arbitrary

The arbitrary library defines an arbitrary category providing predicates for generating random values for selected types to the type object, complementing its type checking predicates. Both the object and the category predicates can be extended by the user with definitions for new types by defining clauses for multifile predicates. This library is notably used in the QuickCheck implementation by the lgtunit tool. See also the documentation of the mutations library for related functionality.

API documentation

Open the ../../docs/library_index.html#arbitrary link in a web browser.

Loading

To load all entities in this library, load the loader.lgt file:

| ?- logtalk_load(arbitrary(loader)).

Testing

To test this library predicates, load the tester.lgt file:

| ?- logtalk_load(arbitrary(tester)).

Several of provided tests are generic and verify correct behavior of all pre-defined and loaded user-defined generators and shrinkers for all ground types.

Pre-defined types

This library defines random generators for most common Logtalk and Prolog types. See the API documentation for a listing of all the pre-defined types.

Usage

The arbitrary category complements the type object and thus its predicates are accessed via this object. For example:

| ?- type::arbitrary(integer, Arbitrary).
Arbitrary = -816
yes

Defining new generators and shrinkers

To define a generator of arbitrary values for a type, define a clause for the arbitrary::arbitrary/1 multifile predicate specifying the type and a clause for the arbitrary::arbitrary/2 multifile predicate generating an arbitrary term of the specified type. As a simple example, assume that we want to define an “odd integer type”. We start by defining both the type checker and the arbitrary generator:

:- multifile(type::type/1).
type::type(odd).

:- multifile(type::check/2).
type::check(odd, Term) :-
    (   var(Term) ->
        throw(instantiation_error)
    ;   integer(Term),
        Term mod 2 =:= 1 ->
        true
    ;   throw(type_error(odd, Term))
    ).

:- multifile(arbitrary::arbitrary/1).
arbitrary::arbitrary(odd).

:- multifile(arbitrary::arbitrary/2).
arbitrary::arbitrary(odd, Arbitrary) :-
    type::arbitrary(integer, Arbitrary0),
    (   Arbitrary0 mod 2 =:= 1 ->
        Arbitrary = Arbitrary0
    ;   Arbitrary is Arbitrary0 + 1
    ).

The arbitrary library also provides meta-types that can simplify the definition of new generators. For example, the odd type above can also be defined using the constrain/2 meta-type to only generate values that satisfy a closure:

arbitrary::arbitrary(odd, Arbitrary) :-
    arbitrary(constrain(integer, [Arbitrary]>>(Arbitrary mod 2 =:= 1)), Arbitrary).

Another example is using the transform/2 meta-type to transform generated values for a base type using a closure. Assuming that we want to generate sorted list of random integers, we can write:

arbitrary::arbitrary(sorted_integer_list, Arbitrary) :-
    arbitrary(transform(list(integer), sort), Arbitrary).

We can also define a clause for the arbitrary::shrinker/1 multifile predicate to declare a new shrinker and a arbitrary::shrink/3 multifile predicate for shrinking arbitrary values for QuickCheck usage:

:- multifile(arbitrary::shrinker/1).
arbitrary::shrinker(odd).

:- multifile(arbitrary::shrink/3).
arbitrary::shrink(odd, Large, Small) :-
    integer(Large),
    (   Large < -1 ->
        Small is Large + 2
    ;   Large > 1,
        Small is Large - 2
    ).

Definitions for the shrink/3 predicate should either succeed or fail but never throw an exception. The shrink_sequence/3 predicate can be used to help testing that shrinking a value results in a finite sequence of values.

It is also possible to define edge cases for a given type for use with QuickCheck implementations. For example:

:- multifile(arbitrary::edge_case/2).
arbitrary::edge_case(odd,  1).
arbitrary::edge_case(odd, -1).

Edge cases are tried before resorting to generating arbitrary values for a type.

A more complex example is generating arbitrary values for a recursive type. A simple example of a recursive type is a binary tree. Assuming that we are working with a binary tree holding integers where each node is represented by a node(Left, Right) compound term, we can define a node(Depth) type where Depth is the maximum depth of the tree. This argument allows us to prevent excessively deep trees:

:- category(binary_tree).

    :- multifile(type::type/1).
    type::type(node(_)).

    :- multifile(type::check/2).
    type::check(node(_), Term) :-
        (   check(Term) ->
            true
        ;   var(Term) ->
            throw(instantiation_error)
        ;   throw(type_error(node(_), Term))
        ).

    check(Term) :-
        (   integer(Term) ->
            true
        ;   compound(Term),
            Term = node(Left, Right),
            check(Left),
            check(Right)
        ).

