• Fixtures
• Test groups
• Testing values
• Testing processes
• Testing invariants against sampled data
• Defining test criteria
• The runtime system
• Integration with ASDF
• Inheritance-based test methods
• Deprecated forms

This document is the manual and users' guide to the 4.1.(x) series of the NST test framework, last updated for 4.1.0. NST is a unit test system for Common Lisp which provides support for test fixture data, stateful setup and cleanup of tests, grouping of tests, and (we think!) a useful runtime interface. Suggestions and comments are welcome. The files in the NST distribution's self-test directory, especially self-test/core/builtin-checks.lisp, holds the NST tests for NST and contain many examples (some of which we have adapted for this manual). Known bugs and infelicities, platform-specific release notes, and other technical materials are available via the link on NST's CLiki page, cliki.net/NST.

Contributors. The primary author of both NST and this manual is John Maraist (Smart Information Flow Technologies, 211 North First Street, Suite 300, Minneapolis, MN 55401; jmaraist at sift.info). Robert P. Goldman provided guidance, comments and suggestions through the development. Other contributors include Michael J.S. Pelican, Steven A. Harp, Michael Atighetchi and Patrick Stein.

Fixtures

Fixtures are data structures and values which may be referred to by name during testing. NST provides the ability to use fixtures across multiple tests and test groups, and to inject fixtures into the runtime namespace for debugging. A set of fixtures is defined using the def-fixtures macro:

(def-fixtures fixture-name ( [ :outer FORM ]
                             [ :inner FORM ]
                             [ :setup FORM ]
                             [ :cleanup FORM ]
                             [ :startup FORM ]
                             [ :finish FORM ]
                             [ :documentation STRING ]
                             [ :cache FLAG ]
                             [ :export-names FLAG ]
                             [ :export-fixture-name FLAG ]
                             [ :export-bound-names FLAG ] )
  ( [ ( [ :cache FLAG ] ) ] NAME [ FORM ] )
  ...
  ( [ ( [ :cache FLAG ] ) ] NAME [ FORM ] ) )

• fixture-name: The name to be associated with this set of fixtures.
• inner: List of declarations to be made inside the let-binding of names of any use of this fixture. Do not include the "declare" keyword here; NST adds these declarations to others, including a special declaration of all bound names.
• outer: List of declarations to be made outside the let-binding of names of any use of this fixture.
• documentation: A documentation string for the fixture set.
• special: Specifies a list of names which should be declared special in the scope within which this set's fixtures are evaluated. The individual names are taken to be single variable names. Each (:fixture NAME) specifies all of the names of the given fixture set. This declaration is generally optional under most platforms, but can help supress spurious warnings. Note that multiple NAMEs may be listed, and these lists and the bare names may be intermixed. If only one name or fixture is specified, it need not be placed in a list
• export-fixture-name: When non-nil, the fixture name will be added to the list of symbols exported by the current package.
• export-bound-names: When non-nil, the names bound by this fixture will be added to the list of symbols exported by the current package.
• export-names: When non-nil, sets the default value to t for the two options above.
• cache: If specified with the group options, when non-nil, the fixture values are cached at their first use, and re-applied at subsequent fixture application rather than being recalculated.

When a fixture is attached to a test or test group, each NAME defined in that fixture becomes available in the body of that test or group as if let* bound to the corresponding FORM. A fixture in one set may refer back to other fixtures in the same set (again as let*) but forward references are not allowed.

The four arguments startup, finish, setup and cleanup specify forms which are run everytime the fixture is applied to a group or test. The startup (respectively finish) form is run before fixtures are bound (after their bindings are released). These forms are useful, for example, to initialize a database connection from which the fixture values are drawn. The setup form is run after inclusion of names from fixture sets, but before any tests from the group. The cleanup form is normally run after the test completes, but while the fixtures are still in scope. Normally, the cleanup form will not be run if the setup form raises an error, and the finish form will not be run if the startup form raises an error; although the user is able to select (perhaps unwisely) a restart which disregards the error.

The names of a fixture and the names it binds can be exported from the package where the fixture is defined using the export-bound-names and export-fixture-name arguments. The default value of both is the value of export-names, whose default value is nil.

The cache option, if non-nil, directs NST to evaluate a fixture's form one single time, and re-use the resulting value on subsequent applications of the fixture. Note that if this value is mutated by the test cases, test behavior may become unpredictable! However this option can considerably improve performance when constant-valued fixtures are applied repeatedly. Caching may be set on or off (the default is off) for the entire fixture set, and the setting may vary for individual fixtures.

Examples of fixture definitions:

(def-fixtures f1 ()
  (c 3)
  (d 'asdfg))
(def-fixtures f2 (:special ((:fixture f1)))
  (d 4)
  (e 'asdfg)
  (f c))
(def-fixtures f3 ()
  ((:cache t)   g (ackermann 1 2))
  ((:cache nil) h (factorial 5)))

To cause a side-effect among the evaluation of a fixture's name definitions, nil can be provided as a fixture name. In uses of the fixture, NST will replace nil with a non-interned symbol; in documentation such as form :whatis, any =nil=s are omitted.

The with-fixtures macro faciliates debugging and other non-NST uses of fixtures sets:

(with-fixtures (FIXTURE FIXTURE ... FIXTURE)
  FORM
  FORM
  ...
  FORM)

This macro evaluates the forms in a namespace expanded with the bindings provided by the fixtures.

