lisp-unit2

2016-05-31

lisp-unit2

lisp-unit2 is a Common Lisp library that supports unit testing. It is a new version of a library of the lisp-unit library written by Chris Riesbeck. There is a long history of testing packages in Lisp, usually called "regression" testers. More recent packages in Lisp and other languages have been inspired by JUnit for Java.

Recently longtime users at Acceleration.net felt motivated to refactor significantly, attempting to make some broad improvements to the library while maintaining its benefits and workflow

Features

  • Written in portable Common Lisp
  • Loadable with ASDF or Quicklisp
  • Simple to define and run tests
  • Redefine functions and macros without reloading tests - tests are recompiled before each run and at definition time
  • Tests have source-location for going to definition
  • Supports testing: return values, printed output, macro expansions, and conditions. Special support for testing floating and rational point numbers
  • Running a single test returns a test-result object. Running many tests returns a test-result-db
  • Tests grouped and accessed by name, tag, and package of test name
  • Signals for starting and completing test runs (both individually and as a group)

Differences from lisp-unit version 1

  • Simplified test retrieval / categorization.
  • Tests are stored by their name's symbol-package (easing the complexity of this package vs the package argument)
  • package arguments now just retrieve all tests in a given package
  • dynamic variable consolidation (eg: one test database with three hashtables instead of three dynamic variables containing hashtables)
  • All output goes to its own stream (and can therefore be manipulated at will). test-stream defaults to standard-output as it always has.
  • Debugging is controlled by debug-hook, continue restart activated around test
  • Tests are named functions that show up on the stack (so that goto-definition in a stack trace brings you to the test definition
  • Contexts can be applied on a test or a run-test(s) to provide common setup/tear down locations (eg: provide a database to tests that need it). Contexts are functions that accept a thunk and apply it inside of a dynamic context eg: (lambda (body-thunk) (let ((*var* 1))(funcall body-thunk)))
  • Signals used throughout (including to drive current output summaries)
  • assertions passing and failing (with abort restart, to cancel assertions)
  • tests starting and completing (with continue restart for resuming test run after an errors)
  • test batches starting and completing
  • Dynamic variables available in signal handlers and all tests unit-test and results
  • Logging used throughout (to ease debugging of lisp-unit2) - compile time
  • TAP output displays better messages about the error and is more consistent with the summary output
  • Better job of reporting compiler warnings and errors: when defining the test (and while running it)
  • Summary output prints test starting messages. Compiler / run-time errors printed into the output is obviously tied to the test in question. (Previously the results printed after the error messages and so it was textually ambiguous to which test the messages applied)
  • lisp-unit2 is no longer loadable as a single file
  • :lisp-unit2 feature is available for conditional compilation
  • Better optimized for running both in a build-environment (jenkins etc) and from the REPL (previously there were many non-standard boolean flags for controlling output and debugging).
  • test bodies compiled with (debug 3) - TODO: revisit this in light of slow tests if that becomes an issue
  • Default ordering for tests and results is the definition order (instead of random or reversed)
  • assert-result -> assert-passes?
  • (define-test name &body) became (define-test name (&key) &body)

How to use lisp-unit2

  1. Load using Quicklisp : (ql:quickload :lisp-unit2) or ASDF : (asdf:load-system :lisp-unit2).
  2. Define some tests (for best luck define tests in their own package by making their name be in a specific package). By having tests in their own package, the test and the fn being tested can share the same name. (Tests are compiled to a function named after the test that runs it and an object in the test database)
(lisp-unit2:define-test my-tests::test-subtraction 
    (:tags '(my-tests::bar))
  (assert-eql 1 (- 2 1)))
  1. Run all tests in your package
;; with-summary provides results while the tests are running
;;;; using the context function
(run-tests :package :my-tests
           :run-contexts #'with-summary-context)

;;;; using the context macro (does the same thing as
;;;; :run-contexts, See Contexts below for an explanation).
(with-summary () (run-tests :package :my-tests))

;; to print the results after the run is complete
(print-summary (run-tests :tags 'my-tests::bar))

