summaryrefslogtreecommitdiff
path: root/Lib/test/test_pprint.py
diff options
context:
space:
mode:
Diffstat (limited to 'Lib/test/test_pprint.py')
-rw-r--r--Lib/test/test_pprint.py456
1 files changed, 432 insertions, 24 deletions
diff --git a/Lib/test/test_pprint.py b/Lib/test/test_pprint.py
index 180ddb0f82..357c5cf0a9 100644
--- a/Lib/test/test_pprint.py
+++ b/Lib/test/test_pprint.py
@@ -1,12 +1,14 @@
# -*- coding: utf-8 -*-
+import collections
+import io
+import itertools
import pprint
+import random
import test.support
-import unittest
import test.test_set
-import random
-import collections
-import itertools
+import types
+import unittest
# list, tuple and dict subclasses that do or don't overwrite __repr__
class list2(list):
@@ -48,6 +50,25 @@ class Unorderable:
def __repr__(self):
return str(id(self))
+# Class Orderable is orderable with any type
+class Orderable:
+ def __init__(self, hash):
+ self._hash = hash
+ def __lt__(self, other):
+ return False
+ def __gt__(self, other):
+ return self != other
+ def __le__(self, other):
+ return self == other
+ def __ge__(self, other):
+ return True
+ def __eq__(self, other):
+ return self is other
+ def __ne__(self, other):
+ return self is not other
+ def __hash__(self):
+ return self._hash
+
class QueryTestCase(unittest.TestCase):
def setUp(self):
@@ -55,6 +76,18 @@ class QueryTestCase(unittest.TestCase):
self.b = list(range(200))
self.a[-12] = self.b
+ def test_init(self):
+ pp = pprint.PrettyPrinter()
+ pp = pprint.PrettyPrinter(indent=4, width=40, depth=5,
+ stream=io.StringIO(), compact=True)
+ pp = pprint.PrettyPrinter(4, 40, 5, io.StringIO())
+ with self.assertRaises(TypeError):
+ pp = pprint.PrettyPrinter(4, 40, 5, io.StringIO(), True)
+ self.assertRaises(ValueError, pprint.PrettyPrinter, indent=-1)
+ self.assertRaises(ValueError, pprint.PrettyPrinter, depth=0)
+ self.assertRaises(ValueError, pprint.PrettyPrinter, depth=-1)
+ self.assertRaises(ValueError, pprint.PrettyPrinter, width=0)
+
def test_basic(self):
# Verify .isrecursive() and .isreadable() w/o recursion
pp = pprint.PrettyPrinter()
@@ -195,10 +228,52 @@ class QueryTestCase(unittest.TestCase):
o = [o1, o2]
expected = """\
[ [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
+ {'first': 1, 'second': 2, 'third': 3}]"""
+ self.assertEqual(pprint.pformat(o, indent=4, width=42), expected)
+ expected = """\
+[ [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
{ 'first': 1,
'second': 2,
'third': 3}]"""
- self.assertEqual(pprint.pformat(o, indent=4, width=42), expected)
+ self.assertEqual(pprint.pformat(o, indent=4, width=41), expected)
+
+ def test_width(self):
+ expected = """\
+[[[[[[1, 2, 3],
+ '1 2']]]],
+ {1: [1, 2, 3],
+ 2: [12, 34]},
+ 'abc def ghi',
+ ('ab cd ef',),
+ set2({1, 23}),
+ [[[[[1, 2, 3],
+ '1 2']]]]]"""
+ o = eval(expected)
+ self.assertEqual(pprint.pformat(o, width=15), expected)
+ self.assertEqual(pprint.pformat(o, width=16), expected)
+ self.assertEqual(pprint.pformat(o, width=25), expected)
+ self.assertEqual(pprint.pformat(o, width=14), """\
+[[[[[[1,
+ 2,
+ 3],
+ '1 '
+ '2']]]],
+ {1: [1,
+ 2,
