pytest-bdd implements a subset of the Gherkin language to enable automating project requirements testing and to facilitate behavioral driven development.
Unlike many other BDD tools, it does not require a separate runner and benefits from the power and flexibility of pytest. It enables unifying unit and functional tests, reduces the burden of continuous integration server configuration and allows the reuse of test setups.
Pytest fixtures written for unit tests can be reused for setup and actions mentioned in feature steps with dependency injection. This allows a true BDD just-enough specification of the requirements without maintaining any context object containing the side effects of Gherkin imperative declarations.
pip install pytest-bdd
The minimum required version of pytest is 4.3.
An example test for a blog hosting software could look like this. Note that pytest-splinter is used to get the browser fixture.
publish_article.feature:
Feature: Blog
A site where you can publish your articles.
Scenario: Publishing the article
Given I'm an author user
And I have an article
When I go to the article page
And I press the publish button
Then I should not see the error message
And the article should be published # Note: will query the database
Note that only one feature is allowed per feature file.
test_publish_article.py:
from pytest_bdd import scenario, given, when, then
@scenario('publish_article.feature', 'Publishing the article')
def test_publish():
pass
@given("I'm an author user")
def author_user(auth, author):
auth['user'] = author.user
@given("I have an article", target_fixture="article")
def article(author):
return create_test_article(author=author)
@when("I go to the article page")
def go_to_article(article, browser):
browser.visit(urljoin(browser.url, '/manage/articles/{0}/'.format(article.id)))
@when("I press the publish button")
def publish_article(browser):
browser.find_by_css('button[name=publish]').first.click()
@then("I should not see the error message")
def no_error_message(browser):
with pytest.raises(ElementDoesNotExist):
browser.find_by_css('.message.error').first
@then("the article should be published")
def article_is_published(article):
article.refresh() # Refresh the object in the SQLAlchemy session
assert article.is_published
Functions decorated with the scenario decorator behave like a normal test function, and they will be executed after all scenario steps.
from pytest_bdd import scenario, given, when, then
@scenario('publish_article.feature', 'Publishing the article')
def test_publish(browser):
assert article.title in browser.html
Note
It is however encouraged to try as much as possible to have your logic only inside the Given, When, Then steps.
Sometimes, one has to declare the same fixtures or steps with different names for better readability. In order to use the same step function with multiple step names simply decorate it multiple times:
@given("I have an article")
@given("there's an article")
def article(author, target_fixture="article"):
return create_test_article(author=author)
Note that the given step aliases are independent and will be executed when mentioned.
For example if you associate your resource to some owner or not. Admin user can’t be an author of the article, but articles should have a default author.
Feature: Resource owner
Scenario: I'm the author
Given I'm an author
And I have an article
Scenario: I'm the admin
Given I'm the admin
And there's an article
Often it's possible to reuse steps giving them a parameter(s). This allows to have single implementation and multiple use, so less code. Also opens the possibility to use same step twice in single scenario and with different arguments! And even more, there are several types of step parameter parsers at your disposal (idea taken from behave implementation):
- string (the default)
- This is the default and can be considered as a null or exact parser. It parses no parameters and matches the step name by equality of strings.
- parse (based on: pypi_parse)
- Provides a simple parser that replaces regular expressions for
step parameters with a readable syntax like
{param:Type}
. The syntax is inspired by the Python builtinstring.format()
function. Step parameters must use the named fields syntax of pypi_parse in step definitions. The named fields are extracted, optionally type converted and then used as step function arguments. Supports type conversions by using type converters passed via extra_types - cfparse (extends: pypi_parse, based on: pypi_parse_type)
- Provides an extended parser with "Cardinality Field" (CF) support.
Automatically creates missing type converters for related cardinality
as long as a type converter for cardinality=1 is provided.
Supports parse expressions like:
*
{values:Type+}
(cardinality=1..N, many) *{values:Type*}
(cardinality=0..N, many0) *{value:Type?}
(cardinality=0..1, optional) Supports type conversions (as above). - re
- This uses full regular expressions to parse the clause text. You will
need to use named groups "(?P<name>...)" to define the variables pulled
from the text and passed to your
step()
function. Type conversion can only be done via converters step decorator argument (see example below).
