Writing good code is hard, or at least it can be. A major challenge is to ensure that your code remains correct as you add new functionality.
Unit testing is a modern, light-weight testing methodology in widespread use in many programming languages. Development that relies on unit tests is often referred to as Test-Driven Development (TDD ). Most Twisted code is tested using TDD.
To gain a solid understanding of unit testing in Python, you should read the unittest -- Unit testing framework chapter of the Python Library Reference.
There is a lot of information available online and in books.
This document is principally a guide to Trial, Twisted’s unit testing framework. Trial is based on Python’s unit testing framework. While we do not aim to give a comprehensive guide to general Python unit testing, it will be helpful to consider a simple non-networked example before expanding to cover networking code that requires the special capabilities of Trial. If you are already familiar with unit test in Python, jump straight to the section specific to testing Twisted code.
Note
In what follows we will make a series of refinements to some simple classes.
In order to keep the examples and source code links complete and to allow you to run Trial on the intermediate results at every stage, I add _N (where the N are successive integers) to file names to keep them separate.
This is a minor visual distraction that should be ignored.
We’ll create a library for arithmetic calculation.
First, create a project structure with a directory called calculus containing an empty __init__.py file.
Then put the following simple class definition API into calculus/base_1.py :
# -*- test-case-name: calculus.test.test_base_1 -*-
class Calculation:
def add(self, a, b):
pass
def subtract(self, a, b):
pass
def multiply(self, a, b):
pass
def divide(self, a, b):
pass
(Ignore the test-case-name comment for now. You’ll see why that’s useful below .)
We’ve written the interface, but not the code. Now we’ll write a set of tests. At this point of development, we’ll be expecting all tests to fail. Don’t worry, that’s part of the point. Once we have a test framework functioning, and we have some decent tests written (and failing!), we’ll go and do the actual development of our calculation API. This is the preferred way to work for many people using TDD - write tests first, make sure they fail, then do development. Others are not so strict and write tests after doing the development.
Create a test directory beneath calculus , with an empty __init__.py file.
In a calculus/test/test_base_1.py, put the following:
from calculus.base_1 import Calculation
from twisted.trial import unittest
class CalculationTestCase(unittest.TestCase):
def test_add(self):
calc = Calculation()
result = calc.add(3, 8)
self.assertEqual(result, 11)
def test_subtract(self):
calc = Calculation()
result = calc.subtract(7, 3)
self.assertEqual(result, 4)
def test_multiply(self):
calc = Calculation()
result = calc.multiply(12, 5)
self.assertEqual(result, 60)
def test_divide(self):
calc = Calculation()
result = calc.divide(12, 5)
self.assertEqual(result, 2)
You should now have the following 4 files:
calculus/__init__.py
calculus/base_1.py
calculus/test/__init__.py
calculus/test/test_base_1.py
To run the tests, you must ensure you are able to load them.
Make sure you are in the directory that the calculus folder is in, if you run ls or dir you should see the folder.
You can test that you can import the calculus package by running python -c import calculus.
If it reports an error (“No module named calculus”), double check you are in the correct directory.
Run python -m twisted.trial calculus.test.test_base_1 from the command line when you are in the directory containing the calculus directory.
You should see the following output (though your files are probably not in /tmp ):
$ python -m twisted.trial calculus.test.test_base_1
calculus.test.test_base_1
CalculationTestCase
test_add ... [FAIL]
test_divide ... [FAIL]
test_multiply ... [FAIL]
test_subtract ... [FAIL]
===============================================================================
[FAIL]
Traceback (most recent call last):
File "/tmp/calculus/test/test_base_1.py", line 8, in test_add
self.assertEqual(result, 11)
twisted.trial.unittest.FailTest: not equal:
a = None
b = 11
calculus.test.test_base_1.CalculationTestCase.test_add
===============================================================================
[FAIL]
Traceback (most recent call last):
File "/tmp/calculus/test/test_base_1.py", line 23, in test_divide
self.assertEqual(result, 2)
twisted.trial.unittest.FailTest: not equal:
a = None
b = 2
calculus.test.test_base_1.CalculationTestCase.test_divide
===============================================================================
[FAIL]
Traceback (most recent call last):
File "/tmp/calculus/test/test_base_1.py", line 18, in test_multiply
self.assertEqual(result, 60)
twisted.trial.unittest.FailTest: not equal:
a = None
b = 60
calculus.test.test_base_1.CalculationTestCase.test_multiply
===============================================================================
[FAIL]
Traceback (most recent call last):
File "/tmp/calculus/test/test_base_1.py", line 13, in test_subtract
self.assertEqual(result, 4)
twisted.trial.unittest.FailTest: not equal:
a = None
b = 4
calculus.test.test_base_1.CalculationTestCase.test_subtract
-------------------------------------------------------------------------------
Ran 4 tests in 0.042s
FAILED (failures=4)
How to interpret this output? You get a list of the individual tests, each followed by its result.