+ 3],
+ 2: [12,
+ 34]},
+ 'abc def '
+ 'ghi',
+ ('ab cd '
+ 'ef',),
+ set2({1,
+ 23}),
+ [[[[[1,
+ 2,
+ 3],
+ '1 '
+ '2']]]]]""")
def test_sorted_dict(self):
# Starting in Python 2.5, pprint sorts dict displays by key regardless
@@ -219,19 +294,51 @@ class QueryTestCase(unittest.TestCase):
r"{5: [[]], 'xy\tab\n': (3,), (): {}}")
def test_ordered_dict(self):
+ d = collections.OrderedDict()
+ self.assertEqual(pprint.pformat(d, width=1), 'OrderedDict()')
+ d = collections.OrderedDict([])
+ self.assertEqual(pprint.pformat(d, width=1), 'OrderedDict()')
words = 'the quick brown fox jumped over a lazy dog'.split()
d = collections.OrderedDict(zip(words, itertools.count()))
self.assertEqual(pprint.pformat(d),
"""\
-{'the': 0,
- 'quick': 1,
- 'brown': 2,
- 'fox': 3,
- 'jumped': 4,
- 'over': 5,
- 'a': 6,
- 'lazy': 7,
- 'dog': 8}""")
+OrderedDict([('the', 0),
+ ('quick', 1),
+ ('brown', 2),
+ ('fox', 3),
+ ('jumped', 4),
+ ('over', 5),
+ ('a', 6),
+ ('lazy', 7),
+ ('dog', 8)])""")
+
+ def test_mapping_proxy(self):
+ words = 'the quick brown fox jumped over a lazy dog'.split()
+ d = dict(zip(words, itertools.count()))
+ m = types.MappingProxyType(d)
+ self.assertEqual(pprint.pformat(m), """\
+mappingproxy({'a': 6,
+ 'brown': 2,
+ 'dog': 8,
+ 'fox': 3,
+ 'jumped': 4,
+ 'lazy': 7,
+ 'over': 5,
+ 'quick': 1,
+ 'the': 0})""")
+ d = collections.OrderedDict(zip(words, itertools.count()))
+ m = types.MappingProxyType(d)
+ self.assertEqual(pprint.pformat(m), """\
+mappingproxy(OrderedDict([('the', 0),
+ ('quick', 1),
+ ('brown', 2),
+ ('fox', 3),
+ ('jumped', 4),
+ ('over', 5),
+ ('a', 6),
+ ('lazy', 7),
+ ('dog', 8)]))""")
+
def test_subclassing(self):
o = {'names with spaces': 'should be presented using repr()',
'others.should.not.be': 'like.this'}
@@ -535,16 +642,35 @@ frozenset2({0,
self.assertEqual(pprint.pformat(dict.fromkeys(keys, 0)),
'{%r: 0, %r: 0}' % tuple(sorted(keys, key=id)))
+ def test_sort_orderable_and_unorderable_values(self):
+ # Issue 22721: sorted pprints is not stable
+ a = Unorderable()
+ b = Orderable(hash(a)) # should have the same hash value
+ # self-test
+ self.assertLess(a, b)
+ self.assertLess(str(type(b)), str(type(a)))
+ self.assertEqual(sorted([b, a]), [a, b])
+ self.assertEqual(sorted([a, b]), [a, b])
+ # set
+ self.assertEqual(pprint.pformat(set([b, a]), width=1),
+ '{%r,\n %r}' % (a, b))
+ self.assertEqual(pprint.pformat(set([a, b]), width=1),
+ '{%r,\n %r}' % (a, b))
+ # dict
+ self.assertEqual(pprint.pformat(dict.fromkeys([b, a]), width=1),
+ '{%r: None,\n %r: None}' % (a, b))
+ self.assertEqual(pprint.pformat(dict.fromkeys([a, b]), width=1),
+ '{%r: None,\n %r: None}' % (a, b))
+
def test_str_wrap(self):
# pprint tries to wrap strings intelligently
fox = 'the quick brown fox jumped over a lazy dog'
- self.assertEqual(pprint.pformat(fox, width=20), """\
-('the quick '
- 'brown fox '
- 'jumped over a '
- 'lazy dog')""")
+ self.assertEqual(pprint.pformat(fox, width=19), """\
+('the quick brown '
+ 'fox jumped over '
+ 'a lazy dog')""")
self.assertEqual(pprint.pformat({'a': 1, 'b': fox, 'c': 2},
- width=26), """\
+ width=25), """\
