D7net Mini Sh3LL v1
Current File : //lib/sudo/../openssh/../python3/dist-packages/uaclient/../more_itertools/tests/test_more.py |
from __future__ import division, print_function, unicode_literals
from decimal import Decimal
from doctest import DocTestSuite
from fractions import Fraction
from functools import partial, reduce
from heapq import merge
from io import StringIO
from itertools import (
chain,
count,
groupby,
islice,
permutations,
product,
repeat,
)
from operator import add, mul, itemgetter
from unittest import TestCase
from six.moves import filter, map, range, zip
import more_itertools as mi
def load_tests(loader, tests, ignore):
# Add the doctests
tests.addTests(DocTestSuite('more_itertools.more'))
return tests
class CollateTests(TestCase):
"""Unit tests for ``collate()``"""
# Also accidentally tests peekable, though that could use its own tests
def test_default(self):
"""Test with the default `key` function."""
iterables = [range(4), range(7), range(3, 6)]
self.assertEqual(
sorted(reduce(list.__add__, [list(it) for it in iterables])),
list(mi.collate(*iterables))
)
def test_key(self):
"""Test using a custom `key` function."""
iterables = [range(5, 0, -1), range(4, 0, -1)]
actual = sorted(
reduce(list.__add__, [list(it) for it in iterables]), reverse=True
)
expected = list(mi.collate(*iterables, key=lambda x: -x))
self.assertEqual(actual, expected)
def test_empty(self):
"""Be nice if passed an empty list of iterables."""
self.assertEqual([], list(mi.collate()))
def test_one(self):
"""Work when only 1 iterable is passed."""
self.assertEqual([0, 1], list(mi.collate(range(2))))
def test_reverse(self):
"""Test the `reverse` kwarg."""
iterables = [range(4, 0, -1), range(7, 0, -1), range(3, 6, -1)]
actual = sorted(
reduce(list.__add__, [list(it) for it in iterables]), reverse=True
)
expected = list(mi.collate(*iterables, reverse=True))
self.assertEqual(actual, expected)
def test_alias(self):
self.assertNotEqual(merge.__doc__, mi.collate.__doc__)
self.assertNotEqual(partial.__doc__, mi.collate.__doc__)
class ChunkedTests(TestCase):
"""Tests for ``chunked()``"""
def test_even(self):
"""Test when ``n`` divides evenly into the length of the iterable."""
self.assertEqual(
list(mi.chunked('ABCDEF', 3)), [['A', 'B', 'C'], ['D', 'E', 'F']]
)
def test_odd(self):
"""Test when ``n`` does not divide evenly into the length of the
iterable.
"""
self.assertEqual(
list(mi.chunked('ABCDE', 3)), [['A', 'B', 'C'], ['D', 'E']]
)
class FirstTests(TestCase):
"""Tests for ``first()``"""
def test_many(self):
"""Test that it works on many-item iterables."""
# Also try it on a generator expression to make sure it works on
# whatever those return, across Python versions.
self.assertEqual(mi.first(x for x in range(4)), 0)
def test_one(self):
"""Test that it doesn't raise StopIteration prematurely."""
self.assertEqual(mi.first([3]), 3)
def test_empty_stop_iteration(self):
"""It should raise StopIteration for empty iterables."""
self.assertRaises(ValueError, lambda: mi.first([]))
def test_default(self):
"""It should return the provided default arg for empty iterables."""
self.assertEqual(mi.first([], 'boo'), 'boo')
class PeekableTests(TestCase):
"""Tests for ``peekable()`` behavor not incidentally covered by testing
``collate()``
"""
def test_peek_default(self):
"""Make sure passing a default into ``peek()`` works."""
p = mi.peekable([])
self.assertEqual(p.peek(7), 7)
def test_truthiness(self):
"""Make sure a ``peekable`` tests true iff there are items remaining in
the iterable.
"""
p = mi.peekable([])
self.assertFalse(p)
p = mi.peekable(range(3))
self.assertTrue(p)
def test_simple_peeking(self):
"""Make sure ``next`` and ``peek`` advance and don't advance the
iterator, respectively.
"""
p = mi.peekable(range(10))
self.assertEqual(next(p), 0)
self.assertEqual(p.peek(), 1)
self.assertEqual(next(p), 1)
def test_indexing(self):
"""
Indexing into the peekable shouldn't advance the iterator.
"""
p = mi.peekable('abcdefghijkl')
# The 0th index is what ``next()`` will return
self.assertEqual(p[0], 'a')
self.assertEqual(next(p), 'a')
# Indexing further into the peekable shouldn't advance the itertor
self.assertEqual(p[2], 'd')
self.assertEqual(next(p), 'b')
# The 0th index moves up with the iterator; the last index follows
self.assertEqual(p[0], 'c')
self.assertEqual(p[9], 'l')
self.assertEqual(next(p), 'c')
self.assertEqual(p[8], 'l')
# Negative indexing should work too
self.assertEqual(p[-2], 'k')
self.assertEqual(p[-9], 'd')
self.assertRaises(IndexError, lambda: p[-10])
def test_slicing(self):
"""Slicing the peekable shouldn't advance the iterator."""
seq = list('abcdefghijkl')
p = mi.peekable(seq)
# Slicing the peekable should just be like slicing a re-iterable
self.assertEqual(p[1:4], seq[1:4])
# Advancing the iterator moves the slices up also
self.assertEqual(next(p), 'a')
self.assertEqual(p[1:4], seq[1:][1:4])
# Implicit starts and stop should work
self.assertEqual(p[:5], seq[1:][:5])
self.assertEqual(p[:], seq[1:][:])
# Indexing past the end should work
self.assertEqual(p[:100], seq[1:][:100])
# Steps should work, including negative
self.assertEqual(p[::2], seq[1:][::2])
self.assertEqual(p[::-1], seq[1:][::-1])
def test_slicing_reset(self):
"""Test slicing on a fresh iterable each time"""
iterable = ['0', '1', '2', '3', '4', '5']
indexes = list(range(-4, len(iterable) + 4)) + [None]
steps = [1, 2, 3, 4, -1, -2, -3, 4]
for slice_args in product(indexes, indexes, steps):
it = iter(iterable)
p = mi.peekable(it)
next(p)
index = slice(*slice_args)
actual = p[index]
expected = iterable[1:][index]
self.assertEqual(actual, expected, slice_args)
def test_slicing_error(self):
iterable = '01234567'
p = mi.peekable(iter(iterable))
# Prime the cache
p.peek()
old_cache = list(p._cache)
# Illegal slice
with self.assertRaises(ValueError):
p[1:-1:0]
# Neither the cache nor the iteration should be affected
self.assertEqual(old_cache, list(p._cache))
self.assertEqual(list(p), list(iterable))
def test_passthrough(self):
"""Iterating a peekable without using ``peek()`` or ``prepend()``
should just give the underlying iterable's elements (a trivial test but
useful to set a baseline in case something goes wrong)"""
expected = [1, 2, 3, 4, 5]
actual = list(mi.peekable(expected))
self.assertEqual(actual, expected)
# prepend() behavior tests
def test_prepend(self):
"""Tests intersperesed ``prepend()`` and ``next()`` calls"""
it = mi.peekable(range(2))
actual = []
# Test prepend() before next()
it.prepend(10)
actual += [next(it), next(it)]
# Test prepend() between next()s
it.prepend(11)
actual += [next(it), next(it)]
# Test prepend() after source iterable is consumed
it.prepend(12)
actual += [next(it)]
expected = [10, 0, 11, 1, 12]
self.assertEqual(actual, expected)
def test_multi_prepend(self):
"""Tests prepending multiple items and getting them in proper order"""
it = mi.peekable(range(5))
actual = [next(it), next(it)]
it.prepend(10, 11, 12)
it.prepend(20, 21)
actual += list(it)
expected = [0, 1, 20, 21, 10, 11, 12, 2, 3, 4]
self.assertEqual(actual, expected)