By default, failures are printed at the end, but this can be changed with the -e (or --rterrors ) option.
One very useful thing in this output is the fully-qualified name of the failed tests.
This appears at the bottom of each =-delimited area of the output.
This allows you to copy and paste it to just run a single test you’re interested in.
In our example, you could run python -m twisted.trial calculus.test.test_base_1.CalculationTestCase.test_subtract from the shell.
Note that trial can use different reporters to modify its output. Run python -m twisted.trial --help-reporters to see a list of reporters.
The tests can be run by Trial in multiple ways:
python -m twisted.trial calculus: run all the tests for the calculus package.
python -m twisted.trial calculus.test: run using Python’s import notation.
python -m twisted.trial calculus.test.test_base_1: as above, for a specific test module.
You can follow that logic by putting your class name and even a method name to only run those specific tests.
python -m twisted.trial --testmodule=calculus/base_1.py: use the test-case-name comment in the first line of calculus/base_1.py to find the tests.
python -m twisted.trial calculus/test: run all the tests in the test directory (not recommended).
python -m twisted.trial calculus/test/test_base_1.py: run a specific test file (not recommended).
The first 3 versions using full qualified names are strongly encouraged: they are much more reliable and they allow you to easily be more selective in your test runs.
You’ll notice that Trial creates a _trial_temp directory in the directory where you run the tests. This has a file called test.log which contains the log output of the tests (created using log.msg or log.err functions).
Examine this file if you add logging to your tests.
Now that we have a working test framework in place, and our tests are failing (as expected) we can go and try to implement the correct API.
We’ll do that in a new version of the above base_1 module, calculus/base_2.py :
# -*- test-case-name: calculus.test.test_base_2 -*-
class Calculation:
def add(self, a, b):
return a + b
def subtract(self, a, b):
return a - b
def multiply(self, a, b):
return a * b
def divide(self, a, b):
return a // b
We’ll also create a new version of test_base_1 which imports and test this new implementation, in calculus/test_base_2.py:
from calculus.base_2 import Calculation
from twisted.trial import unittest
class CalculationTestCase(unittest.TestCase):
def test_add(self):
calc = Calculation()
result = calc.add(3, 8)
self.assertEqual(result, 11)
def test_subtract(self):
calc = Calculation()
result = calc.subtract(7, 3)
self.assertEqual(result, 4)
def test_multiply(self):
calc = Calculation()
result = calc.multiply(12, 5)
self.assertEqual(result, 60)
def test_divide(self):
calc = Calculation()
result = calc.divide(12, 5)
self.assertEqual(result, 2)
is a copy of test_base_1, but with the import changed. Run Trial again as above, and your tests should now pass:
$ python -m twisted.trial calculus.test.test_base_2
Running 4 tests.
calculus.test.test_base
CalculationTestCase
test_add ... [OK]
test_divide ... [OK]
test_multiply ... [OK]
test_subtract ... [OK]
-------------------------------------------------------------------------------
Ran 4 tests in 0.067s
PASSED (successes=4)
You’ll notice that our test file contains redundant code. Let’s get rid
of that. Python’s unit testing framework allows your test class to define a
setUp method that is called before
each test method in the class. This allows you to add attributes
to self that can be used in test
methods. We’ll also add a parameterized test method to further simplify the
code.
Note that a test class may also provide the counterpart of setUp , named tearDown ,
which will be called after each test (whether successful or
not). tearDown is mainly used for post-test
cleanup purposes. We will not use tearDown
until later.
Create calculus/test/test_base_2b.py as follows:
from calculus.base_2 import Calculation
from twisted.trial import unittest
class CalculationTestCase(unittest.TestCase):
def setUp(self):
self.calc = Calculation()
def _test(self, operation, a, b, expected):
result = operation(a, b)
self.assertEqual(result, expected)
def test_add(self):
self._test(self.calc.add, 3, 8, 11)
def test_subtract(self):
self._test(self.calc.subtract, 7, 3, 4)
def test_multiply(self):
self._test(self.calc.multiply, 6, 9, 54)
def test_divide(self):
self._test(self.calc.divide, 12, 5, 2)
Much cleaner, isn’t it?
We’ll now add some additional error tests. Testing just for successful
use of the API is generally not enough, especially if you expect your code
to be used by others. Let’s make sure the Calculation class raises exceptions if someone tries
to call its methods with arguments that cannot be converted to
integers.