{'a': 1,
'b': 'the quick brown '
'fox jumped over '
@@ -556,12 +682,34 @@ frozenset2({0,
# - non-ASCII is allowed
# - an apostrophe doesn't disrupt the pprint
special = "Portons dix bons \"whiskys\"\nà l'avocat goujat\t qui fumait au zoo"
- self.assertEqual(pprint.pformat(special, width=21), """\
-('Portons dix '
- 'bons "whiskys"\\n'
+ self.assertEqual(pprint.pformat(special, width=68), repr(special))
+ self.assertEqual(pprint.pformat(special, width=31), """\
+('Portons dix bons "whiskys"\\n'
+ "à l'avocat goujat\\t qui "
+ 'fumait au zoo')""")
+ self.assertEqual(pprint.pformat(special, width=20), """\
+('Portons dix bons '
+ '"whiskys"\\n'
"à l'avocat "
'goujat\\t qui '
'fumait au zoo')""")
+ self.assertEqual(pprint.pformat([[[[[special]]]]], width=35), """\
+[[[[['Portons dix bons "whiskys"\\n'
+ "à l'avocat goujat\\t qui "
+ 'fumait au zoo']]]]]""")
+ self.assertEqual(pprint.pformat([[[[[special]]]]], width=25), """\
+[[[[['Portons dix bons '
+ '"whiskys"\\n'
+ "à l'avocat "
+ 'goujat\\t qui '
+ 'fumait au zoo']]]]]""")
+ self.assertEqual(pprint.pformat([[[[[special]]]]], width=23), """\
+[[[[['Portons dix '
+ 'bons "whiskys"\\n'
+ "à l'avocat "
+ 'goujat\\t qui '
+ 'fumait au '
+ 'zoo']]]]]""")
# An unwrappable string is formatted as its repr
unwrappable = "x" * 100
self.assertEqual(pprint.pformat(unwrappable, width=80), repr(unwrappable))
@@ -584,7 +732,267 @@ frozenset2({0,
14, 15],
[], [0], [0, 1], [0, 1, 2], [0, 1, 2, 3],
[0, 1, 2, 3, 4]]"""
- self.assertEqual(pprint.pformat(o, width=48, compact=True), expected)
+ self.assertEqual(pprint.pformat(o, width=47, compact=True), expected)
+
+ def test_compact_width(self):
+ levels = 20
+ number = 10
+ o = [0] * number
+ for i in range(levels - 1):
+ o = [o]
+ for w in range(levels * 2 + 1, levels + 3 * number - 1):
+ lines = pprint.pformat(o, width=w, compact=True).splitlines()
+ maxwidth = max(map(len, lines))
+ self.assertLessEqual(maxwidth, w)
+ self.assertGreater(maxwidth, w - 3)
+
+ def test_bytes_wrap(self):
+ self.assertEqual(pprint.pformat(b'', width=1), "b''")
+ self.assertEqual(pprint.pformat(b'abcd', width=1), "b'abcd'")
+ letters = b'abcdefghijklmnopqrstuvwxyz'
+ self.assertEqual(pprint.pformat(letters, width=29), repr(letters))
+ self.assertEqual(pprint.pformat(letters, width=19), """\
+(b'abcdefghijkl'
+ b'mnopqrstuvwxyz')""")
+ self.assertEqual(pprint.pformat(letters, width=18), """\
+(b'abcdefghijkl'
+ b'mnopqrstuvwx'
+ b'yz')""")
+ self.assertEqual(pprint.pformat(letters, width=16), """\
+(b'abcdefghijkl'
+ b'mnopqrstuvwx'
+ b'yz')""")
+ special = bytes(range(16))
+ self.assertEqual(pprint.pformat(special, width=61), repr(special))
+ self.assertEqual(pprint.pformat(special, width=48), """\
+(b'\\x00\\x01\\x02\\x03\\x04\\x05\\x06\\x07\\x08\\t\\n\\x0b'
+ b'\\x0c\\r\\x0e\\x0f')""")
+ self.assertEqual(pprint.pformat(special, width=32), """\
+(b'\\x00\\x01\\x02\\x03'
+ b'\\x04\\x05\\x06\\x07\\x08\\t\\n\\x0b'
+ b'\\x0c\\r\\x0e\\x0f')""")
+ self.assertEqual(pprint.pformat(special, width=1), """\
+(b'\\x00\\x01\\x02\\x03'
+ b'\\x04\\x05\\x06\\x07'
+ b'\\x08\\t\\n\\x0b'