def test_empty(self):
"""Tests prepending in front of an empty iterable"""
it = mi.peekable([])
it.prepend(10)
actual = list(it)
expected = [10]
self.assertEqual(actual, expected)
def test_prepend_truthiness(self):
"""Tests that ``__bool__()`` or ``__nonzero__()`` works properly
with ``prepend()``"""
it = mi.peekable(range(5))
self.assertTrue(it)
actual = list(it)
self.assertFalse(it)
it.prepend(10)
self.assertTrue(it)
actual += [next(it)]
self.assertFalse(it)
expected = [0, 1, 2, 3, 4, 10]
self.assertEqual(actual, expected)
def test_multi_prepend_peek(self):
"""Tests prepending multiple elements and getting them in reverse order
while peeking"""
it = mi.peekable(range(5))
actual = [next(it), next(it)]
self.assertEqual(it.peek(), 2)
it.prepend(10, 11, 12)
self.assertEqual(it.peek(), 10)
it.prepend(20, 21)
self.assertEqual(it.peek(), 20)
actual += list(it)
self.assertFalse(it)
expected = [0, 1, 20, 21, 10, 11, 12, 2, 3, 4]
self.assertEqual(actual, expected)
def test_prepend_after_stop(self):
"""Test resuming iteration after a previous exhaustion"""
it = mi.peekable(range(3))
self.assertEqual(list(it), [0, 1, 2])
self.assertRaises(StopIteration, lambda: next(it))
it.prepend(10)
self.assertEqual(next(it), 10)
self.assertRaises(StopIteration, lambda: next(it))
def test_prepend_slicing(self):
"""Tests interaction between prepending and slicing"""
seq = list(range(20))
p = mi.peekable(seq)
p.prepend(30, 40, 50)
pseq = [30, 40, 50] + seq # pseq for prepended_seq
# adapt the specific tests from test_slicing
self.assertEqual(p[0], 30)
self.assertEqual(p[1:8], pseq[1:8])
self.assertEqual(p[1:], pseq[1:])
self.assertEqual(p[:5], pseq[:5])
self.assertEqual(p[:], pseq[:])
self.assertEqual(p[:100], pseq[:100])
self.assertEqual(p[::2], pseq[::2])
self.assertEqual(p[::-1], pseq[::-1])
def test_prepend_indexing(self):
"""Tests interaction between prepending and indexing"""
seq = list(range(20))
p = mi.peekable(seq)
p.prepend(30, 40, 50)
self.assertEqual(p[0], 30)
self.assertEqual(next(p), 30)
self.assertEqual(p[2], 0)
self.assertEqual(next(p), 40)
self.assertEqual(p[0], 50)
self.assertEqual(p[9], 8)
self.assertEqual(next(p), 50)
self.assertEqual(p[8], 8)
self.assertEqual(p[-2], 18)
self.assertEqual(p[-9], 11)
self.assertRaises(IndexError, lambda: p[-21])
def test_prepend_iterable(self):
"""Tests prepending from an iterable"""
it = mi.peekable(range(5))
# Don't directly use the range() object to avoid any range-specific
# optimizations
it.prepend(*(x for x in range(5)))
actual = list(it)
expected = list(chain(range(5), range(5)))
self.assertEqual(actual, expected)
def test_prepend_many(self):
"""Tests that prepending a huge number of elements works"""
it = mi.peekable(range(5))
# Don't directly use the range() object to avoid any range-specific
# optimizations
it.prepend(*(x for x in range(20000)))
actual = list(it)
expected = list(chain(range(20000), range(5)))
self.assertEqual(actual, expected)
def test_prepend_reversed(self):
"""Tests prepending from a reversed iterable"""
it = mi.peekable(range(3))
it.prepend(*reversed((10, 11, 12)))
actual = list(it)
expected = [12, 11, 10, 0, 1, 2]
self.assertEqual(actual, expected)
class ConsumerTests(TestCase):
"""Tests for ``consumer()``"""
def test_consumer(self):
@mi.consumer
def eater():
while True:
x = yield # noqa
e = eater()
e.send('hi') # without @consumer, would raise TypeError
class DistinctPermutationsTests(TestCase):
def test_distinct_permutations(self):
"""Make sure the output for ``distinct_permutations()`` is the same as
set(permutations(it)).
"""
iterable = ['z', 'a', 'a', 'q', 'q', 'q', 'y']
test_output = sorted(mi.distinct_permutations(iterable))
ref_output = sorted(set(permutations(iterable)))
self.assertEqual(test_output, ref_output)
def test_other_iterables(self):
"""Make sure ``distinct_permutations()`` accepts a different type of
iterables.
"""
# a generator
iterable = (c for c in ['z', 'a', 'a', 'q', 'q', 'q', 'y'])
test_output = sorted(mi.distinct_permutations(iterable))
# "reload" it
iterable = (c for c in ['z', 'a', 'a', 'q', 'q', 'q', 'y'])
ref_output = sorted(set(permutations(iterable)))
self.assertEqual(test_output, ref_output)
# an iterator
iterable = iter(['z', 'a', 'a', 'q', 'q', 'q', 'y'])
test_output = sorted(mi.distinct_permutations(iterable))
# "reload" it
iterable = iter(['z', 'a', 'a', 'q', 'q', 'q', 'y'])
ref_output = sorted(set(permutations(iterable)))
self.assertEqual(test_output, ref_output)
class IlenTests(TestCase):
def test_ilen(self):
"""Sanity-checks for ``ilen()``."""
# Non-empty
self.assertEqual(
mi.ilen(filter(lambda x: x % 10 == 0, range(101))), 11
)
# Empty
self.assertEqual(mi.ilen((x for x in range(0))), 0)
# Iterable with __len__
self.assertEqual(mi.ilen(list(range(6))), 6)
class WithIterTests(TestCase):
def test_with_iter(self):
s = StringIO('One fish\nTwo fish')
initial_words = [line.split()[0] for line in mi.with_iter(s)]
# Iterable's items should be faithfully represented
self.assertEqual(initial_words, ['One', 'Two'])
# The file object should be closed
self.assertEqual(s.closed, True)
class OneTests(TestCase):
def test_basic(self):
it = iter(['item'])
self.assertEqual(mi.one(it), 'item')
def test_too_short(self):
it = iter([])
self.assertRaises(ValueError, lambda: mi.one(it))
self.assertRaises(IndexError, lambda: mi.one(it, too_short=IndexError))
def test_too_long(self):
it = count()
self.assertRaises(ValueError, lambda: mi.one(it)) # burn 0 and 1
self.assertEqual(next(it), 2)
self.assertRaises(
OverflowError, lambda: mi.one(it, too_long=OverflowError)
)
class IntersperseTest(TestCase):
""" Tests for intersperse() """
def test_even(self):
iterable = (x for x in '01')
self.assertEqual(
list(mi.intersperse(None, iterable)), ['0', None, '1']
)
def test_odd(self):
iterable = (x for x in '012')
self.assertEqual(
list(mi.intersperse(None, iterable)), ['0', None, '1', None, '2']
)
def test_nested(self):
element = ('a', 'b')
iterable = (x for x in '012')
actual = list(mi.intersperse(element, iterable))
expected = ['0', ('a', 'b'), '1', ('a', 'b'), '2']
self.assertEqual(actual, expected)
def test_not_iterable(self):
self.assertRaises(TypeError, lambda: mi.intersperse('x', 1))
def test_n(self):
for n, element, expected in [
(1, '_', ['0', '_', '1', '_', '2', '_', '3', '_', '4', '_', '5']),
(2, '_', ['0', '1', '_', '2', '3', '_', '4', '5']),
(3, '_', ['0', '1', '2', '_', '3', '4', '5']),
(4, '_', ['0', '1', '2', '3', '_', '4', '5']),
(5, '_', ['0', '1', '2', '3', '4', '_', '5']),
(6, '_', ['0', '1', '2', '3', '4', '5']),
(7, '_', ['0', '1', '2', '3', '4', '5']),
(3, ['a', 'b'], ['0', '1', '2', ['a', 'b'], '3', '4', '5']),
]:
iterable = (x for x in '012345')
actual = list(mi.intersperse(element, iterable, n=n))
self.assertEqual(actual, expected)
def test_n_zero(self):
self.assertRaises(
ValueError, lambda: list(mi.intersperse('x', '012', n=0))
)
class UniqueToEachTests(TestCase):
"""Tests for ``unique_to_each()``"""
def test_all_unique(self):
"""When all the input iterables are unique the output should match
the input."""