;; The difference is in whether or not the output occurs while the
;; tests are running or after all the tests have run

;; to disable the debugger:
(let (*debugger-hook*)
  (run-tests :tests 'my-tests::test-subtraction))

;; to debug failed assertions with the context function
(run-tests :tests 'my-tests::test-subtraction
           :run-contexts #'with-failure-debugging-context)

;; or use the context macro
(with-failure-debugging ()
  (run-tests :tests 'my-tests::test-subtraction))

See the internal test suite for more and better examples (internal-test/*)

Defining Tests

The define-test macro is the primary way to install a test. define-test creates a function with the same name as the test, which can be called to execute the test. This is nice because there is a direct way to call every test, but also because the test has implicit source destination and thus "go-to-definition" on any printing of the test name will take you directly to the test in question. define-test also creates a unit-test object and inserts it into the test-database.

The test body is compiled at the time of definition (so that any compile warnings or errors are immediately noticeable) and also before every run of the test (so that macro expansions are never out of date).

  • package: if package is provided the test name and all tags are reinterned into that package before definition. This is mostly for porting lisp-unit systems where test-names were not functions and so tests could be named the same as the function they tested.
  • contexts (see Contexts below): a (single or tree of) context fn that will be applied around the test body
  • tags: a list of symbols used to organize tests into groups (not hierarchical)

Undefining Tests

Because lisp-unit2 keeps a database of tests and functions, removing a tests is not as simple as deleteing the test from from your file. Instead a function uninstall-test and a macro undefine-test allow you to remove all the runtime components of a test from lisp-unit2. Uninstall-test accepts the test's symbolic name. Undefine-test is a macro whose form mimics define-test, so that you can simply add the two characters and compile the form to remove the test. This also gives you the ability to easily leave tests in your test file that are inactive currently.

Running Tests

The primary entry point for running tests is lisp-unit2:run-tests. run-tests and get-tests both accept (&key tests tags package reintern-package) as discussed below in "Test Organization". Each test can also be run individually by calling the function named for the test, and by calling lisp-unit2:run-test which each return a single test-result.

run-tests returns a test-results-db, and aside from the keys above accepts,

  • run-contexts: a (single, or list of) context that will be applied around the entire body of running the tests
  • test-contexts: a (single, or list of) context that will be applied around each tests body. This will be around any contexts defined on the unit test
  • name: a name used for this test run. Useful for identifying what defaults to all packages being tested. We attempt to make a good default if nothing is provided

Once you have a test-result-db (or list there of) you can call rerun-tests or rerun-failures to rerun and produce new results.

Test Organization: Names, Tags, and Packages

Tests are organized by their name and by tags. Both of these are symbols in some package. Tests can be retrieved by their name, the package that their name is in, and any of the tags that reference the test.

The most common way to retrieve and run unit tests is run-tests which calls get-tests.

(lisp-unit2:run-tests &key tests tags package reintern-package) (lisp-unit2:get-tests &key tests tags package reintern-package)

Both of these functions accept:

  • tests: a single or list of symbols or unit-test objects
  • tags: a single or list of symbols. All tests in these tags will be returned
  • package: a single or list of packages (names or objects)

If no arguments are provided lisp-unit2 will run all tests in package

In some cases, particularly when converting from lisp-unit(1) we need our tests to be in a different package (because tests are functions in their name's package). In lisp-unit these tests would not conflict with a function named the same (because tests were not functions). To ease conversion, the reintern-package argument will reintern all test names and tags provided into a different package. Define test accepts a package argument that mirrors this functionality. Suggested usage is to either have tests be named differently from the functions they test or to have tests and tags be in an explicitly referenced package, eg: (define-test my-tests::test1 (:tags '(my-tests::tag1)) ...)

Tests are organized into the *test-db* which is an instance test-database. These can be rebound if you needed to write tests about your test framework (see the internal example-tests).

Suites

While lisp-unit does not have any specific notion of a suite, it is believed that the tests are composable enough that explicit test suites are not needed.