We arrive at calculus/test/test_base_3.py:
from calculus.base_3 import Calculation
from twisted.trial import unittest
class CalculationTestCase(unittest.TestCase):
def setUp(self):
self.calc = Calculation()
def _test(self, operation, a, b, expected):
result = operation(a, b)
self.assertEqual(result, expected)
def _test_error(self, operation):
self.assertRaises(TypeError, operation, "foo", 2)
self.assertRaises(TypeError, operation, "bar", "egg")
self.assertRaises(TypeError, operation, [3], [8, 2])
self.assertRaises(TypeError, operation, {"e": 3}, {"r": "t"})
def test_add(self):
self._test(self.calc.add, 3, 8, 11)
def test_subtract(self):
self._test(self.calc.subtract, 7, 3, 4)
def test_multiply(self):
self._test(self.calc.multiply, 6, 9, 54)
def test_divide(self):
self._test(self.calc.divide, 12, 5, 2)
def test_errorAdd(self):
self._test_error(self.calc.add)
def test_errorSubtract(self):
self._test_error(self.calc.subtract)
def test_errorMultiply(self):
self._test_error(self.calc.multiply)
def test_errorDivide(self):
self._test_error(self.calc.divide)
We’ve added four new tests and one general-purpose function, _test_error . This function uses the assertRaises method, which takes an exception class,
a function to run and its arguments, and checks that calling the function
on the arguments does indeed raise the given exception.
If you run the above, you’ll see that not all tests fail. In Python it’s
often valid to add and multiply objects of different and even differing
types, so the code in the add and multiply tests does not raise an exception
and those tests therefore fail. So let’s add explicit type conversion to
our API class. This brings us to calculus/base_3.py :
# -*- test-case-name: calculus.test.test_base_3 -*-
class Calculation:
def _make_ints(self, *args):
try:
return [int(arg) for arg in args]
except ValueError:
raise TypeError("Couldn't coerce arguments to integers: {}".format(*args))
def add(self, a, b):
a, b = self._make_ints(a, b)
return a + b
def subtract(self, a, b):
a, b = self._make_ints(a, b)
return a - b
def multiply(self, a, b):
a, b = self._make_ints(a, b)
return a * b
def divide(self, a, b):
a, b = self._make_ints(a, b)
return a // b
Here the _make_ints helper function tries to convert a list into a list of equivalent integers, and raises a TypeError in case the conversion goes wrong.
Note
The int conversion can also raise a TypeError if passed something of
the wrong type, such as a list. We’ll just let that exception go by, as
TypeError is already what we want in case something goes wrong.
Up to this point we’ve been doing fairly standard Python unit testing.
With only a few cosmetic changes (most importantly, directly importing
unittest instead of using Twisted’s unittest version) we could make the
above tests run using Python’s standard library unit testing framework.
Here we will assume a basic familiarity with Twisted’s network I/O, timing, and Deferred APIs. If you haven’t already read them, you should read the documentation on Writing Servers , Writing Clients , and Deferreds .
Now we’ll get to the real point of this tutorial and take advantage of Trial to test Twisted code.
We’ll now create a custom protocol to invoke our class from a telnet-like session. We’ll remotely call commands with arguments and read back the response. The goal will be to test our network code without creating sockets.
First we’ll write the tests, and then explain what they do. The first version of the remote test code is:
from calculus.remote_1 import RemoteCalculationFactory
from twisted.trial import unittest
from twisted.test import proto_helpers
class RemoteCalculationTestCase(unittest.TestCase):
def setUp(self):
factory = RemoteCalculationFactory()
self.proto = factory.buildProtocol(("127.0.0.1", 0))
self.tr = proto_helpers.StringTransport()
self.proto.makeConnection(self.tr)
def _test(self, operation, a, b, expected):
self.proto.dataReceived("{} {} {}\r\n".format(operation, a, b).encode("utf-8"))
self.assertEqual(int(self.tr.value()), expected)
def test_add(self):
return self._test("add", 7, 6, 13)
def test_subtract(self):
return self._test("subtract", 82, 78, 4)
def test_multiply(self):
return self._test("multiply", 2, 8, 16)
def test_divide(self):
return self._test("divide", 14, 3, 4)
To fully understand this client, it helps a lot to be comfortable with the Factory/Protocol/Transport pattern used in Twisted.
We first create a protocol factory object. Note that we have yet to see
the RemoteCalculationFactory class. It is in
calculus/remote_1.py below. We
call buildProtocol to ask the factory to build us a
protocol object that knows how to talk to our server. We then make a fake
network transport, an instance of twisted.test.proto_helpers.StringTransport
class (note that test packages are generally not part of Twisted’s public API;``twisted.test.proto_helpers`` is an exception). This fake
transport is the key to the communications. It is used to emulate a network
connection without a network. The address and port passed to buildProtocol
are typically used by the factory to choose to immediately deny remote connections; since we’re using a fake transport, we can choose any value that will be acceptable to the factory. In this case the factory just ignores the address, so we don’t need to pick anything in particular.