    :- multifile(arbitrary::arbitrary/1).
    arbitrary::arbitrary(node(_)).

    :- multifile(arbitrary::arbitrary/2).
    arbitrary::arbitrary(node(Depth), Arbitrary) :-
    (   Depth > 1 ->
        NewDepth is Depth - 1,
        type::arbitrary(
            types_frequency([
                integer - 1,
                compound(
                    node,
                    [
                        types([node(NewDepth), integer]),
                        types([node(NewDepth), integer])
                    ]
                ) - 3
            ]),
            Arbitrary
        )
    ;   type::arbitrary(
    integer, Arbitrary)
    ).

:- end_category.

In this second example, we use some of the pre-defined types provided by the library. The types_frequency(Pairs) type supports generating random terms for a type in the Type-Frequency pairs list where the type ie randomly chosen after the types relative frequency. The compound(Name, Types) type supports generating compound term with a given name and random arguments after the given types:

| ?- type::arbitrary(node(4), Arbitrary).
Arbitrary = 907
yes

| ?- type::arbitrary(node(4), Arbitrary).
Arbitrary = node(node(node(522, 509), node(83, 453)), node(454, -197))
yes

| ?- type::arbitrary(node(4), Arbitrary).
Arbitrary = node(node(-875, -866), -254)
yes

| ?- type::arbitrary(node(4), Arbitrary).
Arbitrary = node(-133, -831)
yes

The source code of these examples can be found in the test_files directory. Other examples of arbitrary term generators can be found in the implementation of the optionals and expecteds libraries.

Scoped generators and shrinkers

Declaring a new generator and possibly a shrinker for a custom type rises the possibility of a conflict with third-party defined generators and shrinkers. An alternative is to use the (::)/2 meta-type to define scoped generators and shrinkers. For example:

:- object(scoped).

    % the same predicate is used for both generating and validating
    :- public(custom/1).
    custom(Term) :-
        (   var(Term) ->
            % assume predicate used as a generator
            random::random(Term)
        ;   % assume predicate used as a validator
            float(Term)
        ).

    % a predicate with the same name is used for shrinking
    :- public(custom/2).
    custom(Larger, Small) :-
        Small is Larger / 2.

:- end_object.

Some sample calls:

| ?- type::arbitrary(scoped::custom, Arbitrary).
Arbitrary = 0.5788130906607927
yes

| ?- type::valid(scoped::custom, foo).
no

| ?- type::check(scoped::custom, _).
ERROR: type_error(instantiation_error)

| ?- type::check(scoped::custom, foo).
ERROR: type_error(scoped::custom, foo)

| ?- type::shrink(scoped::custom, 0.42, Smaller).
Smaller = 0.21
yes

The source code of this example can be found in the test_files directory.

Reproducing sequences of arbitrary terms

The arbitrary category provides access to the pseudo-random generator it uses via the get_seed/1 and set_seed/1. This allows sequences of arbitrary values to be reproduced. For example:

| ?- type::get_seed(Seed).
Seed = seed(3172, 9814, 20125)
yes

| ?- type::arbitrary(integer, Arbitrary).
Arbitrary = -816
yes

| ?- type::arbitrary(integer, Arbitrary).
Arbitrary = -113
yes

| ?- type::arbitrary(integer, Arbitrary).
Arbitrary = 446

| ?- type::set_seed(seed(3172, 9814, 20125)).
yes

| ?- type::arbitrary(integer, Arbitrary).
Arbitrary = -816
yes

| ?- type::arbitrary(integer, Arbitrary).
Arbitrary = -113
yes

| ?- type::arbitrary(integer, Arbitrary).
Arbitrary = 446
yes

The seed should be regarded as an opaque term and handled using the get_seed/1 and set_seed/1 predicates. These predicates are notably used in the QuickCheck implementation provided by the lgtunit tool.

Default size of generated terms

The library uses the value 42 for the default size of generated terms for types where size is meaningful and implicit. To override this default value, define a clause for the arbitrary::max_size/1 multifile predicate. The new default size must be a positive integer. For example:

:- multifile(arbitrary::max_size/1).
arbitrary::max_size(7).

When multiple definitions exist, the first valid one found is used. When no definition is valid, the default value of 42 is used.

Known issues

Some Prolog systems either don’t support the null character or provide buggy results when calling char_code/2 with a code of zero. When that’s the case, the null character is excluded when generating arbitrary characters or character codes.

Generating arbitrary Unicode characters (instead of Unicode codepoints) is inherently problematic as the process first generates codepoints and then tries to use the standard char_code/2 to convert them to characters. But, depending on the backend Prolog system and its internal (if any) Unicode normalization, it may not be possible to convert a codepoint to a single character.