iterables = [[1, 2], [3, 4, 5], [6, 7, 8]]
self.assertEqual(mi.unique_to_each(*iterables), iterables)
def test_duplicates(self):
"""When there are duplicates in any of the input iterables that aren't
in the rest, those duplicates should be emitted."""
iterables = ["mississippi", "missouri"]
self.assertEqual(
mi.unique_to_each(*iterables), [['p', 'p'], ['o', 'u', 'r']]
)
def test_mixed(self):
"""When the input iterables contain different types the function should
still behave properly"""
iterables = ['x', (i for i in range(3)), [1, 2, 3], tuple()]
self.assertEqual(mi.unique_to_each(*iterables), [['x'], [0], [3], []])
class WindowedTests(TestCase):
"""Tests for ``windowed()``"""
def test_basic(self):
actual = list(mi.windowed([1, 2, 3, 4, 5], 3))
expected = [(1, 2, 3), (2, 3, 4), (3, 4, 5)]
self.assertEqual(actual, expected)
def test_large_size(self):
"""
When the window size is larger than the iterable, and no fill value is
given,``None`` should be filled in.
"""
actual = list(mi.windowed([1, 2, 3, 4, 5], 6))
expected = [(1, 2, 3, 4, 5, None)]
self.assertEqual(actual, expected)
def test_fillvalue(self):
"""
When sizes don't match evenly, the given fill value should be used.
"""
iterable = [1, 2, 3, 4, 5]
for n, kwargs, expected in [
(6, {}, [(1, 2, 3, 4, 5, '!')]), # n > len(iterable)
(3, {'step': 3}, [(1, 2, 3), (4, 5, '!')]), # using ``step``
]:
actual = list(mi.windowed(iterable, n, fillvalue='!', **kwargs))
self.assertEqual(actual, expected)
def test_zero(self):
"""When the window size is zero, an empty tuple should be emitted."""
actual = list(mi.windowed([1, 2, 3, 4, 5], 0))
expected = [tuple()]
self.assertEqual(actual, expected)
def test_negative(self):
"""When the window size is negative, ValueError should be raised."""
with self.assertRaises(ValueError):
list(mi.windowed([1, 2, 3, 4, 5], -1))
def test_step(self):
"""The window should advance by the number of steps provided"""
iterable = [1, 2, 3, 4, 5, 6, 7]
for n, step, expected in [
(3, 2, [(1, 2, 3), (3, 4, 5), (5, 6, 7)]), # n > step
(3, 3, [(1, 2, 3), (4, 5, 6), (7, None, None)]), # n == step
(3, 4, [(1, 2, 3), (5, 6, 7)]), # line up nicely
(3, 5, [(1, 2, 3), (6, 7, None)]), # off by one
(3, 6, [(1, 2, 3), (7, None, None)]), # off by two
(3, 7, [(1, 2, 3)]), # step past the end
(7, 8, [(1, 2, 3, 4, 5, 6, 7)]), # step > len(iterable)
]:
actual = list(mi.windowed(iterable, n, step=step))
self.assertEqual(actual, expected)
# Step must be greater than or equal to 1
with self.assertRaises(ValueError):
list(mi.windowed(iterable, 3, step=0))
class BucketTests(TestCase):
"""Tests for ``bucket()``"""
def test_basic(self):
iterable = [10, 20, 30, 11, 21, 31, 12, 22, 23, 33]
D = mi.bucket(iterable, key=lambda x: 10 * (x // 10))
# In-order access
self.assertEqual(list(D[10]), [10, 11, 12])
# Out of order access
self.assertEqual(list(D[30]), [30, 31, 33])
self.assertEqual(list(D[20]), [20, 21, 22, 23])
self.assertEqual(list(D[40]), []) # Nothing in here!
def test_in(self):
iterable = [10, 20, 30, 11, 21, 31, 12, 22, 23, 33]
D = mi.bucket(iterable, key=lambda x: 10 * (x // 10))
self.assertTrue(10 in D)
self.assertFalse(40 in D)
self.assertTrue(20 in D)
self.assertFalse(21 in D)
# Checking in-ness shouldn't advance the iterator
self.assertEqual(next(D[10]), 10)
def test_validator(self):
iterable = count(0)
key = lambda x: int(str(x)[0]) # First digit of each number
validator = lambda x: 0 < x < 10 # No leading zeros
D = mi.bucket(iterable, key, validator=validator)
self.assertEqual(mi.take(3, D[1]), [1, 10, 11])
self.assertNotIn(0, D) # Non-valid entries don't return True
self.assertNotIn(0, D._cache) # Don't store non-valid entries
self.assertEqual(list(D[0]), [])
class SpyTests(TestCase):
"""Tests for ``spy()``"""
def test_basic(self):
original_iterable = iter('abcdefg')
head, new_iterable = mi.spy(original_iterable)
self.assertEqual(head, ['a'])
self.assertEqual(
list(new_iterable), ['a', 'b', 'c', 'd', 'e', 'f', 'g']
)
def test_unpacking(self):
original_iterable = iter('abcdefg')
(first, second, third), new_iterable = mi.spy(original_iterable, 3)
self.assertEqual(first, 'a')
self.assertEqual(second, 'b')
self.assertEqual(third, 'c')
self.assertEqual(
list(new_iterable), ['a', 'b', 'c', 'd', 'e', 'f', 'g']
)
def test_too_many(self):
original_iterable = iter('abc')
head, new_iterable = mi.spy(original_iterable, 4)
self.assertEqual(head, ['a', 'b', 'c'])
self.assertEqual(list(new_iterable), ['a', 'b', 'c'])
def test_zero(self):
original_iterable = iter('abc')
head, new_iterable = mi.spy(original_iterable, 0)
self.assertEqual(head, [])
self.assertEqual(list(new_iterable), ['a', 'b', 'c'])
class InterleaveTests(TestCase):
def test_even(self):
actual = list(mi.interleave([1, 4, 7], [2, 5, 8], [3, 6, 9]))
expected = [1, 2, 3, 4, 5, 6, 7, 8, 9]
self.assertEqual(actual, expected)
def test_short(self):
actual = list(mi.interleave([1, 4], [2, 5, 7], [3, 6, 8]))
expected = [1, 2, 3, 4, 5, 6]
self.assertEqual(actual, expected)
def test_mixed_types(self):
it_list = ['a', 'b', 'c', 'd']
it_str = '12345'
it_inf = count()
actual = list(mi.interleave(it_list, it_str, it_inf))
expected = ['a', '1', 0, 'b', '2', 1, 'c', '3', 2, 'd', '4', 3]
self.assertEqual(actual, expected)
class InterleaveLongestTests(TestCase):
def test_even(self):
actual = list(mi.interleave_longest([1, 4, 7], [2, 5, 8], [3, 6, 9]))
expected = [1, 2, 3, 4, 5, 6, 7, 8, 9]
self.assertEqual(actual, expected)
def test_short(self):
actual = list(mi.interleave_longest([1, 4], [2, 5, 7], [3, 6, 8]))
expected = [1, 2, 3, 4, 5, 6, 7, 8]
self.assertEqual(actual, expected)
def test_mixed_types(self):
it_list = ['a', 'b', 'c', 'd']
it_str = '12345'
it_gen = (x for x in range(3))
actual = list(mi.interleave_longest(it_list, it_str, it_gen))
expected = ['a', '1', 0, 'b', '2', 1, 'c', '3', 2, 'd', '4', '5']
self.assertEqual(actual, expected)
class TestCollapse(TestCase):
"""Tests for ``collapse()``"""
def test_collapse(self):
l = [[1], 2, [[3], 4], [[[5]]]]
self.assertEqual(list(mi.collapse(l)), [1, 2, 3, 4, 5])
def test_collapse_to_string(self):
l = [["s1"], "s2", [["s3"], "s4"], [[["s5"]]]]
self.assertEqual(list(mi.collapse(l)), ["s1", "s2", "s3", "s4", "s5"])
def test_collapse_flatten(self):
l = [[1], [2], [[3], 4], [[[5]]]]
self.assertEqual(list(mi.collapse(l, levels=1)), list(mi.flatten(l)))
def test_collapse_to_level(self):
l = [[1], 2, [[3], 4], [[[5]]]]
self.assertEqual(list(mi.collapse(l, levels=2)), [1, 2, 3, 4, [5]])
self.assertEqual(
