with
define-test
, and inside it, make some assertions with is
or matches
.
define-test
is a macro that was generated by the coalton-fiasco-init
form you wrote
earlier. (This is an implementation detail; try not to think too much about it.) For your
very first define-test
, try writing:
(define-test my-first-test ()
(is True))
The is
macro asserts that its first argument evaluates to true. (is True)
will always
succeed, and (is False)
will always fail.
If you have a function always-returns-zero
in your package foo
, and you want to make
sure it returns 0
, you might write:
(define-test test-always-returns-zero ()
(is (== 0 (foo:always-returns-zero))))
When calling boolean predicates in an is
form, like ==
, prefer writing the expected
value as the first operand, and the computed value as the second.
TODO: coalton-testing
doesn't actually privilege that argument order the way FiveAM and
(I assume) Fiasco do. Should we recommend it anyway? I still think it's a nice
style, personally.
is
optionally takes a second argument, a string used to describe the assertion if it
fails. For example, you might write:
(is (== 0 (foo:always-returns-zero))
"ALWAYS-RETURNS-ZERO returned a non-zero value!")
The matches
macro takes two required arguments, a pattern and an expression, and asserts
that the result of the expression matches the pattern. (matches _ ANYTHING)
will always
succeed (assuming anything
compiles and executes without error).
If you have a function one-element-list
which wraps its argument in a one-element list
in your package foo
, you might write:
(define-test test-one-element-list ()
(matches (Cons _ (Nil))
(foo:one-element-list 0)))
Note that this doesn't assert anything about the contents of the list returned my
(foo:one-element-list 0)
, only its structure. If (foo:one-element-list 0)
returns the
list (Cons 0 (Nil))
, the test will pass just the same as if it returned (Cons "zero" (Nil))
.
Like is
, matches
takes an optional string argument used to describe the assertion if
it fails. You might write:
(matches (Cons _ (Nil))
(foo:one-element-list 0)
"ONE-ELEMENT-LIST returned a list with a length other than 1!")
Often, it's useful to invoke a function, destructure its return with match
, and to run
additional assertions in some branches while treating the other branches as failures. In
this case, it's convenient to write an (is False)
assertion with a message describing
the branch incorrectly taken in each of the failed branches. For example, if our
one-element-list
function from above is supposed to wrap specifically its argument in a
list, we'll want to add an ==
assertion. We might write:
(define-test test-one-element-list ()
(match (one-element-list 0)
((Nil) (is False "ONE-ELEMENT-LIST returned an empty list!"))
((Cons elt (Nil)) (is (== 0 elt)))
((Cons _ _) (is False "ONE-ELEMENT-LIST returned a list with more than 1 element!"))))
Now that you've defined a test system and some tests to go in it, you can run your tests
in the REPL using (asdf:test-system "foo")
. Try to avoid doing this while the REPL is in
a Coalton package, or any package that doesn't :use #:cl
- there have been some reports
of breakage when running ASDF operations when *package*
doesn't contain some CL
symbols. If you don't want to change the REPL package, you can evaluate:
(cl:let ((cl:*package* (cl:find-package "CL")))
(asdf:test-system "foo"))