Test groups

The def-test-group form defines a group of the given name, providing one instantiation of the bindings of the given fixtures to each test. Groups can be associated with fixture sets, stateful initialization, and stateful cleanup.

(def-test-group NAME (FIXTURE FIXTURE ...)
  (:aspirational FLAG)
  (:setup FORM FORM ... FORM)
  (:cleanup FORM FORM ... FORM)
  (:startup FORM FORM ... FORM)
  (:finish FORM FORM ... FORM)
  (:each-setup FORM FORM ... FORM)
  (:each-cleanup FORM FORM ... FORM)
  (:include-groups GROUP GROUP ... GROUP)
  (:documentation STRING)
  TEST
  ...
  TEST)

Arguments:

• group-name: Name of the test group being defined
• given-fixtures: List of the names of fixtures and anonymous fixtures to be used with the tests in this group.
• aspirational: An aspirational test is one which verifies some part of an API or code contract which may not yet be implemented. When a group is marked aspirational, all tests within the group are taken to be aspirational as well. At this point, there is no particular processing for aspirational tests and groups, but we hope to implement it at some point in the future.
• forms: Zero or more test forms, given by def-check.
• setup: These forms are run once, before any of the individual tests, but after the fixture names are bound.
• cleanup: These forms are run once, after all of the individual tests, but while the fixture names are still bound.
• startup: These forms are run once, before any of the individual tests and before the fixture names are bound.
• finish: These forms are run once, after all of the individual tests, and after the scope of the bindings to fixture names.
• each-setup: These forms are run before each individual test.
• each-cleanup: These forms are run after each individual test.
• include-group: The test groups named in this form will be run (respectively reported) anytime this group is run (reported).
• documentation: Docstring for the class.

Testing values

Individual unit tests are encoded with the def-test form:

(def-test NAME ( [ :group GROUP-NAME ]
                  [ :setup FORM ]
                  [ :cleanup FORM ]
                  [ :startup FORM ]
                  [ :finish FORM ]
                  [ :fixtures (FIXTURE FIXTURE ... FIXTURE) ]
                  [ :aspirational FLAG ]
                  [ :documentation STRING ] )
    criterion
 FORM ... FORM)

(def-test NAME criterion &body (:seq FORM))

The SETUP, CLEANUP, STARTUP, FINISH and FIXTURES are just as for fixtures and test groups, but apply only to the one test. The CRITERION is a list or symbol specifying the properties which should hold for the =FORM=s.

When a test is not enclosed within a group body, a group name must be provided by the GROUP option. When a test is enclosed within a group body, the GROUP option is not required, but if provided it must agree with the group name.

When there are no SETUP, CLEANUP, STARTUP, FINISH or FIXTURES arguments, the NAME may be given without parentheses. Likewise, any criterion consisting of a single symbol, e.g. (:pass), may be abbreviated as just the symbol without the parentheses, e.g. :pass.

The :documentation form provides a documentation string in the standard Lisp sense. Since documentation strings are stored against names, and since the same name can be used for several tests (so long as they are all in different packages), documentation strings on tests may not be particularly useful.

An aspirational test is one which verifies some part of an API or code contract which may not yet be implemented. When a group is marked aspirational, all tests within the group are taken to be aspirational as well. At this point, there is no particular processing for aspirational tests and groups, but we hope to implement it at some point in the future.

The def-check form is a deprecated synonym for def-test.

Basic criteria

Criterion true

The :true criterion expects one form, which is evaluated at testing time; the criterion requires the result to be non-nil.

Criterion eq

The :eq criterion checks a form using eq. The criterion argument and the form under test are both evaluated at testing time.

Usage:

(:eq TARGET)

Example:

(def-test eq1 (:eq 'b) (cadr '(a b c)))

Criterion symbol

The :symbol criterion checks that its form under test evaluates to a symbol which is eq to the symbol name given as the criterion argument.

Usage:

(:symbol NAME)

Example:

(def-test sym1  (:symbol a) (car '(a b c)))

Example of a test which fails:

(def-test sym1x (:symbol a) (cadr '(a b c)))

Criterion eql

The :eql criterion checks a form using eql. The criterion argument and the form under test are both evaluated at testing time.

Usage:

(:eql TARGET)

Example:

(def-test eql1 (:eql 2) (cadr '(1 2 3)))

Criterion equal

The :equal criterion checks a form using eql. The criterion argument and the form under test are both evaluated at testing time.

Criterion equalp

The :equalp criterion checks a form using equalp. The criterion argument and the form under test are both evaluated at testing time.

Usage:

(:equalp TARGET)

Criterion forms-eq

The :forms-eq criterion compares its two forms under test using eq. The forms are both evaluated at testing time.

Usage:

(:forms-eq)

Example:

(def-test eqforms1 :forms-eq (cadr '(a b c)) (caddr '(a c b)))

Criterion forms-eql

The :forms-eql criterion compares its two forms under test using eql. The two forms under test are both evaluated at testing time.

Usage:

(:forms-eql)

Example:

(def-test eqlforms1 :forms-eql (cadr '(a 3 c)) (caddr '(a c 3)))

Criterion forms-equal

The :forms-equal criterion compares its two forms under test using equal. The forms are both evaluated at testing time.