One suggestion would be to have named functions that run you specific set of tests:

(defun run-{suite} ()
  (lisp-unit2:run-tests
    :name :{suite}
    {stuff to test} ))

(defun run-symbol-munger-and-lisp-unit2-tests () 
  (lisp-unit2:run-tests
    :name :symbol-munger-and-lisp-unit2-tests
    :package '(:symbol-munger-tests :lisp-unit2-tests)))

(defun run-error-and-basic-tests () 
  (lisp-unit2:run-tests
    :name :error-and-basic-tests
    :tests '(symbol-munger-test::test-basic)
    :tags '(:errors lisp-unit2-tests::errors)))

Another suggestion would be to define tests that call other tests:

(define-test suite-1 (:tags '(suites))
  (test-fn-1) ;; calls test-fn-1 unit-test
  (lisp-unit2::run-tests ...) ; runs an entire other set of unit tests)

Assertions

All assert-* macros signal assertion-pass and assertion-fail by comparing their expected results to the actual results during execution. All other values in the assert forms are assumed to be extra data to aid debugging.

assert-{equality-test} macros compare the actual to the expected value of each of the values returned from expected (see: multiple-values) (eg: (assert-eql exp act) => (eql act exp)) This ordering was used so that functions like typep, member, etc could be used as test.

assert-false and assert-true simply check the generic-boolean truth of their first arg (eg: null and (not null)

assert-{signal} and assert-{no-signal} macros are used for testing condition protocols. Signals that are expected/not-expected handled.

Multiple-values and assertions

Actual values are compared to all expected values, that is:

LISP-UNIT2-TESTS> (lisp-unit2:with-assertion-summary ()
                    (assert-eql (values 1 2) (values 1 2 3)))


<No Test>:Passed (ASSERT-EQL (VALUES 1 2) (VALUES 1 2 3))
T
LISP-UNIT2-TESTS> (lisp-unit2:with-assertion-summary ()
                    (assert-eql (values 1 2 3) (values 1 2)))

Failed Form: (ASSERT-EQL (VALUES 1 2 3) (VALUES 1 2))
Expected 1; 2; 3
but saw 1; 2

Debugging

Debugging is controlled by *debugger-hook* (as is usual in common-lisp). You can make lisp-unit simply record the error and move on by binding *debugger-hook* to nil around your run-tests call.

If you would like to debug failed assertions you can wrap your call in with-failure-debugging or apply the with-failure-debugging-context to the unit-test run.

Output and Results

All output is printed to *test-stream* (which by default is *standard-output*). Most forms do not output results by default, instead returning a result object. All results objects can be printed (to *test-stream*) by calling print-summary on the object in question.

print-summary prints information about passing as well as failing tests. print-failure-summary can be called to print only messages about failures, warnings, errors and empty tests (empty tests had no assertions). Care is taken to print tests with their short, but still fully packaged symbol-name (so that go-to-definition) works.

When running interactively with-summary(-context) can provide real-time output, printing start messages and result messages for each test. with-assertion-summary(-context) provides even more detailed output printing a message for each assertion passed or failed.

Test results (from one or many runs) can be captured using with-test-results. The arg: collection-place will copy all the results as they arrive into a location of your choosing (eg: variable, object slot). The arg: summarize? will print a failure summary of each test-run after all of the tests are finished running. This is useful for collecting separate results for many packages or systems (see test-asdf-system-recursive). If no args are provided summarize? is defaulted to true.

TAP Output

Lisp-unit2 provides TAP (test anything protocol) test results (printed in such a way that jenkins tap plugin can parse them).

with-tap-summary prints tap results as the tests run write-tap, write-tap-to-file accept a test-results database and write the TAP results either to test-stream or a file

ASDF

asdf:test-system is assumed to be the canonical way of testing a every system, and so lisp-unit2 makes effort to work well with test-system

Here is an example asdf:test-sytem definition, which will print the verbose summary of the tests as they run.