The default parser is string, so just plain one-to-one match to the keyword definition. Parsers except string, as well as their optional arguments are specified like:
for cfparse parser
from pytest_bdd import parsers
@given(
parsers.cfparse("there are {start:Number} cucumbers",
extra_types=dict(Number=int)),
target_fixture="cucumbers",
)
def given_cucumbers(start):
return dict(start=start, eat=0)
for re parser
from pytest_bdd import parsers
@given(
parsers.re(r"there are (?P<start>\d+) cucumbers"),
converters=dict(start=int),
target_fixture="cucumbers",
)
def given_cucumbers(start):
return dict(start=start, eat=0)
Example:
Feature: Step arguments
Scenario: Arguments for given, when, then
Given there are 5 cucumbers
When I eat 3 cucumbers
And I eat 2 cucumbers
Then I should have 0 cucumbers
The code will look like:
import re
from pytest_bdd import scenario, given, when, then, parsers
@scenario("arguments.feature", "Arguments for given, when, then")
def test_arguments():
pass
@given(parsers.parse("there are {start:d} cucumbers"), target_fixture="cucumbers")
def given_cucumbers(start):
return dict(start=start, eat=0)
@when(parsers.parse("I eat {eat:d} cucumbers"))
def eat_cucumbers(cucumbers, eat):
start_cucumbers["eat"] += eat
@then(parsers.parse("I should have {left:d} cucumbers"))
def should_have_left_cucumbers(cucumbers, left):
assert cucumbers['start'] - cucumbers['eat'] == left
Example code also shows possibility to pass argument converters which may be useful if you need to postprocess step arguments after the parser.
You can implement your own step parser. It's interface is quite simple. The code can looks like:
import re
from pytest_bdd import given, parsers
class MyParser(parsers.StepParser):
"""Custom parser."""
def __init__(self, name, **kwargs):
"""Compile regex."""
super().__init__(name)
self.regex = re.compile(re.sub("%(.+)%", "(?P<\1>.+)", self.name), **kwargs)
def parse_arguments(self, name):
"""Get step arguments.
:return: `dict` of step arguments
"""
return self.regex.match(name).groupdict()
def is_matching(self, name):
"""Match given name with the step name."""
return bool(self.regex.match(name))
@given(parsers.parse("there are %start% cucumbers"), target_fixture="cucumbers")
def given_cucumbers(start):
return dict(start=start, eat=0)
Dependency injection is not a panacea if you have complex structure of your test setup data. Sometimes there's a need such a given step which would imperatively change the fixture only for certain test (scenario), while for other tests it will stay untouched. To allow this, special parameter target_fixture exists in the given decorator:
from pytest_bdd import given
@pytest.fixture
def foo():
return "foo"
@given("I have injecting given", target_fixture="foo")
def injecting_given():
return "injected foo"
@then('foo should be "injected foo"')
def foo_is_foo(foo):
assert foo == 'injected foo'
Feature: Target fixture
Scenario: Test given fixture injection
Given I have injecting given
Then foo should be "injected foo"
In this example existing fixture foo will be overridden by given step I have injecting given only for scenario it's used in.