Usage:

(:forms-equal)

Criterion predicate

The :predicate criterion applies a predicate to the result of evaluating its form under test. The criterion argument is a symbol (unquoted) or a lambda expression; at testing time, the forms under test are evaluated and passed to the denoted function. The criterion expects that the result of the function is non-nil.

Usage:

(:predicate PRED)

Example:

(def-test pred1 (:predicate numberp) 3)

Example of a test which fails:

(def-test pred2 (:predicate eql) (+ 1 2) 3)

Criterion err

The :err criterion evaluates the form under test, expecting the evaluation to raise some condition. If the class argument is supplied, the criterion expects the raised condition to be a subclass. Note that the name of the type should not be quoted; it is not evaluated.

Usage:

(:err &KEY (TYPE CLASS))

Examples:

(def-test err1 (:err :type error) (error "this should be caught"))
(def-test err2 (:err) (error "this should be caught"))

Criterion perf

The :perf criterion evaluates the forms under test at testing time, checking that the evaluation completes within the given time limit.

Usage:

(:perf &KEY (MS MILLISECS) (SEC SECONDS) (MIN MINUTES))

Example:

(def-test perf1 (:perf :min 2) (ack 3 5))

Compound criteria

Criterion not

The :not criterion passes when testing according to subcriterion fails (but does not throw an error).

Usage:

(:not SUBCRITERION)

Example:

(def-test not1 (:not (:symbol b)) 'a)

Criterion all

The :all criterion brings several other criteria under one check, and verifies that they all pass.

Usage:

(:all SUBCRITERION ... SUBCRITERION)

Example:

(def-check not1 ()
    (:all (:predicate even-p)
          (:predicate prime-p))
  2)

Criterion any

The :any criterion passes when any of the subordinate criteria pass.

Usage:

(:any SUBCRITERION ... SUBCRITERION)

Example:

(def-check not1 ()
    (:any (:predicate even-p)
          (:predicate prime-p))
  5)

Criterion apply

The :apply criterion first evaluates the forms under test, applying function to them. The overall criterion passes or fails exactly when the subordinate criterion with the application's multiple result values.

Usage:

(:apply FUNCTION CRITERION)

Example:

(def-test applycheck (:apply cadr (:eql 10)) '(0 10 20))

Criterion check-err

The :check-err criterion is like :err, but proceeds according to the subordinate criterion rather than simply evaluating the input forms.

Usage:

(:check-err CRITERION)

Example:

(def-test check-err1
    (:check-err :forms-eq)
  'asdfgh (error "this should be caught"))

Criterion progn

The :progn criterion first evaluates the =FORM=s in order, and then proceeds with evaluation of the forms under test according to the subordinate criterion.

Usage:

(:progn FORM ... FORM SUBCRITERION)

Example:

(def-test form1 (:progn (setf zz 3) (:eql 3)) zz)

Criterion proj

The :proj criterion rearranges the forms under test by selecting a new list according to the index numbers into the old list. Checking of the reorganized forms continues according to the subordinate criterion.

Usage:

(:proj INDICES CRITERION)

Example:

(def-test proj-1
    (:proj (0 2) :forms-eq)
  'a 3 (car '(a b)))

Criterion with-common-criterion

The :with-common-criterion criterion applies one criterion to several data forms.

Usage:

(:with-common-criterion [ CRITERION | (CRITERION ARG ... ARG) ]
                        (FORM ...  FORM) ... (FORM ...  FORM) )

Criterion applying-common-criterion

The :applying-common-criterion criterion applies one criterion to several pairs of criterion arguments and data forms.

Usage:

(:applying-common-criterion [ CRITERION | (CRITERION ARG ... ARG) ]
                            ( ((ARG ... ARG) (FORM ... FORM))
                              ...
                              ((ARG ... ARG) (FORM ... FORM)) )
                            ...
                            ( ((ARG ... ARG) (FORM ... FORM))
                              ...
                              ((ARG ... ARG) (FORM ... FORM)) ) )

Criteria for multiple values

Criterion values

The :values criterion checks each of the forms under test according to the respective subordinate criterion.

Usage:

(:values SUBCRITERION ... SUBCRITERION)

Criterion value-list

The :value-list criterion converts multiple values into a single list value.

Usage:

(:value-list FURTHER)

Criterion drop-values

The :drop-values criterion checks the primary value according to the subordinate criterion, ignoring any additional returned values from the evaluation of the form under test.

Usage:

(:drop-values CRITERION)

Criteria for lists

Criterion seq

The :seq criterion evaluates its input form, checks each of its elements according to the respective subordinate criterion, and passes when all of them pass.

Usage:

(:seq SUBCRITERION ... SUBCRITERION)

Example:

(def-check seqcheck
    (:seq (:predicate symbolp) (:eql 1) (:symbol d))
  '(a 1 d))

Criterion each

The :each criterion evaluates the form under test, expecting to find a list as a result. Expects that each argument of the list according to the subordinate criterion, and passes when all of these checks pass.

Usage:

(:each CRITERION)

Example:

(def-test each1 (:each (:symbol a)) '(a a a a a))

Criterion permute

The :permute criterion evaluates the form under test, expecting to find a list as a result. The criterion expects to find that some permutation of this list will satisfy the subordinate criterion.