list(mi.collapse(mi.collapse(l, levels=1), levels=1)),
list(mi.collapse(l, levels=2))
)
def test_collapse_to_list(self):
l = (1, [2], (3, [4, (5,)], 'ab'))
actual = list(mi.collapse(l, base_type=list))
expected = [1, [2], 3, [4, (5,)], 'ab']
self.assertEqual(actual, expected)
class SideEffectTests(TestCase):
"""Tests for ``side_effect()``"""
def test_individual(self):
# The function increments the counter for each call
counter = [0]
def func(arg):
counter[0] += 1
result = list(mi.side_effect(func, range(10)))
self.assertEqual(result, list(range(10)))
self.assertEqual(counter[0], 10)
def test_chunked(self):
# The function increments the counter for each call
counter = [0]
def func(arg):
counter[0] += 1
result = list(mi.side_effect(func, range(10), 2))
self.assertEqual(result, list(range(10)))
self.assertEqual(counter[0], 5)
def test_before_after(self):
f = StringIO()
collector = []
def func(item):
print(item, file=f)
collector.append(f.getvalue())
def it():
yield u'a'
yield u'b'
raise RuntimeError('kaboom')
before = lambda: print('HEADER', file=f)
after = f.close
try:
mi.consume(mi.side_effect(func, it(), before=before, after=after))
except RuntimeError:
pass
# The iterable should have been written to the file
self.assertEqual(collector, [u'HEADER\na\n', u'HEADER\na\nb\n'])
# The file should be closed even though something bad happened
self.assertTrue(f.closed)
def test_before_fails(self):
f = StringIO()
func = lambda x: print(x, file=f)
def before():
raise RuntimeError('ouch')
try:
mi.consume(
mi.side_effect(func, u'abc', before=before, after=f.close)
)
except RuntimeError:
pass
# The file should be closed even though something bad happened in the
# before function
self.assertTrue(f.closed)
class SlicedTests(TestCase):
"""Tests for ``sliced()``"""
def test_even(self):
"""Test when the length of the sequence is divisible by *n*"""
seq = 'ABCDEFGHI'
self.assertEqual(list(mi.sliced(seq, 3)), ['ABC', 'DEF', 'GHI'])
def test_odd(self):
"""Test when the length of the sequence is not divisible by *n*"""
seq = 'ABCDEFGHI'
self.assertEqual(list(mi.sliced(seq, 4)), ['ABCD', 'EFGH', 'I'])
def test_not_sliceable(self):
seq = (x for x in 'ABCDEFGHI')
with self.assertRaises(TypeError):
list(mi.sliced(seq, 3))
class SplitAtTests(TestCase):
"""Tests for ``split()``"""
def comp_with_str_split(self, str_to_split, delim):
pred = lambda c: c == delim
actual = list(map(''.join, mi.split_at(str_to_split, pred)))
expected = str_to_split.split(delim)
self.assertEqual(actual, expected)
def test_seperators(self):
test_strs = ['', 'abcba', 'aaabbbcccddd', 'e']
for s, delim in product(test_strs, 'abcd'):
self.comp_with_str_split(s, delim)
class SplitBeforeTest(TestCase):
"""Tests for ``split_before()``"""
def test_starts_with_sep(self):
actual = list(mi.split_before('xooxoo', lambda c: c == 'x'))
expected = [['x', 'o', 'o'], ['x', 'o', 'o']]
self.assertEqual(actual, expected)
def test_ends_with_sep(self):
actual = list(mi.split_before('ooxoox', lambda c: c == 'x'))
expected = [['o', 'o'], ['x', 'o', 'o'], ['x']]
self.assertEqual(actual, expected)
def test_no_sep(self):
actual = list(mi.split_before('ooo', lambda c: c == 'x'))
expected = [['o', 'o', 'o']]
self.assertEqual(actual, expected)
class SplitAfterTest(TestCase):
"""Tests for ``split_after()``"""
def test_starts_with_sep(self):
actual = list(mi.split_after('xooxoo', lambda c: c == 'x'))
expected = [['x'], ['o', 'o', 'x'], ['o', 'o']]
self.assertEqual(actual, expected)
def test_ends_with_sep(self):
actual = list(mi.split_after('ooxoox', lambda c: c == 'x'))
expected = [['o', 'o', 'x'], ['o', 'o', 'x']]
self.assertEqual(actual, expected)
def test_no_sep(self):
actual = list(mi.split_after('ooo', lambda c: c == 'x'))
expected = [['o', 'o', 'o']]
self.assertEqual(actual, expected)
class PaddedTest(TestCase):
"""Tests for ``padded()``"""
def test_no_n(self):
seq = [1, 2, 3]
# No fillvalue
self.assertEqual(mi.take(5, mi.padded(seq)), [1, 2, 3, None, None])
# With fillvalue
self.assertEqual(
mi.take(5, mi.padded(seq, fillvalue='')), [1, 2, 3, '', '']
)
def test_invalid_n(self):
self.assertRaises(ValueError, lambda: list(mi.padded([1, 2, 3], n=-1)))
self.assertRaises(ValueError, lambda: list(mi.padded([1, 2, 3], n=0)))
def test_valid_n(self):
seq = [1, 2, 3, 4, 5]
# No need for padding: len(seq) <= n
self.assertEqual(list(mi.padded(seq, n=4)), [1, 2, 3, 4, 5])
self.assertEqual(list(mi.padded(seq, n=5)), [1, 2, 3, 4, 5])
# No fillvalue
self.assertEqual(
list(mi.padded(seq, n=7)), [1, 2, 3, 4, 5, None, None]
)
# With fillvalue
self.assertEqual(
list(mi.padded(seq, fillvalue='', n=7)), [1, 2, 3, 4, 5, '', '']
)
def test_next_multiple(self):
seq = [1, 2, 3, 4, 5, 6]
# No need for padding: len(seq) % n == 0
self.assertEqual(
list(mi.padded(seq, n=3, next_multiple=True)), [1, 2, 3, 4, 5, 6]
)
# Padding needed: len(seq) < n
self.assertEqual(
list(mi.padded(seq, n=8, next_multiple=True)),
[1, 2, 3, 4, 5, 6, None, None]
)
# No padding needed: len(seq) == n
self.assertEqual(
list(mi.padded(seq, n=6, next_multiple=True)), [1, 2, 3, 4, 5, 6]
)
# Padding needed: len(seq) > n
self.assertEqual(
list(mi.padded(seq, n=4, next_multiple=True)),
[1, 2, 3, 4, 5, 6, None, None]
)
# With fillvalue
self.assertEqual(
list(mi.padded(seq, fillvalue='', n=4, next_multiple=True)),
[1, 2, 3, 4, 5, 6, '', '']
)
class DistributeTest(TestCase):
"""Tests for distribute()"""
def test_invalid_n(self):
self.assertRaises(ValueError, lambda: mi.distribute(-1, [1, 2, 3]))
self.assertRaises(ValueError, lambda: mi.distribute(0, [1, 2, 3]))
def test_basic(self):
iterable = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
for n, expected in [
(1, [iterable]),
(2, [[1, 3, 5, 7, 9], [2, 4, 6, 8, 10]]),
(3, [[1, 4, 7, 10], [2, 5, 8], [3, 6, 9]]),
(10, [[n] for n in range(1, 10 + 1)]),
]:
self.assertEqual(
[list(x) for x in mi.distribute(n, iterable)], expected
)
def test_large_n(self):
iterable = [1, 2, 3, 4]
self.assertEqual(
[list(x) for x in mi.distribute(6, iterable)],
[[1], [2], [3], [4], [], []]
)
class StaggerTest(TestCase):
"""Tests for ``stagger()``"""
def test_default(self):
iterable = [0, 1, 2, 3]
actual = list(mi.stagger(iterable))
expected = [(None, 0, 1), (0, 1, 2), (1, 2, 3)]
self.assertEqual(actual, expected)
def test_offsets(self):
iterable = [0, 1, 2, 3]
for offsets, expected in [
((-2, 0, 2), [('', 0, 2), ('', 1, 3)]),
((-2, -1), [('', ''), ('', 0), (0, 1), (1, 2), (2, 3)]),
((1, 2), [(1, 2), (2, 3)]),
]:
all_groups = mi.stagger(iterable, offsets=offsets, fillvalue='')
