The Practice of TDD: tips&tricks

The practice of TDD: tips&tricks
Antonio Cuni
PyCon Nove
antocuni (PyCon Nove) TDD tips&ticks 1 / 35

View Slide

About me
PyPy core dev since 2006
pdb++, cffi, vmprof, capnpy, ...
@antocuni
http://antocuni.eu

View Slide

About you
Either:
novice programmer
experienced programmer but new to Python and/or
TDD

View Slide

About this talk
Two parts
General TDD principles
Useful patterns and tips

View Slide

The goal of testing
Make sure that your code works
WRONG!

View Slide

The goal of testing
Make sure that your code works
Make sure that your code does NOT break

View Slide

Manual testing (1)
Feature A
Write the code
Start the program
Navigate through N steps
login
click on few links
push a button
CRASH!
(repeat)

View Slide

Manual testing (2)
Feature B
Modify the code
Feature B works! :-)
(Feature A no longer works, but you don’t notice)

View Slide

Automated testing (1)
Write the code for Feature A
Run a command (~0.5 secs)
py.test test_foo.py -k test_feature_A
CRASH
(repeat)

View Slide

Write the code for Feature B
Run a command (~0.5 secs)
py.test test_foo.py -k test_feature_B
It works
py.test test_foo.py
CRASH
ﬁx
Everyone is happy :-)

View Slide

What’s the missing piece?
You have to write the test!
It’s just a program
test frameworks/runners offer a lot of help
unittest, unittest2
nose
“py.test“

View Slide

Test Driven Development
Goal: make sure that our code does not break
What is not tested is broken (aka: Murphy’s law)
Even if it’s not broken right now, it’ll eventually break

View Slide

Tests first
Writing code when no test is failing is forbidden
You should write just the code to make the test
passing
don’t cheat :-)
Each test must run in isolation
bonus track: VCS
commit every time you write/fix a test
write meaningful commit messages
don’t commit if the tests are broken
(unless you are sure it’s the right thing to do :-))
make you confident in your tests
{hg,git} bisect

View Slide

TDD benefits
confidence about the quality of the code
easily spot regressions
easily find by who/when/why a regression was
introduced
"Why the hell did I write this piece of code?"
look at the commit, and the corresponding test
Remove the code, and see if/which tests fail
"One of my most productive days was throwing away 1000
lines of code" (Ken Thompson)
"Deleted code is debugged code" (Jeff Sickel)
The power of refactoring

View Slide

Properties of a good test
It should FAIL before your ﬁx
write the test ﬁrst, then the code
Determinism
NEVER write a test which fails every other run
pay attention e.g. to dictionary order
Easy to READ
executable documentation

View Slide

Readability
A test tells a story
One feature per test
No complex control ﬂow
Clear failures
When a test fails, the poor soul looking at it should be able
to understand why

View Slide

Factorial example
Bad
def test_factorial():
for n in (5, 7):
res = 1
for i in range(1, n+1):
res *= i
assert factorial(n) == res

View Slide

Factorial example

View Slide

Factorial example
Better
assert factorial(5) == 120
assert factorial(7) == 5040

View Slide

Factorial example
Best
assert factorial(5) == 2 * 3 * 4 * 5
assert factorial(7) == 2 * 3 * 4 * 5 * 6 * 7

View Slide

Easy to write
We are all lazy, deal with it
The easiest to write a test, the more likely we’ll do
Invest time in a proper test infrastructure
Test the infrastructure as well :)

View Slide

test_pypy_c example
Bad
def test_call_pypy(tmpdir):
src = """if 1:
def factorial(n):
if n in (0, 1):
return 1
return n * factorial(n-1)
import sys
n = eval(sys.argv[1])
print factorial(n)
"""
pyfile = tmpdir.join("x.py")
pyfile.write(src)
stdout = subprocess.check_output(
["pypy", str(pyfile), "5"])
res = eval(stdout)
assert res == 2*3*4*5

View Slide

test_pypy_c example
Still bad
def execute(tmpdir, src, *args):
pyfile = tmpdir.join("x.py")
pyfile.write(src)
args = ["pypy", str(pyfile)] + list(args)
stdout = subprocess.check_output(args)
return eval(stdout)
def test_call_pypy_2(tmpdir):
src = """if 1:
def factorial(n):
if n in (0, 1):
return 1
import sys
n = eval(sys.argv[1])
print factorial(n)
"""
res = execute(tmpdir, src, "5")

View Slide

test_pypy_c example
Good
class TestCall(PyPyTest):
def test_call(self):
def factorial(n):
if n in (0, 1):
return 1
#
res = self.run(factorial, 5)