Usage:

(:permute CRITERION)

Example:

(def-test permute1 (:permute (:each (:eq 'a))) '(a a))
(def-check permute2
    (:permute (:seq (:symbol b)
                    (:predicate symbolp)
                    (:predicate numberp)))
  '(1 a b))

Criterion alist

The :alist criterion evaluates the form under test, expecting to find an association list as a result. Using the two given function specs to test the keys (during retrieval, via assoc) and the values, the criterion enforces that the association lists contains exactly equivalent keys, mapping to respective equivalent values. Implemented using :alist* plus a check of the list length, which could be incorrect if the criterion lists duplicate keys.

Usage:

(:alist KEY-TEST-FN VALUE-TEST-FN (KEY VALUE) ... (KEY VALUE))

Criteria for vectors

Criterion across

The :across criterion is like :seq, but for a vector instead of a list.

Usage:

(:across SUBCRITERION ... SUBCRITERION)

Example:

(def-check across1
    (:across (:predicate symbolp) (:eql 1))
  (vector 'a 1))

Criteria for classes

Criterion slots

The :slots criterion evaluates its input form, and passes when the value at each given slot satisfies the corresponding subordinate constraint.

Usage:

(:slots (SLOT-NAME SUBCRITERION) ... (SLOT-NAME SUBCRITERION))

Example:

(defclass classcheck ()
  ((s1 :initarg :s1 :reader get-s1)
   (s2 :initarg :s2)
   (s3 :initarg :s3)))
 (def-test slot1
     (:slots (s1 (:eql 10))
             (s2 (:symbol zz))
             (s3 (:seq (:symbol q) (:symbol w)
                       (:symbol e) (:symbol r))))
   (make-instance 'classcheck
     :s1 10 :s2 'zz :s3 '(q w e r)))

Programmatic and debugging criteria

Criterion pass

The :pass is a trivial test which always passes.

Example:

(def-test passing-test :pass 3 4 "sd")

Criterion warn

The :warn criterion issues a warning. The format string and arguments should be suitable for the Lisp format function.

Usage:

(:warn FORMAT-STRING FORM ... FORM)

Example:

(:warn "~{}d is not a perfect square" 5)

Criterion info

The :info criterion adds an informational note to the check result.

Usage:

(:info STRING SUBCRITERION)

Example:

(def-test known-bug (:info "Known bug" (:eql 3)) 4)

Testing processes

The test criteria of the previous section all examined the result of evaluating the forms under test. This section presents NST's criteria for validating the process of a computation, specifying assertions which should hold at the initial, intermediate and final points of the process.

Criterion eval

(def-criterion ( [ :check-warnings FLAG ] [ :muffle-warnings FLAG ]
                 [ :attempt-continue FLAG ] [ :force-continue FLAG ] )
  FORM
  ...
  FORM)

The :eval criterion executes its forms, expecting calls to various assertion functions to check intermediate states of an arbitrarily-long process.

• check-warnings: If non-nil, will add warnings thrown when evaluating the forms under test as NST warnings. The default is t.
• muffle-warnings: If non-nil, will muffle warnings thrown when evaluating the forms under test, so that they are reported only as NST result warnings and if the :check-warnings flag is set. The default is t.
• attempt-continue: If non-nil, will continue evaluation after failed assertions, so long as the failure is not deemed fatal. The default is t.
• force-continue: If non-nil, will continue evaluation after failed assertions even if the failure is not deemed fatal. The default is nil.

Defining tests

The def-eval-test macro abbreviates a call to def-test with a single :eval criterion. Its arguments are just as for def-test and :eval.

(def-eval-test (NAME [ :group GROUP-NAME ]
                     [ :setup FORM ]
                     [ :cleanup FORM ]
                     [ :startup FORM ]
                     [ :finish FORM ]
                     [ :fixtures (FIXTURE ... FIXTURE) ]
                     [ :documentation STRING ]
                     [ :check-warnings FLAG ]
                     [ :muffle-warnings FLAG ]
                     [ :attempt-continue FLAG ]
                     [ :force-continue FLAG ] )
  FORM
  ...
  FORM)

(def-eval-test NAME
  FORM
  ...
  FORM)

Asserting properties

The assert-non-nil function is a unary predicate for use within the forms evaluated for an :eval criterion. It succeeds whenever the null function returns nil.

The assert-not-eq function is a unary predicate for use within the forms evaluated for an :eval criterion. It compares the expected and tested values using eq, succeeds whenever that call returns nil.

The assert-not-equalp function is a unary predicate for use within the forms evaluated for an :eval criterion. It compares the expected and tested values using equalp, succeeds whenever that call returns nil.

The assert-null function is a unary predicate for use within the forms evaluated for an :eval criterion. It succeeds whenever the null function returns non-nil.

The assert-criterion macro asserts that an NST criterion should pass.

(assert-criterion ( [ :msg-format FORMAT-STRING ]
                    [ :msg-args FORMAT-ARGUMENTS ]
                    [ :fatal FLAG [
                    [ :fail-on-warning FLAG ] )
  FORM
  ...
  FORM)

• msg-format: Format string used to build the label of the restart point.
• msg-args: Format arguments used to build the label of the restart point.
• fatal: If non-nil, a failure of this assertion indicates that execution of the test forms should be aborted.
• fail-on-warning: If non-nil, then an NST result which includes a warning indicates failure of this assertion.

The assert-equal function is a unary predicate for use within the forms evaluated for an :eval criterion. It compares the expected and tested values using equal, succeeds whenever that call returns non-nil.

The assert-zero function is a unary predicate for use within the forms evaluated for an :eval criterion. It succeeds whenever the zerop function returns non-nil.