self.assertEqual(list(all_groups), expected)
def test_longest(self):
iterable = [0, 1, 2, 3]
for offsets, expected in [
(
(-1, 0, 1),
[('', 0, 1), (0, 1, 2), (1, 2, 3), (2, 3, ''), (3, '', '')]
),
((-2, -1), [('', ''), ('', 0), (0, 1), (1, 2), (2, 3), (3, '')]),
((1, 2), [(1, 2), (2, 3), (3, '')]),
]:
all_groups = mi.stagger(
iterable, offsets=offsets, fillvalue='', longest=True
)
self.assertEqual(list(all_groups), expected)
class ZipOffsetTest(TestCase):
"""Tests for ``zip_offset()``"""
def test_shortest(self):
a_1 = [0, 1, 2, 3]
a_2 = [0, 1, 2, 3, 4, 5]
a_3 = [0, 1, 2, 3, 4, 5, 6, 7]
actual = list(
mi.zip_offset(a_1, a_2, a_3, offsets=(-1, 0, 1), fillvalue='')
)
expected = [('', 0, 1), (0, 1, 2), (1, 2, 3), (2, 3, 4), (3, 4, 5)]
self.assertEqual(actual, expected)
def test_longest(self):
a_1 = [0, 1, 2, 3]
a_2 = [0, 1, 2, 3, 4, 5]
a_3 = [0, 1, 2, 3, 4, 5, 6, 7]
actual = list(
mi.zip_offset(a_1, a_2, a_3, offsets=(-1, 0, 1), longest=True)
)
expected = [
(None, 0, 1),
(0, 1, 2),
(1, 2, 3),
(2, 3, 4),
(3, 4, 5),
(None, 5, 6),
(None, None, 7),
]
self.assertEqual(actual, expected)
def test_mismatch(self):
iterables = [0, 1, 2], [2, 3, 4]
offsets = (-1, 0, 1)
self.assertRaises(
ValueError,
lambda: list(mi.zip_offset(*iterables, offsets=offsets))
)
class SortTogetherTest(TestCase):
"""Tests for sort_together()"""
def test_key_list(self):
"""tests `key_list` including default, iterables include duplicates"""
iterables = [
['GA', 'GA', 'GA', 'CT', 'CT', 'CT'],
['May', 'Aug.', 'May', 'June', 'July', 'July'],
[97, 20, 100, 70, 100, 20]
]
self.assertEqual(
mi.sort_together(iterables),
[
('CT', 'CT', 'CT', 'GA', 'GA', 'GA'),
('June', 'July', 'July', 'May', 'Aug.', 'May'),
(70, 100, 20, 97, 20, 100)
]
)
self.assertEqual(
mi.sort_together(iterables, key_list=(0, 1)),
[
('CT', 'CT', 'CT', 'GA', 'GA', 'GA'),
('July', 'July', 'June', 'Aug.', 'May', 'May'),
(100, 20, 70, 20, 97, 100)
]
)
self.assertEqual(
mi.sort_together(iterables, key_list=(0, 1, 2)),
[
('CT', 'CT', 'CT', 'GA', 'GA', 'GA'),
('July', 'July', 'June', 'Aug.', 'May', 'May'),
(20, 100, 70, 20, 97, 100)
]
)
self.assertEqual(
mi.sort_together(iterables, key_list=(2,)),
[
('GA', 'CT', 'CT', 'GA', 'GA', 'CT'),
('Aug.', 'July', 'June', 'May', 'May', 'July'),
(20, 20, 70, 97, 100, 100)
]
)
def test_invalid_key_list(self):
"""tests `key_list` for indexes not available in `iterables`"""
iterables = [
['GA', 'GA', 'GA', 'CT', 'CT', 'CT'],
['May', 'Aug.', 'May', 'June', 'July', 'July'],
[97, 20, 100, 70, 100, 20]
]
self.assertRaises(
IndexError, lambda: mi.sort_together(iterables, key_list=(5,))
)
def test_reverse(self):
"""tests `reverse` to ensure a reverse sort for `key_list` iterables"""
iterables = [
['GA', 'GA', 'GA', 'CT', 'CT', 'CT'],
['May', 'Aug.', 'May', 'June', 'July', 'July'],
[97, 20, 100, 70, 100, 20]
]
self.assertEqual(
mi.sort_together(iterables, key_list=(0, 1, 2), reverse=True),
[('GA', 'GA', 'GA', 'CT', 'CT', 'CT'),
('May', 'May', 'Aug.', 'June', 'July', 'July'),
(100, 97, 20, 70, 100, 20)]
)
def test_uneven_iterables(self):
"""tests trimming of iterables to the shortest length before sorting"""
iterables = [['GA', 'GA', 'GA', 'CT', 'CT', 'CT', 'MA'],
['May', 'Aug.', 'May', 'June', 'July', 'July'],
[97, 20, 100, 70, 100, 20, 0]]
self.assertEqual(
mi.sort_together(iterables),
[
('CT', 'CT', 'CT', 'GA', 'GA', 'GA'),
('June', 'July', 'July', 'May', 'Aug.', 'May'),
(70, 100, 20, 97, 20, 100)
]
)
class DivideTest(TestCase):
"""Tests for divide()"""
def test_invalid_n(self):
self.assertRaises(ValueError, lambda: mi.divide(-1, [1, 2, 3]))
self.assertRaises(ValueError, lambda: mi.divide(0, [1, 2, 3]))
def test_basic(self):
iterable = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
for n, expected in [
(1, [iterable]),
(2, [[1, 2, 3, 4, 5], [6, 7, 8, 9, 10]]),
(3, [[1, 2, 3, 4], [5, 6, 7], [8, 9, 10]]),
(10, [[n] for n in range(1, 10 + 1)]),
]:
self.assertEqual(
[list(x) for x in mi.divide(n, iterable)], expected
)
def test_large_n(self):
iterable = [1, 2, 3, 4]
self.assertEqual(
[list(x) for x in mi.divide(6, iterable)],
[[1], [2], [3], [4], [], []]
)
class TestAlwaysIterable(TestCase):
"""Tests for always_iterable()"""
def test_single(self):
self.assertEqual(list(mi.always_iterable(1)), [1])
def test_strings(self):
for obj in ['foo', b'bar', u'baz']:
actual = list(mi.always_iterable(obj))
expected = [obj]
self.assertEqual(actual, expected)
def test_base_type(self):
dict_obj = {'a': 1, 'b': 2}
str_obj = '123'
# Default: dicts are iterable like they normally are
default_actual = list(mi.always_iterable(dict_obj))
default_expected = list(dict_obj)
self.assertEqual(default_actual, default_expected)
# Unitary types set: dicts are not iterable
custom_actual = list(mi.always_iterable(dict_obj, base_type=dict))
custom_expected = [dict_obj]
self.assertEqual(custom_actual, custom_expected)
# With unitary types set, strings are iterable
str_actual = list(mi.always_iterable(str_obj, base_type=None))
str_expected = list(str_obj)
self.assertEqual(str_actual, str_expected)
def test_iterables(self):
self.assertEqual(list(mi.always_iterable([0, 1])), [0, 1])
self.assertEqual(
list(mi.always_iterable([0, 1], base_type=list)), [[0, 1]]
)
self.assertEqual(
list(mi.always_iterable(iter('foo'))), ['f', 'o', 'o']
)
self.assertEqual(list(mi.always_iterable([])), [])
def test_none(self):
self.assertEqual(list(mi.always_iterable(None)), [])
def test_generator(self):
def _gen():
yield 0
yield 1
self.assertEqual(list(mi.always_iterable(_gen())), [0, 1])
class AdjacentTests(TestCase):
def test_typical(self):
actual = list(mi.adjacent(lambda x: x % 5 == 0, range(10)))
expected = [(True, 0), (True, 1), (False, 2), (False, 3), (True, 4),
(True, 5), (True, 6), (False, 7), (False, 8), (False, 9)]
self.assertEqual(actual, expected)
def test_empty_iterable(self):
actual = list(mi.adjacent(lambda x: x % 5 == 0, []))
expected = []
self.assertEqual(actual, expected)
def test_length_one(self):
actual = list(mi.adjacent(lambda x: x % 5 == 0, [0]))
expected = [(True, 0)]
self.assertEqual(actual, expected)
actual = list(mi.adjacent(lambda x: x % 5 == 0, [1]))
expected = [(False, 1)]
self.assertEqual(actual, expected)
def test_consecutive_true(self):
"""Test that when the predicate matches multiple consecutive elements
it doesn't repeat elements in the output"""
actual = list(mi.adjacent(lambda x: x % 5 < 2, range(10)))
expected = [(True, 0), (True, 1), (True, 2), (False, 3), (True, 4),