Testing protocols without the use of real network connections is both simple and recommended when testing Twisted code. Even though there are many tests in Twisted that use the network, most good tests don’t. The problem with unit tests and networking is that networks aren’t reliable. We cannot know that they will exhibit reasonable behavior all the time. This creates intermittent test failures due to network vagaries. Right now we’re trying to test our Twisted code, not network reliability. By setting up and using a fake transport, we can write 100% reliable tests. We can also test network failures in a deterministic manner, another important part of your complete test suite.
The final key to understanding this client code is the _test method. The call to dataReceived simulates data arriving on the network
transport. But where does it arrive? It’s handed to the lineReceived method of the protocol instance (in
calculus/remote_1.py below). So the client
is essentially tricking the server into thinking it has received the
operation and the arguments over the network. The server (once again, see
below) hands over the work to its CalculationProxy object which in turn hands it to its
Calculation instance. The result is written
back via sendLine (into the fake string
transport object), and is then immediately available to the client, who
fetches it with tr.value() and checks that it
has the expected value. So there’s quite a lot going on behind the scenes
in the two-line _test method above.
Finally , let’s see the implementation of this protocol. Put the
following into calculus/remote_1.py :
# -*- test-case-name: calculus.test.test_remote_1 -*-
from twisted.protocols import basic
from twisted.internet import protocol
from calculus.base_3 import Calculation
class CalculationProxy:
def __init__(self):
self.calc = Calculation()
for m in ["add", "subtract", "multiply", "divide"]:
setattr(self, "remote_{}".format(m), getattr(self.calc, m))
class RemoteCalculationProtocol(basic.LineReceiver):
def __init__(self):
self.proxy = CalculationProxy()
def lineReceived(self, line):
op, a, b = line.decode("utf-8").split()
a = int(a)
b = int(b)
op = getattr(self.proxy, "remote_{}".format(op))
result = op(a, b)
self.sendLine(str(result).encode("utf-8"))
class RemoteCalculationFactory(protocol.Factory):
protocol = RemoteCalculationProtocol
def main():
from twisted.internet import reactor
from twisted.python import log
import sys
log.startLogging(sys.stdout)
reactor.listenTCP(0, RemoteCalculationFactory())
reactor.run()
if __name__ == "__main__":
main()
As mentioned, this server creates a protocol that inherits from basic.LineReceiver , and then a
factory that uses it as protocol. The only trick is the CalculationProxy object, which calls Calculation methods through remote_* methods. This pattern is used frequently in
Twisted, because it is very explicit about what methods you are making
accessible.
If you run this test (python -m twisted.trial calculus.test.test_remote_1 ), everything should be fine. You can also
run a server to test it with a telnet client. To do that, call python calculus/remote_1.py . You should have the following output:
2008-04-25 10:53:27+0200 [-] Log opened.
2008-04-25 10:53:27+0200 [-] __main__.RemoteCalculationFactory starting on 46194
2008-04-25 10:53:27+0200 [-] Starting factory <__main__.RemoteCalculationFactory instance at 0x846a0cc>
46194 is replaced by a random port. You can then call telnet on it:
$ telnet localhost 46194
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
add 4123 9423
13546
It works!
Of course, what we build is not particularly useful for now: we’ll now build a client for our server, to be able to use it inside a Python program. And it will serve our next purpose.
Create calculus/test/test_client_1.py:
from calculus.client_1 import RemoteCalculationClient
from twisted.trial import unittest
from twisted.test import proto_helpers
class ClientCalculationTestCase(unittest.TestCase):
def setUp(self):
self.tr = proto_helpers.StringTransport()
self.proto = RemoteCalculationClient()
self.proto.makeConnection(self.tr)
def _test(self, operation, a, b, expected):
d = getattr(self.proto, operation)(a, b)
self.assertEqual(
self.tr.value(), "{} {} {}\r\n".format(operation, a, b).encode("utf-8")
)
self.tr.clear()
d.addCallback(self.assertEqual, expected)
self.proto.dataReceived(
"{}\r\n".format(
expected,
).encode("utf-8")
)
return d
def test_add(self):
return self._test("add", 7, 6, 13)
def test_subtract(self):
return self._test("subtract", 82, 78, 4)
def test_multiply(self):
return self._test("multiply", 2, 8, 16)
def test_divide(self):
return self._test("divide", 14, 3, 4)
It’s really symmetric to the server test cases. The only tricky part is that we don’t use a client factory. We’re lazy, and it’s not very useful in the client part, so we instantiate the protocol directly.
Incidentally, we have introduced a very important concept here: the tests now return a Deferred object, and the assertion is done in a callback. When a test returns a Deferred, the reactor is run until the Deferred fires and its callbacks run. The important thing to do here is to not forget to return the Deferred . If you do, your tests will pass even if nothing is asserted. That’s also why it’s important to make tests fail first: if your tests pass whereas you know they shouldn’t, there is a problem in your tests.