Sometimes it is also useful to let when and then steps to provide a fixture as well. A common use case is when we have to assert the outcome of an HTTP request:
# test_blog.py
from pytest_bdd import scenarios, given, when, then
from my_app.models import Article
scenarios("blog.feature")
@given("there is an article", target_fixture="article")
def there_is_an_article():
return Article()
@when("I request the deletion of the article", target_fixture="request_result")
def there_should_be_a_new_article(article, http_client):
return http_client.delete(f"/articles/{article.uid}")
@then("the request should be successful")
def article_is_published(request_result):
assert request_result.status_code == 200
# blog.feature
Feature: Blog
Scenario: Deleting the article
Given there is an article
When I request the deletion of the article
Then the request should be successful
As Gherkin, pytest-bdd supports multiline steps (aka PyStrings). But in much cleaner and powerful way:
Feature: Multiline steps
Scenario: Multiline step using sub indentation
Given I have a step with:
Some
Extra
Lines
Then the text should be parsed with correct indentation
Step is considered as multiline one, if the next line(s) after it's first line, is indented relatively to the first line. The step name is then simply extended by adding further lines with newlines. In the example above, the Given step name will be:
'I have a step with:\nSome\nExtra\nLines'
You can of course register step using full name (including the newlines), but it seems more practical to use step arguments and capture lines after first line (or some subset of them) into the argument:
import re
from pytest_bdd import given, then, scenario, parsers
@scenario(
'multiline.feature',
'Multiline step using sub indentation',
)
def test_multiline():
pass
@given(parsers.parse("I have a step with:\n{text}"), target_fixture="i_have_text")
def i_have_text(text):
return text
@then("the text should be parsed with correct indentation")
def text_should_be_correct(i_have_text, text):
assert i_have_text == text == 'Some\nExtra\nLines'
Note that then step definition (text_should_be_correct) in this example uses text fixture which is provided by a given step (i_have_text) argument with the same name (text). This possibility is described in the `Step arguments are fixtures as well!`_ section.
If you have relatively large set of feature files, it's boring to manually bind scenarios to the tests using the scenario decorator. Of course with the manual approach you get all the power to be able to additionally parametrize the test, give the test function a nice name, document it, etc, but in the majority of the cases you don't need that. Instead you want to bind all scenarios found in the feature folder(s) recursively automatically. For this - there's a scenarios helper.
from pytest_bdd import scenarios
# assume 'features' subfolder is in this file's directory
scenarios('features')
That's all you need to do to bind all scenarios found in the features folder! Note that you can pass multiple paths, and those paths can be either feature files or feature folders.
from pytest_bdd import scenarios
# pass multiple paths/files
scenarios('features', 'other_features/some.feature', 'some_other_features')
But what if you need to manually bind certain scenario, leaving others to be automatically bound? Just write your scenario in a normal way, but ensure you do it BEFORE the call of scenarios helper.
from pytest_bdd import scenario, scenarios
@scenario('features/some.feature', 'Test something')
def test_something():
pass
# assume 'features' subfolder is in this file's directory
scenarios('features')
In the example above test_something scenario binding will be kept manual, other scenarios found in the features folder will be bound automatically.
Scenarios can be parametrized to cover few cases. In Gherkin the variable
templates are written using corner braces as <somevalue>
.
Gherkin scenario outlines are supported by pytest-bdd
exactly as it's described in the behave docs.
Example:
Feature: Scenario outlines
Scenario Outline: Outlined given, when, then
Given there are <start> cucumbers
When I eat <eat> cucumbers
Then I should have <left> cucumbers
Examples:
| start | eat | left |
| 12 | 5 | 7 |
The more features and scenarios you have, the more important becomes the question about their organization. The things you can do (and that is also a recommended way):
- organize your feature files in the folders by semantic groups:
features │ ├──frontend │ │ │ └──auth │ │ │ └──login.feature └──backend │ └──auth │ └──login.feature
This looks fine, but how do you run tests only for certain feature? As pytest-bdd uses pytest, and bdd scenarios are actually normal tests. But test files are separate from the feature files, the mapping is up to developers, so the test files structure can look completely different:
tests │ └──functional │ └──test_auth.py │ └ """Authentication tests.""" from pytest_bdd import scenario @scenario('frontend/auth/login.feature') def test_logging_in_frontend(): pass @scenario('backend/auth/login.feature') def test_logging_in_backend(): pass
For picking up tests to run we can use tests selection technique. The problem is that you have to know how your tests are organized, knowing only the feature files organization is not enough. cucumber tags introduce standard way of categorizing your features and scenarios, which pytest-bdd supports. For example, we could have:
@login @backend
Feature: Login
@successful
Scenario: Successful login
pytest-bdd uses pytest markers as a storage of the tags for the given scenario test, so we can use standard test selection:
pytest -m "backend and login and successful"
The feature and scenario markers are not different from standard pytest markers, and the @
symbol is stripped out
automatically to allow test selector expressions. If you want to have bdd-related tags to be distinguishable from the
other test markers, use prefix like bdd.