The assert-equalp function is a unary predicate for use within the forms evaluated for an :eval criterion. It compares the expected and tested values using equalp, succeeds whenever that call returns non-nil.

The assert-not-equal function is a unary predicate for use within the forms evaluated for an :eval criterion. It compares the expected and tested values using equal, succeeds whenever that call returns nil.

The assert-eq function is a unary predicate for use within the forms evaluated for an :eval criterion. It compares the expected and tested values using eq, succeeds whenever that call returns non-nil.

The assert-eql function is a unary predicate for use within the forms evaluated for an :eval criterion. It compares the expected and tested values using eql, succeeds whenever that call returns non-nil.

The assert-not-eql function is a unary predicate for use within the forms evaluated for an :eval criterion. It compares the expected and tested values using eql, succeeds whenever that call returns nil.

Defining new assertion functions

Macro macro def-unary-predicate-assert creates an assertion function using the result of a call to a unary predicate. A non-nil result from the predicate corresponds to a successful assertion.

(def-unary-predicate-assert ASSERT-FN PREDICATE DEFAULT-MESSAGE
                            [ :message-defvar NAME ]
                            [ :pred-name NAME ]
                            [ :doc-state-flag BOOL ] )

• assert-fn: The name of the assertion function being defined.
• predicate: The predicate used to define the assertion function. It should take a single argument.
• default-message: Format string used by default for reporting failures of this assertion. It should expect to be used in a call to format with one additional argument, the value being tested.
• message-defvar: The name of a global variable into which the default message will be stored. If this argument is omitted, the result of a call to gensym is used.
• pred-name: This argument is used only for documenting the underlying predicate in the assertion function's docstring. By default, it is the same as the predicate.

Macro def-binary-predicate-assert uses a binary predicate as the basis for an assertion function just as def-unary-predicate-assert uses a unary predicate. This macro's arguments are just as for def-unary-predicate-assert.

Macro def-unary-negated-predicate-assert uses the negated result of a unary predicate as the basis of an assertion function. This macro's arguments are just as for def-unary-predicate-assert.

Macro def-binary-negated-predicate-assert uses the negated result of a binary predicate as the basis for an assertion function just as def-unary-negated-predicate-assert uses the negated result of a unary predicate. This macro's arguments are just as for def-unary-predicate-assert.

A simpler process checker

Criterion process

The :process criterion allows simple interleaving of Lisp function calls and NST checks, to allow checking of intermediate states of an arbitrarily-long process.

This criterion takes as its body a list of forms. The first element of each form should be a symbol:

• :eval: Heads a list of forms which should be evaluated.
• :check: Heads a list of criteria which should be checked.
• :failcheck: If checks to this point have generated any errors or failures, then the process criterion is aborted.
• :errcheck: If checks to this point have generated any errors (but not failures), then the process criterion is aborted.

The :process criterion takes no value arguments in a def-test.

For example:

(def-test process-1
    (:process (:eval (setf zzz 0))
              (:check (:true-form (eql zzz 0)))
              (:eval (incf zzz))
              (:check (:true-form (eql zzz 1)))
              (:eval (incf zzz))
              (:check (:true-form (eql zzz 2)))))

Testing invariants against sampled data

The sample criterion provides random generation of data for validating program properties. Our approach is based on Claessen and Hughes's Quickcheck (Koen Claessen and John Hughes, ``QuickCheck: a lightweight tool for random testing of Haskell programs,'' from Proceedings of the International Conference on Functional Programming, 2000. QuickCheck papers, code and other resources are available at www.cs.chalmers.se/~rjmh/QuickCheck).

This style of testing is somewhat more complicated than specific tests on single, bespoke forms. There are two distinct efforts, which we address in the next two sections: describing how the sample data is to be generated, and specifying the test itself.

Generating sample data

Data generation is centered around the generic function arbitrary.

This function takes a single argument, which determines the type of the value to be generated. For simple types, the name of the type (or the class object, such as returned by find-class by itself is a complete specification. For more complicated types, arbitrary can also take a list argument, where the first element gives the type and the remaining elements are keyword argument providing additional requirements for the generated value.

NST provides methods of arbitrary for many standard Lisp types. Non-scalar types are associated with additional keyword arguments for constraints on value generation.

• Standard numeric types number, real, rational, integer, float, fixnum, bignum, ratio, short-float, single-float, double-float, long-float, complex.

  These standard numeric types are not associated with additional keyword arguments.

  (nst:arbitrary t)
  (nst:arbitrary 'complex)
  (nst:arbitrary 'integer)
  (nst:arbitrary 'ratio)
  (nst:arbitrary 'single-float)

  Note that short-float, single-float, double-float are not available on Allegro Lisp.

• Types character and string:
  • Argument noncontrol. Excludes the control characters associated with ASCII code 0 through 31.

  • Argument range. Allows the range of characters to be restricted to a particular subset:

    Value	Meaning
    :standard	Codes up to 96
    :ascii	Codes through 127
    :ascii-ext	Codes through 255

    Omitted or with any other value, characters with any code up to char-code-limit can result. Examples:

    (nst:arbitrary 'character)
    (nst:arbitrary '(character :noncontrol t
                               :range :standard))

• Type symbol.
  • Argument existing. If non-nil, requires that the result be a previously-interned symbol.
  • Argument exported. Requires that the result be not only a previously-interned symbol, but also one exported by its package. Ignored if existing is explicitly set to nil.
  • Argument package. Specifies the package from which the symbol will be generated. If omitted, a package is selected at random from the existing ones.
  • Argument nonnull. If non-nil, allows arbitrary to ignore other restriction to guarantee returning a non-nil symbol. When null, arbitrary may return nil.
  • Argument gensym. If non-nil, and if arbitrary is explicitly set to nil, returns a new uninterned symbol.