We’ll now add the remote client class to produce calculus/client_1.py :
# -*- test-case-name: calculus.test.test_client_1 -*-
from twisted.protocols import basic
from twisted.internet import defer
class RemoteCalculationClient(basic.LineReceiver):
def __init__(self):
self.results = []
def lineReceived(self, line):
d = self.results.pop(0)
d.callback(int(line))
def _sendOperation(self, op, a, b):
d = defer.Deferred()
self.results.append(d)
line = "{} {} {}".format(op, a, b).encode("utf-8")
self.sendLine(line)
return d
def add(self, a, b):
return self._sendOperation("add", a, b)
def subtract(self, a, b):
return self._sendOperation("subtract", a, b)
def multiply(self, a, b):
return self._sendOperation("multiply", a, b)
def divide(self, a, b):
return self._sendOperation("divide", a, b)
When testing code that involves the passage of time, waiting e.g. for a two hour timeout to occur in a test is not very realistic. Twisted provides a solution to this, the Clock class that allows one to simulate the passage of time.
As an example we’ll test the code for client request timeout: since our client uses TCP it can hang for a long time (firewall, connectivity problems, etc…). So generally we need to implement timeouts on the client side. Basically it’s just that we send a request, don’t receive a response and expect a timeout error to be triggered after a certain duration.
from calculus.client_2 import RemoteCalculationClient, ClientTimeoutError
from twisted.internet import task
from twisted.trial import unittest
from twisted.test import proto_helpers
class ClientCalculationTestCase(unittest.TestCase):
def setUp(self):
self.tr = proto_helpers.StringTransportWithDisconnection()
self.clock = task.Clock()
self.proto = RemoteCalculationClient()
self.tr.protocol = self.proto
self.proto.callLater = self.clock.callLater
self.proto.makeConnection(self.tr)
def _test(self, operation, a, b, expected):
d = getattr(self.proto, operation)(a, b)
self.assertEqual(
self.tr.value(), "{} {} {}\r\n".format(operation, a, b).encode("utf-8")
)
self.tr.clear()
d.addCallback(self.assertEqual, expected)
self.proto.dataReceived("{}\r\n".format(expected).encode("utf-8"))
return d
def test_add(self):
return self._test("add", 7, 6, 13)
def test_subtract(self):
return self._test("subtract", 82, 78, 4)
def test_multiply(self):
return self._test("multiply", 2, 8, 16)
def test_divide(self):
return self._test("divide", 14, 3, 4)
def test_timeout(self):
d = self.proto.add(9, 4)
self.assertEqual(self.tr.value(), b"add 9 4\r\n")
self.clock.advance(self.proto.timeOut)
return self.assertFailure(d, ClientTimeoutError)
What happens here? We instantiate our protocol as usual, the only trick
is to create the clock, and assign proto.callLater to
clock.callLater . Thus, every callLater
call in the protocol will finish before clock.advance() returns.
In the new test (test_timeout), we call clock.advance , that simulates an advance in time
(logically it’s similar to a time.sleep call). And
we just have to verify that our Deferred got a timeout error.
Let’s implement that in our code.
# -*- test-case-name: calculus.test.test_client_2 -*-
from twisted.protocols import basic
from twisted.internet import defer, reactor
class ClientTimeoutError(Exception):
pass
class RemoteCalculationClient(basic.LineReceiver):
callLater = reactor.callLater
timeOut = 60
def __init__(self):
self.results = []
def lineReceived(self, line):
d, callID = self.results.pop(0)
callID.cancel()
d.callback(int(line))
def _cancel(self, d):
d.errback(ClientTimeoutError())
def _sendOperation(self, op, a, b):
d = defer.Deferred()
callID = self.callLater(self.timeOut, self._cancel, d)
self.results.append((d, callID))
line = "{} {} {}".format(op, a, b).encode("utf-8")
self.sendLine(line)
return d
def add(self, a, b):
return self._sendOperation("add", a, b)
def subtract(self, a, b):
return self._sendOperation("subtract", a, b)
def multiply(self, a, b):
return self._sendOperation("multiply", a, b)
def divide(self, a, b):
return self._sendOperation("divide", a, b)
If everything completed successfully, it is important to remember to cancel the DelayedCall returned by callLater .
This chapter is mainly intended for people who want to have sockets or processes created in their tests. If it’s still not obvious, you must try to avoid using them, because it ends up with a lot of problems, one of them being intermittent failures. And intermittent failures are the plague of automated tests.
