spec framework

pip install kallikrein==0.22.0a17



kallikrein is a test framework inspired by specs2, focused on functional purity and composability.


pip install kallikrein


Spec classes need only define a docstring containing interpolation expressions that specify which of the class' methods are specs:

from kallikrein import k
from kallikrein.matchers import contain, greater_equal, forall
from amino import List

class ASpec:
    '''example specifications

    this test is simple
    simple spec $simple

    these tests are nested
    successful nesting $nested
    failed spec $failure

    def setup(self):
        self.a = 3

    def simple(self):
        return k(3).must(greater_equal(self.a))

    def nested(self):
        return k(List(1, 2, 3)).must(contain(greater_equal(self.a)))

    def failure(self):
        return k(List('abc', 'abc', 'ac')).must(forall(contain('b')))

There is no magic involved, the names in the docstring are simply used with getattr.

The expectations aren't evaluated in-place, but after having been returned to the spec runner. There is an alternative, impure version called unsafe_k that raises an exception.

Logical operations on expectations are possible:

k(3).must(equal(3)) & k(4).must(greater_equal(2)) | k(List(2)).must(contain(2))

associativity via parentheses is not yet implemented.

THe same applies for matchers:

k(3).must(equal(3) & greater_equal(2))

If a spec class has setup and teardown methods, they are called once before and after each individual spec.

The decorator kallikrein.pending can be used to mark a spec as pending:

def not_implemented_yet(self):

To deactivate a spec completely, simply put a comment character # in front of the line in the docstring.


% klk mod.path.to.ASpec

The output looks like this:


Selection of specs works as well by specifying a file name. Optionally, a line number or method name can be appended to run a single case:

% klk mod/path/to.py:18
% klk mod.path.to.ASpec.simple

Modules, packages, directories and files are valid as well, in which case a recursive search is done returning all valid specs specified in the modules' __all__ attribute:

% klk mod.path
% klk mod/path/to



This matcher simply checks that the expectable is an instance of the class supplied as an argument:


have_type can also take a matcher as a second argument, which uses the feature of chained matchers that works by implementing the typeclass ChainMatcher:

k(List(1, 2, 3)).must(have_type(List)(contain(greater(2))))

There are convenience extensions for Maybe and Either:



This matcher compares two strings or lists of strings (mixed works as well) and prints a diff of the lines instead of the whole lists:



must expects a Match instance for its argument which is produced by a Matcher when called with a target value and produces a MatchResult when evaluated. Subclassing Matcher and implementing the match and match_nested methods is a simple way to create a custom matcher, but there is a much more flexible concept available.

The TCMatcher class uses amino's typeclass system to allow special treatment of any type by any matcher without having to reimplement the matchers. A typeclass matcher consists of two string templates for assembling the success and failure messages and two instances of the typeclasses Predicate and Nesting.

Defining classes for those typeclasses for a specific type automatically registers them as handlers for that type in the desired matcher.

This is the implementation of contain for reference:

from collections.abc import Container, Iterable

class PredContain(Predicate):

class NestContain(Nesting):

is_container = L(issubclass)(_, Container)
is_collection = L(issubclass)(_, Iterable)

class PredContainCollection(PredContain, pred=is_container):

    def check(self, exp: Collection[A], target: A) -> Boolean:
        return Boolean(target in exp)

class NestContainCollection(NestContain, pred=is_collection):

    def match(self, exp: Collection[A], target: Match) -> List[MatchResult[B]]:
        return List.wrap([target.evaluate(e) for e in exp])

    def wrap(self, name: str, exp: Collection[A], nested: List[MatchResult[B]]) -> MatchResult[A]:
        return ExistsMatchResult(name, exp, nested)

success = '`{}` contains `{}`'
failure = '`{}` does not contain `{}`'
contain = matcher('contain', success, failure, PredContain, NestContain)

class PredContainMaybe(PredContain, tpe=Maybe):

    def check(self, exp: Maybe[A], target: A) -> Boolean:
        return Boolean(exp.contains(target))

The PredContain and NestContain classes are used to link instances for specific types to the contain matcher. The matcher checks all available instances for eligibility for the type of the checked expectable and calls the check, match and wrap methods on the respective instances.

In this example, the instances use a predicate function to check whether a type can be handled by them, in this case, if they are virtual subclasses of Container or Iterable. The simple way would be to pass tpe=list to the metaclass constructor instead of pred=is_container, but that would not allow any other iterable type to be matched. The instance PredContainMaybe demonstrates the use of the tpe variant and shows that additional instances for arbitrary types can be added without having to change the internal logic of kallikrein.

The internal part of TCMatcher constructs a SimpleMatchResult from the result of Predicate.check, indicating success of the match, and the two strings supplied to the constructor that describe the success and failure.

Because nested matches must be handled specifically to the matcher, the MatchResult must be constructed in the implementation. ExistsMatchResult is one possible variant; it receives the list of nested match results (one for each list element) and creates a detailed error message, succeeding if at least one nested MatchResult is successful.