+ b'\\x0c\\r\\x0e\\x0f')""")
+ self.assertEqual(pprint.pformat({'a': 1, 'b': letters, 'c': 2},
+ width=21), """\
+{'a': 1,
+ 'b': b'abcdefghijkl'
+ b'mnopqrstuvwx'
+ b'yz',
+ 'c': 2}""")
+ self.assertEqual(pprint.pformat({'a': 1, 'b': letters, 'c': 2},
+ width=20), """\
+{'a': 1,
+ 'b': b'abcdefgh'
+ b'ijklmnop'
+ b'qrstuvwxyz',
+ 'c': 2}""")
+ self.assertEqual(pprint.pformat([[[[[[letters]]]]]], width=25), """\
+[[[[[[b'abcdefghijklmnop'
+ b'qrstuvwxyz']]]]]]""")
+ self.assertEqual(pprint.pformat([[[[[[special]]]]]], width=41), """\
+[[[[[[b'\\x00\\x01\\x02\\x03\\x04\\x05\\x06\\x07'
+ b'\\x08\\t\\n\\x0b\\x0c\\r\\x0e\\x0f']]]]]]""")
+ # Check that the pprint is a usable repr
+ for width in range(1, 64):
+ formatted = pprint.pformat(special, width=width)
+ self.assertEqual(eval(formatted), special)
+ formatted = pprint.pformat([special] * 2, width=width)
+ self.assertEqual(eval(formatted), [special] * 2)
+
+ def test_bytearray_wrap(self):
+ self.assertEqual(pprint.pformat(bytearray(), width=1), "bytearray(b'')")
+ letters = bytearray(b'abcdefghijklmnopqrstuvwxyz')
+ self.assertEqual(pprint.pformat(letters, width=40), repr(letters))
+ self.assertEqual(pprint.pformat(letters, width=28), """\
+bytearray(b'abcdefghijkl'
+ b'mnopqrstuvwxyz')""")
+ self.assertEqual(pprint.pformat(letters, width=27), """\
+bytearray(b'abcdefghijkl'
+ b'mnopqrstuvwx'
+ b'yz')""")
+ self.assertEqual(pprint.pformat(letters, width=25), """\
+bytearray(b'abcdefghijkl'
+ b'mnopqrstuvwx'
+ b'yz')""")
+ special = bytearray(range(16))
+ self.assertEqual(pprint.pformat(special, width=72), repr(special))
+ self.assertEqual(pprint.pformat(special, width=57), """\
+bytearray(b'\\x00\\x01\\x02\\x03\\x04\\x05\\x06\\x07\\x08\\t\\n\\x0b'
+ b'\\x0c\\r\\x0e\\x0f')""")
+ self.assertEqual(pprint.pformat(special, width=41), """\
+bytearray(b'\\x00\\x01\\x02\\x03'
+ b'\\x04\\x05\\x06\\x07\\x08\\t\\n\\x0b'
+ b'\\x0c\\r\\x0e\\x0f')""")
+ self.assertEqual(pprint.pformat(special, width=1), """\
+bytearray(b'\\x00\\x01\\x02\\x03'
+ b'\\x04\\x05\\x06\\x07'
+ b'\\x08\\t\\n\\x0b'
+ b'\\x0c\\r\\x0e\\x0f')""")
+ self.assertEqual(pprint.pformat({'a': 1, 'b': letters, 'c': 2},
+ width=31), """\
+{'a': 1,
+ 'b': bytearray(b'abcdefghijkl'
+ b'mnopqrstuvwx'
+ b'yz'),
+ 'c': 2}""")
+ self.assertEqual(pprint.pformat([[[[[letters]]]]], width=37), """\
+[[[[[bytearray(b'abcdefghijklmnop'
+ b'qrstuvwxyz')]]]]]""")
+ self.assertEqual(pprint.pformat([[[[[special]]]]], width=50), """\
+[[[[[bytearray(b'\\x00\\x01\\x02\\x03\\x04\\x05\\x06\\x07'
+ b'\\x08\\t\\n\\x0b\\x0c\\r\\x0e\\x0f')]]]]]""")
+
+ def test_default_dict(self):
+ d = collections.defaultdict(int)
+ self.assertEqual(pprint.pformat(d, width=1), "defaultdict(<class 'int'>, {})")
+ words = 'the quick brown fox jumped over a lazy dog'.split()
+ d = collections.defaultdict(int, zip(words, itertools.count()))
+ self.assertEqual(pprint.pformat(d),
+"""\
+defaultdict(<class 'int'>,
+ {'a': 6,
+ 'brown': 2,
+ 'dog': 8,
+ 'fox': 3,
+ 'jumped': 4,
+ 'lazy': 7,
+ 'over': 5,
+ 'quick': 1,
+ 'the': 0})""")
+
+ def test_counter(self):
+ d = collections.Counter()