To actually test that, we’ll launch a server with our protocol.
from calculus.remote_1 import RemoteCalculationFactory
from calculus.client_2 import RemoteCalculationClient
from twisted.trial import unittest
from twisted.internet import reactor, protocol
class RemoteRunCalculationTestCase(unittest.TestCase):
def setUp(self):
factory = RemoteCalculationFactory()
self.port = reactor.listenTCP(0, factory, interface="127.0.0.1")
self.client = None
def tearDown(self):
if self.client is not None:
self.client.transport.loseConnection()
return self.port.stopListening()
def _test(self, op, a, b, expected):
creator = protocol.ClientCreator(reactor, RemoteCalculationClient)
def cb(client):
self.client = client
return getattr(self.client, op)(a, b).addCallback(
self.assertEqual, expected
)
return creator.connectTCP("127.0.0.1", self.port.getHost().port).addCallback(cb)
def test_add(self):
return self._test("add", 5, 9, 14)
def test_subtract(self):
return self._test("subtract", 47, 13, 34)
def test_multiply(self):
return self._test("multiply", 7, 3, 21)
def test_divide(self):
return self._test("divide", 84, 10, 8)
Recent versions of Trial will fail loudly if you remove the stopListening call, which is good.
Also, you should be aware that tearDown will be called in any case, after success or failure.
So don’t expect every object you created in the test method to be present, because your tests may have failed in the middle.
Trial also has a addCleanup method, which makes these kind of cleanups easy and removes the need for tearDown.
For example, you could remove the code in _test this way:
def setUp(self):
factory = RemoteCalculationFactory()
self.port = reactor.listenTCP(0, factory, interface="127.0.0.1")
self.addCleanup(self.port.stopListening)
def _test(self, op, a, b, expected):
creator = protocol.ClientCreator(reactor, RemoteCalculationClient)
def cb(client):
self.addCleanup(self.client.transport.loseConnection)
return getattr(client, op)(a, b).addCallback(self.assertEqual, expected)
return creator.connectTCP('127.0.0.1', self.port.getHost().port).addCallback(cb)
This removes the need of a tearDown method, and you don’t have to check for
the value of self.client: you only call addCleanup when the client is
created.
Currently, if you send an invalid command or invalid arguments to our server, it logs an exception and closes the connection. This is a perfectly valid behavior, but for the sake of this tutorial, we want to return an error to the user if they send invalid operators, and log any errors on server side. So we’ll want a test like this:
def test_invalidParameters(self):
self.proto.dataReceived('add foo bar\r\n')
self.assertEqual(self.tr.value(), "error\r\n")
# -*- test-case-name: calculus.test.test_remote_1 -*-
from twisted.protocols import basic
from twisted.internet import protocol
from twisted.python import log
from calculus.base_3 import Calculation
class CalculationProxy:
def __init__(self):
self.calc = Calculation()
for m in ["add", "subtract", "multiply", "divide"]:
setattr(self, "remote_{}".format(m), getattr(self.calc, m))
class RemoteCalculationProtocol(basic.LineReceiver):
def __init__(self):
self.proxy = CalculationProxy()
def lineReceived(self, line):
op, a, b = line.decode("utf-8").split()
op = getattr(
self.proxy,
"remote_{}".format(
op,
),
)
try:
result = op(a, b)
except TypeError:
log.err()
self.sendLine(b"error")
else:
self.sendLine(str(result).encode("utf-8"))
class RemoteCalculationFactory(protocol.Factory):
protocol = RemoteCalculationProtocol
def main():
from twisted.internet import reactor
from twisted.python import log
import sys
log.startLogging(sys.stdout)
reactor.listenTCP(0, RemoteCalculationFactory())
reactor.run()
if __name__ == "__main__":
main()
If you try something like that, it will not work. Here is the output you should have:
$ python -m twisted.trial calculus.test.test_remote_3.RemoteCalculationTestCase.test_invalidParameters
calculus.test.test_remote_3
RemoteCalculationTestCase
test_invalidParameters ... [ERROR]
===============================================================================
[ERROR]: calculus.test.test_remote_3.RemoteCalculationTestCase.test_invalidParameters
Traceback (most recent call last):
File "/tmp/calculus/remote_2.py", line 27, in lineReceived
result = op(a, b)
File "/tmp/calculus/base_3.py", line 11, in add
a, b = self._make_ints(a, b)
File "/tmp/calculus/base_3.py", line 8, in _make_ints
raise TypeError
exceptions.TypeError:
-------------------------------------------------------------------------------
Ran 1 tests in 0.004s
FAILED (errors=1)
At first, you could think there is a problem, because you catch this
exception. But in fact Trial doesn’t let you do that without controlling it:
you must expect logged errors and clean them. To do that, you have to use the
flushLoggedErrors method. You call it with the
exception you expect, and it returns the list of exceptions logged since the
start of the test. Generally, you’ll want to check that this list has the
expected length, or possibly that each exception has an expected message. We do
the former in our test:
from calculus.remote_2 import RemoteCalculationFactory
from twisted.trial import unittest
from twisted.test import proto_helpers
class RemoteCalculationTestCase(unittest.TestCase):
def setUp(self):
factory = RemoteCalculationFactory()
self.proto = factory.buildProtocol(("127.0.0.1", 0))
self.tr = proto_helpers.StringTransport()
self.proto.makeConnection(self.tr)
def _test(self, operation, a, b, expected):
self.proto.dataReceived("{} {} {}\r\n".format(operation, a, b).encode("utf-8"))
self.assertEqual(int(self.tr.value()), expected)
def test_add(self):
return self._test("add", 7, 6, 13)
def test_subtract(self):
return self._test("subtract", 82, 78, 4)
def test_multiply(self):
return self._test("multiply", 2, 8, 16)
def test_divide(self):
return self._test("divide", 14, 3, 4)
def test_invalidParameters(self):
self.proto.dataReceived(b"add foo bar\r\n")
self.assertEqual(self.tr.value(), b"error\r\n")
errors = self.flushLoggedErrors(TypeError)
self.assertEqual(len(errors), 1)
A bug was left over during the development of the timeout (probably several bugs, but that’s not the point), concerning the reuse of the protocol when you got a timeout: the connection is not dropped, so you can get timeout forever. Generally a user will come to you saying “I have this strange problem on my crappy network. It seems you could solve it with doing XXX at YYY.”