View Slide

test_pypy_c example
@pytest.mark.usefixtures(’initargs’)
class PyPyTest(object):
@pytest.fixture
def initargs(self, tmpdir):
self.tmpdir = tmpdir
def run(self, fn, *args):
fnsrc = textwrap.dedent(inspect.getsource(fn))
boilerplate = textwrap.dedent("""
import sys
args = map(eval, sys.argv[1:])
print {fnname}(*args)
""").format(fnname=fn.__name__)
src = fnsrc + boilerplate
args = map(str, args)
return execute(self.tmpdir, src, *args)

View Slide

test_pypy_c example
pypy/module/pypyjit/test/model.py
https://bitbucket.org/pypy/pypy/src/
5a1232a1c2e6e01422713659c57ae3bcd5db3f70/pypy/
module/pypyjit/test_pypy_c/model.py?at=default&
fileviewer=file-view-default
pypy/module/pypyjit/test/test_00model.py
5a1232a1c2e6e01422713659c57ae3bcd5db3f70/pypy/
module/pypyjit/test_pypy_c/test_00_model.py?at=
default&fileviewer=file-view-default
capnpy/testing/compiler/support.py
https://github.com/antocuni/capnpy/blob/master/
capnpy/testing/compiler/support.py

View Slide

Decoupling components
Useful in general
In particular for GUIs
Leads to a better design

View Slide

Google for tkinter tutorial
http:
//python-textbok.readthedocs.io/en/1.
0/Introduction_to_GUI_Programming.html

View Slide

class Calculator:
def __init__(self, master):
...
self.total = 0
self.entered_number = 0
self.total_label_text = IntVar()
self.total_label_text.set(self.total)
self.add_button = ...
self.subtract_button = ...
self.reset_button = ...
def update(self, method):
if method == "add":
self.total += self.entered_number
elif method == "subtract":
self.total -= self.entered_number
else: # reset
self.total = 0
self.total_label_text.set(self.total)
self.entry.delete(0, END)

View Slide

class Calculator(object):
def __init__(self):
self.total = 0
def add(self, x):
self.total += x
def sub(self, x):
self.total -= x
def reset(self):
self.total = 0

View Slide

class GUI(object):
def __init__(self, master):
self.calculator = Calculator()
...
self.add_button = ...
self.subtract_button = ...
self.reset_button = ...
self.update_total()
def update_total(self):
self.total_label_text.set(self.calculator.total)
def update(self, method):
if method == "add":
self.calculator.add(self.entered_number)
elif method == "subtract":
self.calculator.sub(self.entered_number)
else: # reset
self.calculator.reset()
self.update_total()
self.entry.delete(0, END)

View Slide

def test_Calculator():
calc = Calculator()
assert calc.total == 0
calc.add(3)
calc.add(5)
calc.sub(7)
calc.sub(10)
assert calc.total == -9
calc.reset()

View Slide

Notation
Chapter 9: Notation

View Slide

Notation
ops = [
MergePoint(’merge_point’, [sum, n1], []),
ResOperation(’guard_class’, [n1, ConstAddr(node_vtable, cpu)], []),
ResOperation(’getfield_gc__4’, [n1, ConstInt(ofs_value)], [v]),
ResOperation(’int_sub’, [v, ConstInt(1)], [v2]),
ResOperation(’int_add’, [sum, v], [sum2]),
ResOperation(’new_with_vtable’, [ConstInt(size_of_node),
ConstAddr(node_vtable, cpu)], [n2]),
ResOperation(’setfield_gc__4’, [n2, ConstInt(ofs_value), v2], []),
Jump(’jump’, [sum2, n2], []),
]

View Slide

Notation
5a1232a1c2e6e01422713659c57ae3bcd5db3f70/rpython/jit/metainterp/
optimizeopt/test/test_optimizeopt.py?at=default&fileviewer=
file-view-default
def test_remove_guard_class_1(self):
ops = """
[p0]
guard_class(p0, ConstClass(node_vtable)) []
jump(p0)
"""
preamble = """
[p0]
jump(p0)
"""
expected = """
[p0]
jump(p0)
"""
self.optimize_loop(ops, expected, expected_preamble=preamble)

View Slide

Q&A
@antocuni
https:
//bitbucket.org/antocuni/misc/src/
704c99980401dbb638387ded33e6c23157a683
talks/tdd-tips-and-tricks/?at=
default

View Slide

The Practice of TDD: tips&tricks

The Practice of TDD: tips&tricks

Antonio Cuni

More Decks by Antonio Cuni

Other Decks in Programming

Featured

Transcript