Note that if you use pytest --strict option, all bdd tags mentioned in the feature files should be also in the
markers setting of the pytest.ini config. Also for tags please use names which are python-compatible variable
names, eg starts with a non-number, underscore alphanumeric, etc. That way you can safely use tags for tests filtering.
You can customize how tags are converted to pytest marks by implementing the
pytest_bdd_apply_tag
hook and returning True
from it:
def pytest_bdd_apply_tag(tag, function):
if tag == 'todo':
marker = pytest.mark.skip(reason="Not implemented yet")
marker(function)
return True
else:
# Fall back to the default behavior of pytest-bdd
return None
Test setup is implemented within the Given section. Even though these steps are executed imperatively to apply possible side-effects, pytest-bdd is trying to benefit of the PyTest fixtures which is based on the dependency injection and makes the setup more declarative style.
@given("I have a beautiful article", target_fixture="article")
def article():
return Article(is_beautiful=True)
The target PyTest fixture "article" gets the return value and any other step can depend on it.
Feature: The power of PyTest
Scenario: Symbolic name across steps
Given I have a beautiful article
When I publish this article
When step is referring the article to publish it.
@when("I publish this article")
def publish_article(article):
article.publish()
Many other BDD toolkits operate a global context and put the side effects there. This makes it very difficult to implement the steps, because the dependencies appear only as the side-effects in the run-time and not declared in the code. The publish article step has to trust that the article is already in the context, has to know the name of the attribute it is stored there, the type etc.
In pytest-bdd you just declare an argument of the step function that it depends on and the PyTest will make sure to provide it.
Still side effects can be applied in the imperative style by design of the BDD.
Feature: News website
Scenario: Publishing an article
Given I have a beautiful article
And my article is published
Functional tests can reuse your fixture libraries created for the unit-tests and upgrade them by applying the side effects.
@pytest.fixture
def article():
return Article(is_beautiful=True)
@given("I have a beautiful article")
def i_have_a_beautiful_article(article):
pass
@given("my article is published")
def published_article(article):
article.publish()
return article
This way side-effects were applied to our article and PyTest makes sure that all steps that require the "article" fixture will receive the same object. The value of the "published_article" and the "article" fixtures is the same object.
Fixtures are evaluated only once within the PyTest scope and their values are cached.
It's often the case that to cover certain feature, you'll need multiple scenarios. And it's logical that the setup for those scenarios will have some common parts (if not equal). For this, there are backgrounds. pytest-bdd implements Gherkin backgrounds for features.
Feature: Multiple site support
Background:
Given a global administrator named "Greg"
And a blog named "Greg's anti-tax rants"
And a customer named "Wilson"
And a blog named "Expensive Therapy" owned by "Wilson"
Scenario: Wilson posts to his own blog
Given I am logged in as Wilson
When I try to post to "Expensive Therapy"
Then I should see "Your article was published."
Scenario: Greg posts to a client's blog
Given I am logged in as Greg
When I try to post to "Expensive Therapy"
Then I should see "Your article was published."
In this example, all steps from the background will be executed before all the scenario's own given steps, adding possibility to prepare some common setup for multiple scenarios in a single feature. About background best practices, please read here.
Note
There is only step "Given" should be used in "Background" section, steps "When" and "Then" are prohibited, because their purpose are related to actions and consuming outcomes, that is conflict with "Background" aim - prepare system for tests or "put the system in a known state" as "Given" does it. The statement above is applied for strict Gherkin mode, which is enabled by default.
Sometimes scenarios define new names for the existing fixture that can be inherited (reused). For example, if we have pytest fixture:
@pytest.fixture
def article():
"""Test article."""
return Article()
Then this fixture can be reused with other names using given():
@given('I have beautiful article')
def i_have_an_article(article):
"""I have an article."""