Actually, this bug can be corrected several ways. But if you correct it without adding tests, one day you’ll face a big problem: regression. So the first step is adding a failing test.
from calculus.client_3 import RemoteCalculationClient, ClientTimeoutError
from twisted.internet import task
from twisted.trial import unittest
from twisted.test import proto_helpers
class ClientCalculationTestCase(unittest.TestCase):
def setUp(self):
self.tr = proto_helpers.StringTransportWithDisconnection()
self.clock = task.Clock()
self.proto = RemoteCalculationClient()
self.tr.protocol = self.proto
self.proto.callLater = self.clock.callLater
self.proto.makeConnection(self.tr)
def _test(self, operation, a, b, expected):
d = getattr(self.proto, operation)(a, b)
self.assertEqual(
self.tr.value(), "{} {} {}\r\n".format(operation, a, b).encode("utf-8")
)
self.tr.clear()
d.addCallback(self.assertEqual, expected)
self.proto.dataReceived("{}\r\n".format(expected).encode("utf-8"))
return d
def test_add(self):
return self._test("add", 7, 6, 13)
def test_subtract(self):
return self._test("subtract", 82, 78, 4)
def test_multiply(self):
return self._test("multiply", 2, 8, 16)
def test_divide(self):
return self._test("divide", 14, 3, 4)
def test_timeout(self):
d = self.proto.add(9, 4)
self.assertEqual(self.tr.value(), b"add 9 4\r\n")
self.clock.advance(self.proto.timeOut)
return self.assertFailure(d, ClientTimeoutError)
def test_timeoutConnectionLost(self):
called = []
def lost(arg):
called.append(True)
self.proto.connectionLost = lost
d = self.proto.add(9, 4)
self.assertEqual(self.tr.value(), b"add 9 4\r\n")
self.clock.advance(self.proto.timeOut)
def check(ignore):
self.assertEqual(called, [True])
return self.assertFailure(d, ClientTimeoutError).addCallback(check)
What have we done here?
We switched to StringTransportWithDisconnection. This transport manages
loseConnection and forwards it to its protocol.
We assign the protocol to the transport via the protocol attribute.
We check that after a timeout our connection has closed.
For doing that, we then use the TimeoutMixin
class, that does almost everything we want. The great thing is that it almost
changes nothing to our class.
# -*- test-case-name: calculus.test.test_client -*-
from twisted.protocols import basic, policies
from twisted.internet import defer
class ClientTimeoutError(Exception):
pass
class RemoteCalculationClient(basic.LineReceiver, policies.TimeoutMixin):
def __init__(self):
self.results = []
self._timeOut = 60
def lineReceived(self, line):
self.setTimeout(None)
d = self.results.pop(0)
d.callback(int(line))
def timeoutConnection(self):
for d in self.results:
d.errback(ClientTimeoutError())
self.transport.loseConnection()
def _sendOperation(self, op, a, b):
d = defer.Deferred()
self.results.append(d)
line = "{} {} {}".format(op, a, b).encode("utf-8")
self.sendLine(line)
self.setTimeout(self._timeOut)
return d
def add(self, a, b):
return self._sendOperation("add", a, b)
def subtract(self, a, b):
return self._sendOperation("subtract", a, b)
def multiply(self, a, b):
return self._sendOperation("multiply", a, b)
def divide(self, a, b):
return self._sendOperation("divide", a, b)
Above we learned about returning Deferreds from test methods in order to make
assertions about their results, or side-effects that only happen after they
fire. This can be useful, but we don’t actually need the feature in this
example. Because we were careful to use Clock , we
don’t need the global reactor to run in our tests. Instead of returning the
Deferred with a callback attached to it which performs the necessary assertions,
we can use a testing helper,
successResultOf (and
the corresponding error-case helper
failureResultOf ), to
extract its result and make assertions against it directly. Compared to
returning a Deferred, this avoids the problem of forgetting to return the
Deferred, improves the stack trace reported when the assertion fails, and avoids
the complexity of using global reactor (which, for example, may then require
cleanup).
