2000-03-28 19:51:17 -04:00
|
|
|
test_extcall
|
|
|
|
|
() {}
|
|
|
|
|
(1,) {}
|
|
|
|
|
(1, 2) {}
|
|
|
|
|
(1, 2, 3) {}
|
|
|
|
|
(1, 2, 3, 4, 5) {}
|
2000-03-28 19:53:22 -04:00
|
|
|
(1, 2, 3, 4, 5) {}
|
|
|
|
|
(1, 2, 3, 4, 5) {}
|
2001-01-21 14:52:02 -04:00
|
|
|
(1, 2, 3) {'a': 4, 'b': 5}
|
|
|
|
|
(1, 2, 3, 4, 5) {'a': 6, 'b': 7}
|
|
|
|
|
(1, 2, 3, 6, 7) {'a': 8, 'b': 9, 'x': 4, 'y': 5}
|
2001-01-15 18:14:16 -04:00
|
|
|
TypeError: g() takes at least 1 argument (0 given)
|
|
|
|
|
TypeError: g() takes at least 1 argument (0 given)
|
|
|
|
|
TypeError: g() takes at least 1 argument (0 given)
|
2000-03-28 19:51:17 -04:00
|
|
|
1 () {}
|
|
|
|
|
1 (2,) {}
|
|
|
|
|
1 (2, 3) {}
|
|
|
|
|
1 (2, 3, 4, 5) {}
|
2000-03-30 19:55:31 -04:00
|
|
|
0 (1, 2) {}
|
2001-01-21 14:52:02 -04:00
|
|
|
1 () {'a': 1, 'b': 2, 'c': 3, 'd': 4}
|
|
|
|
|
{'a': 1, 'b': 2, 'c': 3}
|
|
|
|
|
{'a': 1, 'b': 2, 'c': 3}
|
2001-01-15 18:14:16 -04:00
|
|
|
g() got multiple values for keyword argument 'x'
|
|
|
|
|
g() got multiple values for keyword argument 'b'
|
|
|
|
|
f() keywords must be strings
|
2000-10-24 16:57:45 -03:00
|
|
|
h() got an unexpected keyword argument 'e'
|
2001-01-15 18:14:16 -04:00
|
|
|
h() argument after * must be a sequence
|
2001-04-11 10:53:35 -03:00
|
|
|
dir() argument after * must be a sequence
|
2001-08-16 05:32:39 -03:00
|
|
|
NoneType object argument after * must be a sequence
|
2001-01-15 18:14:16 -04:00
|
|
|
h() argument after ** must be a dictionary
|
2001-04-11 10:53:35 -03:00
|
|
|
dir() argument after ** must be a dictionary
|
2001-08-16 05:32:39 -03:00
|
|
|
NoneType object argument after ** must be a dictionary
|
2001-04-11 10:53:35 -03:00
|
|
|
dir() got multiple values for keyword argument 'b'
|
2002-04-03 18:41:51 -04:00
|
|
|
3 512 True
|
2000-10-30 15:41:33 -04:00
|
|
|
3
|
|
|
|
|
3
|
2001-01-15 18:14:16 -04:00
|
|
|
za () {} -> za() takes exactly 1 argument (0 given)
|
|
|
|
|
za () {'a': 'aa'} -> ok za aa B D E V a
|
|
|
|
|
za () {'d': 'dd'} -> za() got an unexpected keyword argument 'd'
|
2001-01-21 14:52:02 -04:00
|
|
|
za () {'a': 'aa', 'd': 'dd'} -> za() got an unexpected keyword argument 'd'
|
Get rid of the superstitious "~" in dict hashing's "i = (~hash) & mask".
The comment following used to say:
/* We use ~hash instead of hash, as degenerate hash functions, such
as for ints <sigh>, can have lots of leading zeros. It's not
really a performance risk, but better safe than sorry.
12-Dec-00 tim: so ~hash produces lots of leading ones instead --
what's the gain? */
That is, there was never a good reason for doing it. And to the contrary,
as explained on Python-Dev last December, it tended to make the *sum*
(i + incr) & mask (which is the first table index examined in case of
collison) the same "too often" across distinct hashes.
Changing to the simpler "i = hash & mask" reduced the number of string-dict
collisions (== # number of times we go around the lookup for-loop) from about
6 million to 5 million during a full run of the test suite (these are
approximate because the test suite does some random stuff from run to run).
The number of collisions in non-string dicts also decreased, but not as
dramatically.
Note that this may, for a given dict, change the order (wrt previous
releases) of entries exposed by .keys(), .values() and .items(). A number
of std tests suffered bogus failures as a result. For dicts keyed by
small ints, or (less so) by characters, the order is much more likely to be
in increasing order of key now; e.g.,
>>> d = {}
>>> for i in range(10):
... d[i] = i
...