(True, 5), (True, 6), (True, 7), (False, 8), (False, 9)]
self.assertEqual(actual, expected)
def test_distance(self):
actual = list(mi.adjacent(lambda x: x % 5 == 0, range(10), distance=2))
expected = [(True, 0), (True, 1), (True, 2), (True, 3), (True, 4),
(True, 5), (True, 6), (True, 7), (False, 8), (False, 9)]
self.assertEqual(actual, expected)
actual = list(mi.adjacent(lambda x: x % 5 == 0, range(10), distance=3))
expected = [(True, 0), (True, 1), (True, 2), (True, 3), (True, 4),
(True, 5), (True, 6), (True, 7), (True, 8), (False, 9)]
self.assertEqual(actual, expected)
def test_large_distance(self):
"""Test distance larger than the length of the iterable"""
iterable = range(10)
actual = list(mi.adjacent(lambda x: x % 5 == 4, iterable, distance=20))
expected = list(zip(repeat(True), iterable))
self.assertEqual(actual, expected)
actual = list(mi.adjacent(lambda x: False, iterable, distance=20))
expected = list(zip(repeat(False), iterable))
self.assertEqual(actual, expected)
def test_zero_distance(self):
"""Test that adjacent() reduces to zip+map when distance is 0"""
iterable = range(1000)
predicate = lambda x: x % 4 == 2
actual = mi.adjacent(predicate, iterable, 0)
expected = zip(map(predicate, iterable), iterable)
self.assertTrue(all(a == e for a, e in zip(actual, expected)))
def test_negative_distance(self):
"""Test that adjacent() raises an error with negative distance"""
pred = lambda x: x
self.assertRaises(
ValueError, lambda: mi.adjacent(pred, range(1000), -1)
)
self.assertRaises(
ValueError, lambda: mi.adjacent(pred, range(10), -10)
)
def test_grouping(self):
"""Test interaction of adjacent() with groupby_transform()"""
iterable = mi.adjacent(lambda x: x % 5 == 0, range(10))
grouper = mi.groupby_transform(iterable, itemgetter(0), itemgetter(1))
actual = [(k, list(g)) for k, g in grouper]
expected = [
(True, [0, 1]),
(False, [2, 3]),
(True, [4, 5, 6]),
(False, [7, 8, 9]),
]
self.assertEqual(actual, expected)
def test_call_once(self):
"""Test that the predicate is only called once per item."""
already_seen = set()
iterable = range(10)
def predicate(item):
self.assertNotIn(item, already_seen)
already_seen.add(item)
return True
actual = list(mi.adjacent(predicate, iterable))
expected = [(True, x) for x in iterable]
self.assertEqual(actual, expected)
class GroupByTransformTests(TestCase):
def assertAllGroupsEqual(self, groupby1, groupby2):
"""Compare two groupby objects for equality, both keys and groups."""
for a, b in zip(groupby1, groupby2):
key1, group1 = a
key2, group2 = b
self.assertEqual(key1, key2)
self.assertListEqual(list(group1), list(group2))
self.assertRaises(StopIteration, lambda: next(groupby1))
self.assertRaises(StopIteration, lambda: next(groupby2))
def test_default_funcs(self):
"""Test that groupby_transform() with default args mimics groupby()"""
iterable = [(x // 5, x) for x in range(1000)]
actual = mi.groupby_transform(iterable)
expected = groupby(iterable)
self.assertAllGroupsEqual(actual, expected)
def test_valuefunc(self):
iterable = [(int(x / 5), int(x / 3), x) for x in range(10)]
# Test the standard usage of grouping one iterable using another's keys
grouper = mi.groupby_transform(
iterable, keyfunc=itemgetter(0), valuefunc=itemgetter(-1)
)
actual = [(k, list(g)) for k, g in grouper]
expected = [(0, [0, 1, 2, 3, 4]), (1, [5, 6, 7, 8, 9])]
self.assertEqual(actual, expected)
grouper = mi.groupby_transform(
iterable, keyfunc=itemgetter(1), valuefunc=itemgetter(-1)
)
actual = [(k, list(g)) for k, g in grouper]
expected = [(0, [0, 1, 2]), (1, [3, 4, 5]), (2, [6, 7, 8]), (3, [9])]
self.assertEqual(actual, expected)
# and now for something a little different
d = dict(zip(range(10), 'abcdefghij'))
grouper = mi.groupby_transform(
range(10), keyfunc=lambda x: x // 5, valuefunc=d.get
)
actual = [(k, ''.join(g)) for k, g in grouper]
expected = [(0, 'abcde'), (1, 'fghij')]
self.assertEqual(actual, expected)
def test_no_valuefunc(self):
iterable = range(1000)
def key(x):
return x // 5
actual = mi.groupby_transform(iterable, key, valuefunc=None)
expected = groupby(iterable, key)
self.assertAllGroupsEqual(actual, expected)
actual = mi.groupby_transform(iterable, key) # default valuefunc
expected = groupby(iterable, key)
self.assertAllGroupsEqual(actual, expected)
class NumericRangeTests(TestCase):
def test_basic(self):
for args, expected in [
((4,), [0, 1, 2, 3]),
((4.0,), [0.0, 1.0, 2.0, 3.0]),
((1.0, 4), [1.0, 2.0, 3.0]),
((1, 4.0), [1, 2, 3]),
((1.0, 5), [1.0, 2.0, 3.0, 4.0]),
((0, 20, 5), [0, 5, 10, 15]),
((0, 20, 5.0), [0.0, 5.0, 10.0, 15.0]),
((0, 10, 3), [0, 3, 6, 9]),
((0, 10, 3.0), [0.0, 3.0, 6.0, 9.0]),
((0, -5, -1), [0, -1, -2, -3, -4]),
((0.0, -5, -1), [0.0, -1.0, -2.0, -3.0, -4.0]),
((1, 2, Fraction(1, 2)), [Fraction(1, 1), Fraction(3, 2)]),
((0,), []),
((0.0,), []),
((1, 0), []),
((1.0, 0.0), []),
((Fraction(2, 1),), [Fraction(0, 1), Fraction(1, 1)]),
((Decimal('2.0'),), [Decimal('0.0'), Decimal('1.0')]),
]:
actual = list(mi.numeric_range(*args))
self.assertEqual(actual, expected)
self.assertTrue(
all(type(a) == type(e) for a, e in zip(actual, expected))
)
def test_arg_count(self):
self.assertRaises(TypeError, lambda: list(mi.numeric_range()))
self.assertRaises(
TypeError, lambda: list(mi.numeric_range(0, 1, 2, 3))
)
def test_zero_step(self):
self.assertRaises(
ValueError, lambda: list(mi.numeric_range(1, 2, 0))
)
class CountCycleTests(TestCase):
def test_basic(self):
expected = [
(0, 'a'), (0, 'b'), (0, 'c'),
(1, 'a'), (1, 'b'), (1, 'c'),
(2, 'a'), (2, 'b'), (2, 'c'),
]
for actual in [
mi.take(9, mi.count_cycle('abc')), # n=None
list(mi.count_cycle('abc', 3)), # n=3
]:
self.assertEqual(actual, expected)
def test_empty(self):
self.assertEqual(list(mi.count_cycle('')), [])
self.assertEqual(list(mi.count_cycle('', 2)), [])
def test_negative(self):
self.assertEqual(list(mi.count_cycle('abc', -3)), [])
class LocateTests(TestCase):
def test_default_pred(self):
iterable = [0, 1, 1, 0, 1, 0, 0]
actual = list(mi.locate(iterable))
expected = [1, 2, 4]
self.assertEqual(actual, expected)
def test_no_matches(self):
iterable = [0, 0, 0]
actual = list(mi.locate(iterable))
expected = []
self.assertEqual(actual, expected)
def test_custom_pred(self):
iterable = ['0', 1, 1, '0', 1, '0', '0']
pred = lambda x: x == '0'
actual = list(mi.locate(iterable, pred))
expected = [0, 3, 5, 6]
self.assertEqual(actual, expected)
class StripFunctionTests(TestCase):
def test_hashable(self):
iterable = list('www.example.com')
pred = lambda x: x in set('cmowz.')