It is possible to define some common steps in the parent conftest.py and simply expect them in the child test file.
common_steps.feature:
Scenario: All steps are declared in the conftest
Given I have a bar
Then bar should have value "bar"
conftest.py:
from pytest_bdd import given, then
@given("I have a bar", target_fixture="bar")
def bar():
return "bar"
@then('bar should have value "bar"')
def bar_is_bar(bar):
assert bar == "bar"
test_common.py:
@scenario("common_steps.feature", "All steps are declared in the conftest")
def test_conftest():
pass
There are no definitions of the steps in the test file. They were collected from the parent conftest.py.
Here is the list of steps that are implemented inside of the pytest-bdd:
- given
- trace - enters the pdb debugger via pytest.set_trace()
- when
- trace - enters the pdb debugger via pytest.set_trace()
- then
- trace - enters the pdb debugger via pytest.set_trace()
By default, pytest-bdd will use current module's path as base path for finding feature files, but this behaviour can be changed in the pytest configuration file (i.e. pytest.ini, tox.ini or setup.cfg) by declaring the new base path in the bdd_features_base_dir key. The path is interpreted as relative to the working directory when starting pytest. You can also override features base path on a per-scenario basis, in order to override the path for specific tests.
pytest.ini:
[pytest]
bdd_features_base_dir = features/
tests/test_publish_article.py:
from pytest_bdd import scenario
@scenario("foo.feature", "Foo feature in features/foo.feature")
def test_foo():
pass
@scenario(
"foo.feature",
"Foo feature in tests/local-features/foo.feature",
features_base_dir="./local-features/",
)
def test_foo_local():
pass
The features_base_dir parameter can also be passed to the @scenario decorator.
If you want to avoid retyping the feature file name when defining your scenarios in a test file, use functools.partial
.
This will make your life much easier when defining multiple scenarios in a test file. For example:
test_publish_article.py:
from functools import partial
import pytest_bdd
scenario = partial(pytest_bdd.scenario, "/path/to/publish_article.feature")
@scenario("Publishing the article")
def test_publish():
pass
@scenario("Publishing the article as unprivileged user")
def test_publish_unprivileged():
pass
You can learn more about functools.partial in the Python docs.
pytest-bdd exposes several pytest hooks which might be helpful building useful reporting, visualization, etc on top of it:
- pytest_bdd_before_scenario(request, feature, scenario) - Called before scenario is executed
- pytest_bdd_after_scenario(request, feature, scenario) - Called after scenario is executed (even if one of steps has failed)
- pytest_bdd_before_step(request, feature, scenario, step, step_func) - Called before step function is executed and it's arguments evaluated
- pytest_bdd_before_step_call(request, feature, scenario, step, step_func, step_func_args) - Called before step function is executed with evaluated arguments
- pytest_bdd_after_step(request, feature, scenario, step, step_func, step_func_args) - Called after step function is successfully executed
- pytest_bdd_step_error(request, feature, scenario, step, step_func, step_func_args, exception) - Called when step function failed to execute
- pytest_bdd_step_func_lookup_error(request, feature, scenario, step, exception) - Called when step lookup failed
Tools recommended to use for browser testing:
- pytest-splinter - pytest splinter integration for the real browser testing
It's important to have nice reporting out of your bdd tests. Cucumber introduced some kind of standard for json format which can be used for, for example, by this Jenkins plugin.
To have an output in json format:
pytest --cucumberjson=<path to json report>
This will output an expanded (meaning scenario outlines will be expanded to several scenarios) cucumber format.