(defmethod asdf:perform ((o asdf:test-op) (c (eql (find-system :symbol-munger))))
  (asdf:oos 'asdf:load-op :symbol-munger-test)
  (let ((*package* (find-package :symbol-munger-test)))
    (eval (read-from-string "
            (lisp-unit2:with-summary ()
             (lisp-unit2:run-tests
              :package :symbol-munger-test
              :name :symbol-munger
              :run-context))
      "))))

Additionally, lisp-unit2 provides test-asdf-system-recursive which accepts a (list or single) system name, looks up all its dependencies and calls asdf:tests-system on each listed system. Any lisp-unit or lisp-unit2 test-results-dbs are collected and returned at the end. A failure summary is also printed for each result db (so that after running many tests you are presented with a short synopsis of what ran and what failed.

There is an interop layer for converting test results from other systems to lisp-unit2 test results, so that we can gather and summarize more information. Currently this is only implemented for lisp-unit1, but patches would be welcome to allow collecting test results from any other lisp test systems. (This is currently a bit tedious, simplification patches welcome as well, see interop.lisp).

Signals

Signals are used throughout lisp-unit2 to communicate progress and results throughout lisp unit

  • missing-test: when asked to run a test that does not exist
  • test-start: when a unit test starts, contains a ref to the unit-test
  • test-complete: when a test completes contains a ref to results for that test (test-result)
  • all-tests-start, all-tests-complete: signaled at the beginning and end of run-tests, contains a ref to the test-results-db
  • collect-test-results: used in it interop to send results to with-test-results
  • assertion-pass: signaled when an assertion passes
  • assertion-fail: signaled when an assertion fails

All signals have an abort interrupt which simply cancels the signal. This is mostly used for meta-testing (ie: testing lisp-unit2), but there are conceivably other uses.

Contexts / Fixtures

Contexts allow you to manipulate the dynamic state of a given unit-test or test-run. These are functions of a single required argument (a thunk), that they execute inside of a new/changed dynamic environment.

A good example is the with-failure-debugging-context, which simply invokes the debugger whenever we signal a failed assertion

(defun with-failure-debugging-context (body-fn)
  "A context that invokes the debugger on failed assertions"
  (handler-bind ((assertion-fail #'invoke-debugger))
    (funcall body-fn)))

Both define-test and run-tests accept a tree of contexts that is flattened and turned into a single context function (see combine-contexts). This function then executes the body-fn for us (see: do-contexts).

This should allow any type of manipulation of the current lisp-unit2 environment through access to handling signals and setting up and tearing down dynamic environments.

Example Contexts:

  • rolled-back-database-context: open a connection to the dev database and a transaction that we will abort at the end
  • http-context: establish a dummy http context for use in testing web connected things
  • html-output-context: establish global html building objects needed for html output functions to work
  • dev-mode-context: bind any dynamic variables related to DEV/LIVE mode
  • failure-debugging-context: (as above) handling signals and debugging them instead of only recording them

Example test-with-contexts defintion:

(defmacro db-render-test (name (&rest args) &body body)
  `(lisp-unit2:define-test ,name
    (:tags ',args
     :contexts
     (list #'test-context #'dom-context #'database-context )
     :package :test-objects)
    ,@body))

Data Model

  • test-database: a list and some indexes of all installed tests
  • unit-test: an object containing the name, code, contexts, and documentation for a single unit-test
  • test-results-db: a collection of results from a single run-tests call
  • test-result: the result of running a single unit test
  • failure-result: information useful for debugging failures (eg: unit-test, the form that failed, the expected and actual results)

Remaining Tasks

  • Expanded internal testing.

Future Features

  • Benchmarking tools
  • More interop with other test systems
  • Database layer to store test results to a database (so that we can track bugs / test durations) over time (goes with benchmarking)

0.2.0 Acknowledgments

  • Russ Tyndall - Acceleration.net

0.9.5 Acknowledgments

Author
Russ Tyndall <russ@acceleration.net>, Acceleration.net, Thomas M. Hermann <thomas.m.hermann@odonata-research.com>
License
MIT
Categories
accelerationnet, test framework