self.assertEqual(list(mi.lstrip(iterable, pred)), list('example.com'))
self.assertEqual(list(mi.rstrip(iterable, pred)), list('www.example'))
self.assertEqual(list(mi.strip(iterable, pred)), list('example'))
def test_not_hashable(self):
iterable = [
list('http://'), list('www'), list('.example'), list('.com')
]
pred = lambda x: x in [list('http://'), list('www'), list('.com')]
self.assertEqual(list(mi.lstrip(iterable, pred)), iterable[2:])
self.assertEqual(list(mi.rstrip(iterable, pred)), iterable[:3])
self.assertEqual(list(mi.strip(iterable, pred)), iterable[2: 3])
def test_math(self):
iterable = [0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2]
pred = lambda x: x <= 2
self.assertEqual(list(mi.lstrip(iterable, pred)), iterable[3:])
self.assertEqual(list(mi.rstrip(iterable, pred)), iterable[:-3])
self.assertEqual(list(mi.strip(iterable, pred)), iterable[3:-3])
class IsliceExtendedTests(TestCase):
def test_all(self):
iterable = ['0', '1', '2', '3', '4', '5']
indexes = list(range(-4, len(iterable) + 4)) + [None]
steps = [1, 2, 3, 4, -1, -2, -3, 4]
for slice_args in product(indexes, indexes, steps):
try:
actual = list(mi.islice_extended(iterable, *slice_args))
except Exception as e:
self.fail((slice_args, e))
expected = iterable[slice(*slice_args)]
self.assertEqual(actual, expected, slice_args)
def test_zero_step(self):
with self.assertRaises(ValueError):
list(mi.islice_extended([1, 2, 3], 0, 1, 0))
class ConsecutiveGroupsTest(TestCase):
def test_numbers(self):
iterable = [-10, -8, -7, -6, 1, 2, 4, 5, -1, 7]
actual = [list(g) for g in mi.consecutive_groups(iterable)]
expected = [[-10], [-8, -7, -6], [1, 2], [4, 5], [-1], [7]]
self.assertEqual(actual, expected)
def test_custom_ordering(self):
iterable = ['1', '10', '11', '20', '21', '22', '30', '31']
ordering = lambda x: int(x)
actual = [list(g) for g in mi.consecutive_groups(iterable, ordering)]
expected = [['1'], ['10', '11'], ['20', '21', '22'], ['30', '31']]
self.assertEqual(actual, expected)
def test_exotic_ordering(self):
iterable = [
('a', 'b', 'c', 'd'),
('a', 'c', 'b', 'd'),
('a', 'c', 'd', 'b'),
('a', 'd', 'b', 'c'),
('d', 'b', 'c', 'a'),
('d', 'c', 'a', 'b'),
]
ordering = list(permutations('abcd')).index
actual = [list(g) for g in mi.consecutive_groups(iterable, ordering)]
expected = [
[('a', 'b', 'c', 'd')],
[('a', 'c', 'b', 'd'), ('a', 'c', 'd', 'b'), ('a', 'd', 'b', 'c')],
[('d', 'b', 'c', 'a'), ('d', 'c', 'a', 'b')],
]
self.assertEqual(actual, expected)
class DifferenceTest(TestCase):
def test_normal(self):
iterable = [10, 20, 30, 40, 50]
actual = list(mi.difference(iterable))
expected = [10, 10, 10, 10, 10]
self.assertEqual(actual, expected)
def test_custom(self):
iterable = [10, 20, 30, 40, 50]
actual = list(mi.difference(iterable, add))
expected = [10, 30, 50, 70, 90]
self.assertEqual(actual, expected)
def test_roundtrip(self):
original = list(range(100))
accumulated = mi.accumulate(original)
actual = list(mi.difference(accumulated))
self.assertEqual(actual, original)
def test_one(self):
self.assertEqual(list(mi.difference([0])), [0])
def test_empty(self):
self.assertEqual(list(mi.difference([])), [])
class SeekableTest(TestCase):
def test_exhaustion_reset(self):
iterable = [str(n) for n in range(10)]
s = mi.seekable(iterable)
self.assertEqual(list(s), iterable) # Normal iteration
self.assertEqual(list(s), []) # Iterable is exhausted
s.seek(0)
self.assertEqual(list(s), iterable) # Back in action
def test_partial_reset(self):
iterable = [str(n) for n in range(10)]
s = mi.seekable(iterable)
self.assertEqual(mi.take(5, s), iterable[:5]) # Normal iteration
s.seek(1)
self.assertEqual(list(s), iterable[1:]) # Get the rest of the iterable
def test_forward(self):
iterable = [str(n) for n in range(10)]
s = mi.seekable(iterable)
self.assertEqual(mi.take(1, s), iterable[:1]) # Normal iteration
s.seek(3) # Skip over index 2
self.assertEqual(list(s), iterable[3:]) # Result is similar to slicing
s.seek(0) # Back to 0
self.assertEqual(list(s), iterable) # No difference in result
def test_past_end(self):
iterable = [str(n) for n in range(10)]
s = mi.seekable(iterable)
self.assertEqual(mi.take(1, s), iterable[:1]) # Normal iteration
s.seek(20)
self.assertEqual(list(s), []) # Iterable is exhausted
s.seek(0) # Back to 0
self.assertEqual(list(s), iterable) # No difference in result
def test_elements(self):
iterable = map(str, count())
s = mi.seekable(iterable)
mi.take(10, s)
elements = s.elements()
self.assertEqual(
[elements[i] for i in range(10)], [str(n) for n in range(10)]
)
self.assertEqual(len(elements), 10)
mi.take(10, s)
self.assertEqual(list(elements), [str(n) for n in range(20)])
class SequenceViewTests(TestCase):
def test_init(self):
view = mi.SequenceView((1, 2, 3))
self.assertEqual(repr(view), "SequenceView((1, 2, 3))")
self.assertRaises(TypeError, lambda: mi.SequenceView({}))
def test_update(self):
seq = [1, 2, 3]
view = mi.SequenceView(seq)
self.assertEqual(len(view), 3)
self.assertEqual(repr(view), "SequenceView([1, 2, 3])")
seq.pop()
self.assertEqual(len(view), 2)
self.assertEqual(repr(view), "SequenceView([1, 2])")
def test_indexing(self):
seq = ('a', 'b', 'c', 'd', 'e', 'f')
view = mi.SequenceView(seq)
for i in range(-len(seq), len(seq)):
self.assertEqual(view[i], seq[i])
def test_slicing(self):
seq = ('a', 'b', 'c', 'd', 'e', 'f')
view = mi.SequenceView(seq)
n = len(seq)
indexes = list(range(-n - 1, n + 1)) + [None]