To enable gherkin-formatted output on terminal, use
pytest --gherkin-terminal-reporter
For newcomers it's sometimes hard to write all needed test code without being frustrated. To simplify their life, simple code generator was implemented. It allows to create fully functional but of course empty tests and step definitions for given a feature file. It's done as a separate console script provided by pytest-bdd package:
pytest-bdd generate <feature file name> .. <feature file nameN>
It will print the generated code to the standard output so you can easily redirect it to the file:
pytest-bdd generate features/some.feature > tests/functional/test_some.py
For more experienced users, there's smart code generation/suggestion feature. It will only generate the test code which is not yet there, checking existing tests and step definitions the same way it's done during the test execution. The code suggestion tool is called via passing additional pytest arguments:
pytest --generate-missing --feature features tests/functional
The output will be like:
============================= test session starts ============================== platform linux2 -- Python 2.7.6 -- py-1.4.24 -- pytest-2.6.2 plugins: xdist, pep8, cov, cache, bdd, bdd, bdd collected 2 items Scenario is not bound to any test: "Code is generated for scenarios which are not bound to any tests" in feature "Missing code generation" in /tmp/pytest-552/testdir/test_generate_missing0/tests/generation.feature -------------------------------------------------------------------------------- Step is not defined: "I have a custom bar" in scenario: "Code is generated for scenario steps which are not yet defined(implemented)" in feature "Missing code generation" in /tmp/pytest-552/testdir/test_generate_missing0/tests/generation.feature -------------------------------------------------------------------------------- Please place the code above to the test file(s): @scenario('tests/generation.feature', 'Code is generated for scenarios which are not bound to any tests') def test_Code_is_generated_for_scenarios_which_are_not_bound_to_any_tests(): """Code is generated for scenarios which are not bound to any tests.""" @given("I have a custom bar") def I_have_a_custom_bar(): """I have a custom bar."""
As as side effect, the tool will validate the files for format errors, also some of the logic bugs, for example the ordering of the types of the steps.
The primary focus of the pytest-bdd is the compatibility with the latest gherkin developments e.g. multiple scenario outline example tables with tags support etc.
In order to provide the best compatibility it is best to support the features described in the official gherkin reference. This means deprecation of some non-standard features that were implemented in pytest-bdd.
The example tables on the feature level are no longer supported. The tests should be parametrized with the example tables on the scenario level.
Vertical example tables are no longer supported since the official gherkin doesn't support them. The example tables should have horizontal orientation.
Step parsed arguments conflicted with the fixtures. Now they no longer define fixture. If the fixture has to be defined by the step the target_fixture param should be used.
Templated steps (e.g. @given("there are <start> cucumbers")
) should now the use step argument parsers in order to match the scenario outlines and get the values from the example tables. The values from the example tables are no longer passed as fixtures, although if you define your step to use a parser, the parameters will be still provided as fixtures.
# Old step definition:
@given("there are <start> cucumbers")
def given_cucumbers(start):
pass
# New step definition:
@given(parsers.parse("there are {start} cucumbers"))
def given_cucumbers(start):
pass
Scenario example_converters are removed in favor of the converters provided on the step level:
# Old code:
@given("there are <start> cucumbers")
def given_cucumbers(start):
return {"start": start}
@scenario("outline.feature", "Outlined", example_converters={"start": float})
def test_outline():
pass
# New code:
@given(parsers.parse("there are {start} cucumbers"), converters={"start": float})
def given_cucumbers(start):
return {"start": start}
@scenario("outline.feature", "Outlined")
def test_outline():
pass
The significant downside of combining scenario outline and pytest parametrization approach was inability to see the test table from the feature file.
Given steps are no longer fixtures. In case it is needed to make given step setup a fixture the target_fixture parameter should be used.
@given("there's an article", target_fixture="article")
def there_is_an_article():
return Article()
Given steps no longer have fixture parameter. In fact the step may depend on multiple fixtures. Just normal step declaration with the dependency injection should be used.
@given("there's an article")
def there_is_an_article(article):
pass
Strict gherkin option is removed, so the strict_gherkin
parameter can be removed from the scenario decorators
as well as bdd_strict_gherkin
from the ini files.
Step validation handlers for the hook pytest_bdd_step_validation_error
should be removed.
This software is licensed under the MIT license.
© 2013-2014 Oleg Pidsadnyi, Anatoly Bubenkov and others