• Type scalar. The common supertype of the above numeric types, character, string and symbol. This specifier is not associated with additional keyword arguments.

• Type cons.
  • Arguments car and cdr should be additional type specifications, used direct the generation of respectively the left and right elements of the result. Each defaults to t.
• Type list and vector.
  • Argument length specifies the length of the structure. If omitted, will be randomly generated.
  • Argument elem directs the generation of the container's elements. For both, the default element type is t.
• Type array.
  • Argument elem. As for list and vector.
  • Argument dimens. Should be a list of nonnegative integers specifying the length of each dimension of the array. If omitted, will be randomly generated.
  • Argument rank. Specifies the number of dimensions. If omitted but dimens is given, will be set to the length of dimens. If both rank and dimens are omitted, then both are randomly generated.
• Type hash-table.
  • Argument size. Specifies the number of entries in the table. If omitted, will be randomly generated.
  • Argument test. Specifies the hash table's test function. If omitted, will be randomly selected from eq, eql, equal and equalp.
  • Arguments key and val direct the generation of the table's keys and values, respectively. For the keys, the default element type is t when the test function is texttt{eq} or eql, and scalar otherwise. For the values, the default element type is t.

• Type t. The common supertype of all Lisp types. This specifier is not associated with additional keyword arguments.

Beyond those standard Lisp types, NST provides the type scalar as a supertype of the numeric types plus character, string and symbol. Users may extend this definition to include additional type specifications, as we discuss below. Types are not associated with scalar are referred to as compound (although there is no corresponding type specification). To avoid generating structures too large to hold in memory, NST provides the global variable *max-compound-structure-depth* and the macro compound-structure.

The *max-compound-structure-depth* variable sets the maximum nesting depth of compound data structures: beyond that depth, scalar rather than t is the default element generator. This restriction does not apply to explicitly specified element types, only to the use of defaults.

The compound-structure macro wraps substructure which should be considered compound for the limits set by *max-compound-structure-depth*.

New type specifications for invariant-testing. are defined with the def-arbitrary-instance-type macro.

(def-arbitrary-instance-type ( SPEC-NAME [ :params FORMALS ]
                                         [ :scalar BOOL ]
                                         [ :key KEY ] )
  FORM
  FORM
  ...
  FORM)

• formals: Formal parameter definition used to pass subcomponent types.

• scalar: When a non-null value is provided for the :scalar argument, the new specifier is taken to be generable by the :scalar specification.

  (def-arbitrary-instance-type (ratio :scalar t)
    (/ (arbitrary 'integer)
       (let ((raw (arbitrary (find-class 'integer))))
         (cond
           ((< raw 0) raw)
           (t (+ 1 raw))))))

• key: The :key argument gives a list of keyword arguments which may accompany the new specification. For the cons type, keyword arguments allow specifications for the left and right components:

  (def-arbitrary-instance-type
    (cons :key ((car t car-supp-p) (cdr t cdr-supp-p)))
    (compound-structure
     (when (and (not car-supp-p)
                (>= *current-compound-structure-depth*
                    *max-compound-structure-depth*))
       (setf car 'scalar))
     (when (and (not cdr-supp-p)
                (>= *current-compound-structure-depth*
                    *max-compound-structure-depth*))
       (setf cdr 'scalar))
     (cons (arbitrary car) (arbitrary cdr))))

• form: Construct and return (as if through progn the arbtrary instance.

Invariants as tests

Defining test criteria

The criteria used in test forms decide whether, when and how to use the forms under test and the forms and subcriteria provided to each test criterion. Criteria receive their arguments as forms, and may examine them with or without evaluation, as the particular criterion requires. NST provides two mechanisms for defining new criteria, and a number of support functions for use within these definitions. The simpler, but more limited, way to define a new criterion is by specifying how it should be rewritten to another criterion. The def-criterion-alias macro provides this mechanism, which we discuss in Section . The def-criterion macro provides the more general mechanism for criteria definition, where Lisp code produces a result report from the forms under test and criterion's forms and subcriteria. We discuss def-criterion in Section . We discuss the NST API for creating these result reports in Section , and for recursive processing of subcriteria in Section .

Aliases over criteria

The simplest mechanism for defining a new criterion involves simply defining one criterion to rewrite as another using def-criterion-alias.

(def-criterion-alias (name (:seq arg))
  [ doc-string ]
  expansion)

The body of the expansion should be a Lisp form which, when evaluated, returns an S-expression quoting the new criterion which the rewrite should produce. The arg are passed as for Lisp macros: they are not evaluated and are most typically comma-inserted into a backquoted result. For example:

(def-criterion-alias (:forms-eq) `(:predicate eq))
(def-criterion-alias (:symbol name) `(:eq ',name))

Reporting forms

NST provides functions both for building test reports, and for adding information to a report.

The make-success-report function indicates a successful test result.

(make-success-report)

Note that some older examples show (make-check-result), (emit-success) or (check-result). The former is an internal function and should not be used from outside the core NST files. The latter two are deprecated.