from calculus.client_3 import RemoteCalculationClient, ClientTimeoutError
from twisted.internet import task
from twisted.trial import unittest
from twisted.test import proto_helpers
class ClientCalculationTestCase(unittest.TestCase):
def setUp(self):
self.tr = proto_helpers.StringTransportWithDisconnection()
self.clock = task.Clock()
self.proto = RemoteCalculationClient()
self.tr.protocol = self.proto
self.proto.callLater = self.clock.callLater
self.proto.makeConnection(self.tr)
def _test(self, operation, a, b, expected):
d = getattr(self.proto, operation)(a, b)
self.assertEqual(
self.tr.value(), "{} {} {}\r\n".format(operation, a, b).encode("utf-8")
)
self.tr.clear()
self.proto.dataReceived("{}\r\n".format(expected).encode("utf-8"))
self.assertEqual(expected, self.successResultOf(d))
def test_add(self):
self._test("add", 7, 6, 13)
def test_subtract(self):
self._test("subtract", 82, 78, 4)
def test_multiply(self):
self._test("multiply", 2, 8, 16)
def test_divide(self):
self._test("divide", 14, 3, 4)
def test_timeout(self):
d = self.proto.add(9, 4)
self.assertEqual(self.tr.value(), b"add 9 4\r\n")
self.clock.advance(self.proto.timeOut)
self.failureResultOf(d).trap(ClientTimeoutError)
def test_timeoutConnectionLost(self):
called = []
def lost(arg):
called.append(True)
self.proto.connectionLost = lost
d = self.proto.add(9, 4)
self.assertEqual(self.tr.value(), b"add 9 4\r\n")
self.clock.advance(self.proto.timeOut)
def check(ignore):
self.assertEqual(called, [True])
self.failureResultOf(d).trap(ClientTimeoutError)
self.assertEqual(called, [True])
This version of the code makes the same assertions, but no longer returns any
Deferreds from any test methods. Instead of making assertions about the result
of the Deferred in a callback, it makes the assertions as soon as
it knows the Deferred is supposed to have a result (in
the _test method and in test_timeout
and test_timeoutConnectionLost ). The possibility
of knowing exactly when a Deferred is supposed to have a test is what
makes successResultOf useful in unit testing, but prevents it from being
applicable to non-testing purposes.
successResultOf will raise an exception (failing the test) if
the Deferred passed to it does not have a result, or has a failure
result. Similarly, failureResultOf will raise an exception (also
failing the test) if the Deferred passed to it does not have a
result, or has a success result. There is a third helper method for testing the
final case,
assertNoResult ,
which only raises an exception (failing the test) if the Deferred passed
to it has a result (either success or failure).
In the course of writing and running your tests, it is often helpful to
employ the use of a debugger. This can be particularly helpful in tracking down
where the source of a troublesome bug is in your code. Python’s standard library
includes a debugger in the form of the pdb module.
Running your tests with pdb is as simple as invoking
twisted with the --debug option, which will start pdb at the beginning of the execution of your test
suite.
Trial also provides a --debugger option which can
run your test suite using another debugger instead. To specify a debugger other
than pdb , pass in the fully-qualified name of an
object that provides the same interface as pdb .
Most third-party debuggers tend to implement an interface similar to pdb , or at least provide a wrapper object that
does. For example, invoking Trial with the extra arguments --debug --debugger pudb will open the PuDB debugger instead, provided
it is properly installed.
Code coverage is one of the aspects of software testing that shows how much your tests cross (cover) the code of your program. There are different kinds of measures: path coverage, condition coverage, statement coverage… We’ll only consider statement coverage here, whether a line has been executed or not.
Trial has an option to generate the statement coverage of your tests. This option is –coverage. It creates a coverage directory in _trial_temp, with a file .cover for every module used during the tests. The ones interesting for us are calculus.base.cover and calculus.remote.cover. Each line starts with a counter showing how many times the line was executed during the tests, or the marker ‘>>>>>>’ if the line was not covered. If you went through the whole tutorial to this point, you should have complete coverage :).
Again, this is only another useful pointer, but it doesn’t mean your code is perfect: your tests should consider every possible input and output, to get full coverage (condition, path, etc.) as well .
So what did you learn in this document?
How to use the Trial command-line tool to run your tests
How to use string transports to test individual clients and servers without creating sockets
If you really want to create sockets, how to cleanly do it so that it doesn’t have bad side effects
And some small tips you can’t live without.
If one of the topics still looks cloudy to you, please give us your feedback! You can file tickets to improve this document - learn how to contribute on the Twisted web site.