steps = list(range(-n, n + 1))
steps.remove(0)
for slice_args in product(indexes, indexes, steps):
i = slice(*slice_args)
self.assertEqual(view[i], seq[i])
def test_abc_methods(self):
# collections.Sequence should provide all of this functionality
seq = ('a', 'b', 'c', 'd', 'e', 'f', 'f')
view = mi.SequenceView(seq)
# __contains__
self.assertIn('b', view)
self.assertNotIn('g', view)
# __iter__
self.assertEqual(list(iter(view)), list(seq))
# __reversed__
self.assertEqual(list(reversed(view)), list(reversed(seq)))
# index
self.assertEqual(view.index('b'), 1)
# count
self.assertEqual(seq.count('f'), 2)
class RunLengthTest(TestCase):
def test_encode(self):
iterable = (int(str(n)[0]) for n in count(800))
actual = mi.take(4, mi.run_length.encode(iterable))
expected = [(8, 100), (9, 100), (1, 1000), (2, 1000)]
self.assertEqual(actual, expected)
def test_decode(self):
iterable = [('d', 4), ('c', 3), ('b', 2), ('a', 1)]
actual = ''.join(mi.run_length.decode(iterable))
expected = 'ddddcccbba'
self.assertEqual(actual, expected)
class ExactlyNTests(TestCase):
"""Tests for ``exactly_n()``"""
def test_true(self):
"""Iterable has ``n`` ``True`` elements"""
self.assertTrue(mi.exactly_n([True, False, True], 2))
self.assertTrue(mi.exactly_n([1, 1, 1, 0], 3))
self.assertTrue(mi.exactly_n([False, False], 0))
self.assertTrue(mi.exactly_n(range(100), 10, lambda x: x < 10))
def test_false(self):
"""Iterable does not have ``n`` ``True`` elements"""
self.assertFalse(mi.exactly_n([True, False, False], 2))
self.assertFalse(mi.exactly_n([True, True, False], 1))
self.assertFalse(mi.exactly_n([False], 1))
self.assertFalse(mi.exactly_n([True], -1))
self.assertFalse(mi.exactly_n(repeat(True), 100))
def test_empty(self):
"""Return ``True`` if the iterable is empty and ``n`` is 0"""
self.assertTrue(mi.exactly_n([], 0))
self.assertFalse(mi.exactly_n([], 1))
class AlwaysReversibleTests(TestCase):
"""Tests for ``always_reversible()``"""
def test_regular_reversed(self):
self.assertEqual(list(reversed(range(10))),
list(mi.always_reversible(range(10))))
self.assertEqual(list(reversed([1, 2, 3])),
list(mi.always_reversible([1, 2, 3])))
self.assertEqual(reversed([1, 2, 3]).__class__,
mi.always_reversible([1, 2, 3]).__class__)
def test_nonseq_reversed(self):
# Create a non-reversible generator from a sequence
with self.assertRaises(TypeError):
reversed(x for x in range(10))
self.assertEqual(list(reversed(range(10))),
list(mi.always_reversible(x for x in range(10))))
self.assertEqual(list(reversed([1, 2, 3])),
list(mi.always_reversible(x for x in [1, 2, 3])))
self.assertNotEqual(reversed((1, 2)).__class__,
mi.always_reversible(x for x in (1, 2)).__class__)
class CircularShiftsTests(TestCase):
def test_empty(self):
# empty iterable -> empty list
self.assertEqual(list(mi.circular_shifts([])), [])
def test_simple_circular_shifts(self):
# test the a simple iterator case
self.assertEqual(
mi.circular_shifts(range(4)),
[(0, 1, 2, 3), (1, 2, 3, 0), (2, 3, 0, 1), (3, 0, 1, 2)]
)
def test_duplicates(self):
# test non-distinct entries
self.assertEqual(
mi.circular_shifts([0, 1, 0, 1]),
[(0, 1, 0, 1), (1, 0, 1, 0), (0, 1, 0, 1), (1, 0, 1, 0)]
)
class MakeDecoratorTests(TestCase):
def test_basic(self):
slicer = mi.make_decorator(islice)
@slicer(1, 10, 2)
def user_function(arg_1, arg_2, kwarg_1=None):
self.assertEqual(arg_1, 'arg_1')
self.assertEqual(arg_2, 'arg_2')
self.assertEqual(kwarg_1, 'kwarg_1')
return map(str, count())
it = user_function('arg_1', 'arg_2', kwarg_1='kwarg_1')
actual = list(it)
expected = ['1', '3', '5', '7', '9']
self.assertEqual(actual, expected)
def test_result_index(self):
def stringify(*args, **kwargs):
self.assertEqual(args[0], 'arg_0')
iterable = args[1]
self.assertEqual(args[2], 'arg_2')
self.assertEqual(kwargs['kwarg_1'], 'kwarg_1')
return map(str, iterable)
stringifier = mi.make_decorator(stringify, result_index=1)
@stringifier('arg_0', 'arg_2', kwarg_1='kwarg_1')
def user_function(n):
return count(n)
it = user_function(1)
actual = mi.take(5, it)
expected = ['1', '2', '3', '4', '5']
self.assertEqual(actual, expected)
def test_wrap_class(self):
seeker = mi.make_decorator(mi.seekable)
@seeker()
def user_function(n):
return map(str, range(n))
it = user_function(5)
self.assertEqual(list(it), ['0', '1', '2', '3', '4'])
it.seek(0)
self.assertEqual(list(it), ['0', '1', '2', '3', '4'])
class MapReduceTests(TestCase):
def test_default(self):
iterable = (str(x) for x in range(5))
keyfunc = lambda x: int(x) // 2
actual = sorted(mi.map_reduce(iterable, keyfunc).items())
expected = [(0, ['0', '1']), (1, ['2', '3']), (2, ['4'])]
self.assertEqual(actual, expected)
def test_valuefunc(self):
iterable = (str(x) for x in range(5))
keyfunc = lambda x: int(x) // 2
valuefunc = int
actual = sorted(mi.map_reduce(iterable, keyfunc, valuefunc).items())
expected = [(0, [0, 1]), (1, [2, 3]), (2, [4])]
self.assertEqual(actual, expected)
def test_reducefunc(self):
iterable = (str(x) for x in range(5))
keyfunc = lambda x: int(x) // 2
valuefunc = int
reducefunc = lambda value_list: reduce(mul, value_list, 1)
actual = sorted(
mi.map_reduce(iterable, keyfunc, valuefunc, reducefunc).items()
)
expected = [(0, 0), (1, 6), (2, 4)]
self.assertEqual(actual, expected)
def test_ret(self):
d = mi.map_reduce([1, 0, 2, 0, 1, 0], bool)
self.assertEqual(d, {False: [0, 0, 0], True: [1, 2, 1]})
self.assertRaises(KeyError, lambda: d[None].append(1))
AnonSec - 2021 | Recode By D7net