The make-failure-report function returns a report of test failure.

(make-failure-report FORMAT ARGS)

The format-string and args are as to the Common Lisp function format. The emit-failure function is an older, deprecated version of this function.

Function make-warning-report is like make-failure-report, but provides supplimentary information as a warning.

(make-warning-report [ :format FORMAT-STRING ] [ :args ARG-FORM-LIST ])

The emit-warning function is an older, deprecated version of this function.

Function make-error-report produces a report of an error during test execution.

(make-error-report [ :format FORMAT-STRING ] [ :args ARG-FORM-LIST ] )

For use within user-defined NST criteria: add an error to a result.

(add-error RESULT-REPORT [ :format FORMAT-STRING ] [ :args ARGUMENT-LIST ] )

For use within user-defined NST criteria: add a failure to a result.

(add-failure RESULT-REPORT [ :format FORMAT-STRING ] [ :args ARGUMENT-LIST ])

For use within user-defined NST criteria: add an info note to a result.

(add-info RESULT-REPORT INFO-ITEM)

For use within user-defined NST criteria: add a warning to a result.

The add-warning function adds an warning to a result record. The item can be any of a Lisp warning, an NST check-note or a format string; in the first two cases, no additional arguments should be provided.

(add-warning RESULT-REPORT [ :format ITEM ] [ :args ARGUMENT-LIST ] )

The helper function wrap-thrown-lisp-warning creates an NST check-note object from a standard Lisp warning.

Processing subcriteria

The criterion itself can contain subcriteria which can be incorporated into the main criterion's assessment. NST provides two functions which trigger testing by a subcriterion, each returning the check's result report.

The check-criterion-on-value function can be called from within a criterion body to verify that a value adheres to a criterion.

(check-criterion-on-value CRITERION VALUE)

This function verifies that the values return by evaluating the form adheres to the criterion.

(check-criterion-on-form CRITERION FORM)

General criteria definitions

The def-criterion macro defines a new criterion for use in NST tests. These criteria definitions are like generic function method definitions with two sets of formal parameters: the forms provided as the actual parameters of the criterion itself, and the values arising from the evaluation of the forms under test.

(def-criterion (name criterion-lambda-list values-lambda-list)
  [ doc-string )
  form
  form
  ...
  form)

• name: Name of the criterion.
• criterion-lambda-list: Lambda list for the arguments to the criterion. Optionally, the first element of the list is a symbol specifying the parameter-passing semantics for the criterion arguments: :values for call-by-value, or :forms for call-by-name (the default). The list may include the keywords =&key, &optional, &body and &rest but may not use &whole or &environment. Apart from this restriction, in the former case the list may be any ordinary lambda list as for defun, and in the latter case the list may be any macro lambda list as for defmacro.
• values-lambda-list: Lambda list for the forms under test. Optionally, the first element of the list is a symbol specifying the parameter-passing semantics for the criterion arguments: :values for call-by-value (the default), or :form for call-by-name. In the former case, the list may include the keywords &key, &optional, &body and &rest, but not &whole or &environment; apart from that restriction, list may be any ordinary lambda list as for defun. In the latter case, the remainder of the list must contain exactly one symbol, to which a form which would evaluate to the values under test will be bound. If the criterion ignores the values, then instead of a lambda list, this argument may be the symbol :ignore. On many platforms, listing a dummy parameter which is then declare=d =ignore or ignorable will produce a style warning: the body of a def-criterion should not be assumed to correspond directly to the body of a defmethod; in general there will be surrounding =destructuring-bind=s.
• documentation: An optional documentation string for the criterion.
• form: The body of the criterion definition should return a test result report contructed with the make-success-report, etc. functions.

Examples:

(def-criterion (:true () (bool))
  (if bool
      (make-success-report)
      (make-failure-report :format "Expected non-null, got: ~s"
                    :args (list bool))))

(def-criterion (:eql (target) (actual))
  (if (eql (eval target) actual)
      (make-success-report)
      (make-failure-report :format "Not eql to value of ~s"
                    :args (list target))))

The runtime system

The runtime system provides several operations for scheduling and running tests, and debugging failing and erring tests.

User-level NST operations are accessible from the REPL via the nst-cmd macro.

(nst-cmd NST-COMMAND ARG ... ARG)

Where a particular system supports the facility (Currently Allegro, and SBCL under ACL-REPL) the top-level alias :nst provides a shorthand to this function. Calling nst or nst-cmd without a command argument repeats the last test-executing command. For the sake of brevity we use the :nst shorthand below.

Command help

The :help command gives a complete inventory of runtime system commands.

Usage:

:nst :help

There are a number of commands for running tests, but most of the time only one will be needed:

Command run

The :run command executes all tests in the named package, or in the named group, or runs the named test. It is not necessary to prefix the name with a package prefix. The name does not need to be prefix-qualified, but if the name is ambiguous then :run will simply report the possible interpretations.

Usage:

:nst :run NAME

Command run-package

The :run-package command executes all tests associated with groups in the named packages, and reports the test results afterwards.

Usage:

:nst :run-package (:SEQ PACKAGE)

Command run-group

The :run-group command executes all tests associated with the name groups, and reports the test results afterwards. The group name should be package-qualified.

Usage:

:nst :run-group (:SEQ GROUP)

Command run-test

The :run-test command executes the given test. Both the group and test name should be package-qualified.