+ self.assertEqual(pprint.pformat(d, width=1), "Counter()")
+ d = collections.Counter('senselessness')
+ self.assertEqual(pprint.pformat(d, width=40),
+"""\
+Counter({'s': 6,
+ 'e': 4,
+ 'n': 2,
+ 'l': 1})""")
+
+ def test_chainmap(self):
+ d = collections.ChainMap()
+ self.assertEqual(pprint.pformat(d, width=1), "ChainMap({})")
+ words = 'the quick brown fox jumped over a lazy dog'.split()
+ items = list(zip(words, itertools.count()))
+ d = collections.ChainMap(dict(items))
+ self.assertEqual(pprint.pformat(d),
+"""\
+ChainMap({'a': 6,
+ 'brown': 2,
+ 'dog': 8,
+ 'fox': 3,
+ 'jumped': 4,
+ 'lazy': 7,
+ 'over': 5,
+ 'quick': 1,
+ 'the': 0})""")
+ d = collections.ChainMap(dict(items), collections.OrderedDict(items))
+ self.assertEqual(pprint.pformat(d),
+"""\
+ChainMap({'a': 6,
+ 'brown': 2,
+ 'dog': 8,
+ 'fox': 3,
+ 'jumped': 4,
+ 'lazy': 7,
+ 'over': 5,
+ 'quick': 1,
+ 'the': 0},
+ OrderedDict([('the', 0),
+ ('quick', 1),
+ ('brown', 2),
+ ('fox', 3),
+ ('jumped', 4),
+ ('over', 5),
+ ('a', 6),
+ ('lazy', 7),
+ ('dog', 8)]))""")
+
+ def test_deque(self):
+ d = collections.deque()
+ self.assertEqual(pprint.pformat(d, width=1), "deque([])")
+ d = collections.deque(maxlen=7)
+ self.assertEqual(pprint.pformat(d, width=1), "deque([], maxlen=7)")
+ words = 'the quick brown fox jumped over a lazy dog'.split()
+ d = collections.deque(zip(words, itertools.count()))
+ self.assertEqual(pprint.pformat(d),
+"""\
+deque([('the', 0),
+ ('quick', 1),
+ ('brown', 2),
+ ('fox', 3),
+ ('jumped', 4),
+ ('over', 5),
+ ('a', 6),
+ ('lazy', 7),
+ ('dog', 8)])""")
+ d = collections.deque(zip(words, itertools.count()), maxlen=7)
+ self.assertEqual(pprint.pformat(d),
+"""\
+deque([('brown', 2),
+ ('fox', 3),
+ ('jumped', 4),
+ ('over', 5),
+ ('a', 6),
+ ('lazy', 7),
+ ('dog', 8)],
+ maxlen=7)""")
+
+ def test_user_dict(self):
+ d = collections.UserDict()
+ self.assertEqual(pprint.pformat(d, width=1), "{}")
+ words = 'the quick brown fox jumped over a lazy dog'.split()
+ d = collections.UserDict(zip(words, itertools.count()))
+ self.assertEqual(pprint.pformat(d),
+"""\
+{'a': 6,
+ 'brown': 2,
+ 'dog': 8,
+ 'fox': 3,
+ 'jumped': 4,
+ 'lazy': 7,
+ 'over': 5,
+ 'quick': 1,
+ 'the': 0}""")
+
+ def test_user_dict(self):
+ d = collections.UserList()
+ self.assertEqual(pprint.pformat(d, width=1), "[]")
+ words = 'the quick brown fox jumped over a lazy dog'.split()
+ d = collections.UserList(zip(words, itertools.count()))
+ self.assertEqual(pprint.pformat(d),
+"""\
+[('the', 0),
+ ('quick', 1),
+ ('brown', 2),
+ ('fox', 3),
+ ('jumped', 4),
+ ('over', 5),
+ ('a', 6),
+ ('lazy', 7),
+ ('dog', 8)]""")
+
+ def test_user_string(self):
+ d = collections.UserString('')
+ self.assertEqual(pprint.pformat(d, width=1), "''")
+ d = collections.UserString('the quick brown fox jumped over a lazy dog')
+ self.assertEqual(pprint.pformat(d, width=20),
+"""\
+('the quick brown '
+ 'fox jumped over '
+ 'a lazy dog')""")
+ self.assertEqual(pprint.pformat({1: d}, width=20),
+"""\
+{1: 'the quick '
+ 'brown fox '
+ 'jumped over a '
+ 'lazy dog'}""")
class DottedPrettyPrinter(pprint.PrettyPrinter):