Usage:

:nst :run-test GROUP TEST

One further command for running a test is useful when writing and debugging the tests themselves:

Command apply

The :apply command assesses whether a test criterion prints the uses to which a particular name has been applied in an NST session.

Usage:

:nst :apply NAME

• The apply criterion first evaluates the forms under test, applying FUNCTION to them. The overall criterion passes or fails exactly when the subordinate CRITERION with the application's multiple result values.

  (:apply FUNCTION CRITERION)

  Example:

  (def-test applycheck (:apply cadr (:eql 10)) '(0 10 20))

There are two commands for (re)printing the results of tests:

Command report

The :report command summarizes successes, failures and errors in tests. It reports either for the named artifact, or for all recently-run tests.

Usage:

:nst :run
:nst :run PACKAGE
:nst :run GROUP
:nst :run GROUP TEST

Command detail

The :detail command gives detailed information about individual test results.

Usage:

:nst :detail
:nst :detail PACKAGE
:nst :detail GROUP
:nst :detail GROUP TEST

The set and unset display and adjust NST's configuration.

Command set

The :set command assigns or displays the values of NST runtime switches.

Usage:

:nst :set PROPERTY
:nst :set PROPERTY VALUE

There are currently three properties which can be manipulated by set and unset:

The :verbosity switch controls the level of NST's output.

Usage:

:nst :verbose SETTING

Valid settings are:

• :silent (aka nil)
• :quiet (aka :default)
• :verbose (aka t)
• :vverbose
• :trace

The :report and :detail commands operate by setting minimum levels of verbosity.

The :debug-on-error switch controls NST's behavior on errors. When non-nil, NST will break into the debugger when it encounters an error.

Usage:

:nst :debug-on-error FLAG

The :debug command is a short-cut for setting this property.

The :debug-on-fail switch controls NST's behavior when a test fails When non-nil, NST will break into the debugger when it encounters a failing test.

Usage:

:nst :debug-on-fail FLAG

This behavior is less useful than it may seem; by the time the results of the test are examined for failure, the stack from the actual form evaluation will usually have been released. Still, this switch is useful for inspecting the environment in which a failing test was run.

Note that both :debug-on-error and :debug-on-fail apply in the case of an error; if the latter is set but the former is not, then the debugger will be entered after an erring test completes.

The :debug command is a short-cut for setting this property.

Command unset

The :unset command clears the values of NST runtime switches.

Usage:

:nst :unset PROPERTY

The :backtraces switch, when non-=nil=, directs NST to attempt to capture the Lisp backtrace of errors in tests.

Usage:

:nst :backtraces FLAG

This property is only available on platform which allow programmatic examination of backtraces, which is not standardized in Common Lisp; currently we have implemented this feature on Allegro only.

This property has a complicated default setting. Firstly, if the symbol ='common-lisp-user::*nst-generate-backtraces*= is bound when NST loads, NST will use its value as the initial value for this property. Otherwise by default, on MacOS systems the property initializes to nil because of a known error on that system, but this setting can be overriden by the property :nst-unsafe-allegro-backtraces. Finally, if none of these issues apply, the initial value is t.

This property is only available on platform which allow programmatic examination of backtraces, which is not standardized in Common Lisp; currently we have implemented this feature on Allegro only.This property has a complicated default setting. Firstly, if the symbol ='common-lisp-user::*nst-generate-backtraces*= is bound when NST loads, NST will use its value as the initial value for this property. Otherwise by default, on MacOS systems the property initializes to nil because of a known error on that system, but this setting can be overriden by the property :nst-unsafe-allegro-backtraces. Finally, if none of these issues apply, the initial value is ==.

The above NST commands are governed by a number of global variables. In general, interactive use of NST should not require direct access to these variables, but when automating NST operations may require changing, or creating a new dynamic scope for, their settings.

User variable *debug-on-error*: if non-nil, will break into the Lisp REPL debugger upon encountering an unexpected error. If nil, will record the error and continue with other tests.

User variable *debug-on-fail*: if non-nil, will break into the Lisp REPL debugger upon encountering a test which fails. If nil, will record the failure and continue with other tests. This variable is useful inspecting the dynamic environment under which a test was evaluated.

User variable *default-report-verbosity* determines the default value for *nst-verbosity* when printing reports (2 by default).

Determines the output stream to which NST should print its output (*standard-output* by default).

Command open

The :open command injects the binding given by the fixture set into the current package.

Usage:

:nst :open (:SEQ FIXTURE)

Example:

CL-USER(75): (nst:def-fixtures small-fixture ()
                (fix-var1 3)
                (fix-var2 'asdfg))
NIL
CL-USER(76): (boundp 'fix-var1)
NIL
CL-USER(77): :nst :open small-fixture
Opened fixture SMALL-FIXTURE.
CL-USER(78): fix-var1
3
CL-USER(79):

Fixtures can be opened into a different package than where they were first defined, but these bindings are in addition to the bindings in the original package, and are made by a symbol import to the additional package.

Integration with ASDF

NST's integration with ASDF is a work in progress. This section described the current integration, the ways we expect it to change, and a less-flexible and lower-level, but likely more stable, alternative integration technique.

NST's ASDF systems

From version 1.2.2, the system :asdf-nst provides two classes for ASDF system definitions, asdf:nst-test-runner and asdf:nst-test-holder.

Up to NST 1.2.1 :asdf-nst provided