>>> d
{0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9}
>>>
Unfortunately. people may latch on to that in small examples and draw a
bogus conclusion.
test_support.py
Moved test_extcall's sortdict() into test_support, made it stronger,
and imported sortdict into other std tests that needed it.
test_unicode.py
Excluced cp875 from the "roundtrip over range(128)" test, because
cp875 doesn't have a well-defined inverse for unicode("?", "cp875").
See Python-Dev for excruciating details.
Cookie.py
Chaged various output functions to sort dicts before building
strings from them.
test_extcall
Fiddled the expected-result file. This remains sensitive to native
dict ordering, because, e.g., if there are multiple errors in a
keyword-arg dict (and test_extcall sets up many cases like that), the
specific error Python complains about first depends on native dict
ordering.
2001-05-12 21:19:31 -03:00
|
|
|
za () {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> za() got an unexpected keyword argument 'b'
|
2001-01-15 18:14:16 -04:00
|
|
|
za (1, 2) {} -> za() takes exactly 1 argument (2 given)
|
|
|
|
|
za (1, 2) {'a': 'aa'} -> za() takes exactly 1 non-keyword argument (2 given)
|
|
|
|
|
za (1, 2) {'d': 'dd'} -> za() takes exactly 1 non-keyword argument (2 given)
|
2001-01-21 14:52:02 -04:00
|
|
|
za (1, 2) {'a': 'aa', 'd': 'dd'} -> za() takes exactly 1 non-keyword argument (2 given)
|
|
|
|
|
za (1, 2) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> za() takes exactly 1 non-keyword argument (2 given)
|
2001-01-15 18:14:16 -04:00
|
|
|
za (1, 2, 3, 4, 5) {} -> za() takes exactly 1 argument (5 given)
|
|
|
|
|
za (1, 2, 3, 4, 5) {'a': 'aa'} -> za() takes exactly 1 non-keyword argument (5 given)
|
|
|
|
|
za (1, 2, 3, 4, 5) {'d': 'dd'} -> za() takes exactly 1 non-keyword argument (5 given)
|
2001-01-21 14:52:02 -04:00
|
|
|
za (1, 2, 3, 4, 5) {'a': 'aa', 'd': 'dd'} -> za() takes exactly 1 non-keyword argument (5 given)
|
|
|
|
|
za (1, 2, 3, 4, 5) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> za() takes exactly 1 non-keyword argument (5 given)
|
2001-01-15 18:14:16 -04:00
|
|
|
zade () {} -> zade() takes at least 1 argument (0 given)
|
|
|
|
|
zade () {'a': 'aa'} -> ok zade aa B d e V a
|
|
|
|
|
zade () {'d': 'dd'} -> zade() takes at least 1 non-keyword argument (0 given)
|
2001-01-21 14:52:02 -04:00
|
|
|
zade () {'a': 'aa', 'd': 'dd'} -> ok zade aa B dd e V d
|
|
|
|
|
zade () {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zade() got an unexpected keyword argument 'b'
|
2001-01-15 18:14:16 -04:00
|
|
|
zade (1, 2) {} -> ok zade 1 B 2 e V e
|
|
|
|
|
zade (1, 2) {'a': 'aa'} -> zade() got multiple values for keyword argument 'a'
|
|
|
|
|
zade (1, 2) {'d': 'dd'} -> zade() got multiple values for keyword argument 'd'
|
Get rid of the superstitious "~" in dict hashing's "i = (~hash) & mask".
The comment following used to say:
/* We use ~hash instead of hash, as degenerate hash functions, such
as for ints <sigh>, can have lots of leading zeros. It's not
really a performance risk, but better safe than sorry.
12-Dec-00 tim: so ~hash produces lots of leading ones instead --
what's the gain? */
That is, there was never a good reason for doing it. And to the contrary,
as explained on Python-Dev last December, it tended to make the *sum*
(i + incr) & mask (which is the first table index examined in case of
collison) the same "too often" across distinct hashes.
Changing to the simpler "i = hash & mask" reduced the number of string-dict
collisions (== # number of times we go around the lookup for-loop) from about
6 million to 5 million during a full run of the test suite (these are
approximate because the test suite does some random stuff from run to run).
The number of collisions in non-string dicts also decreased, but not as
dramatically.
Note that this may, for a given dict, change the order (wrt previous
releases) of entries exposed by .keys(), .values() and .items(). A number
of std tests suffered bogus failures as a result. For dicts keyed by
small ints, or (less so) by characters, the order is much more likely to be
in increasing order of key now; e.g.,
>>> d = {}
>>> for i in range(10):
... d[i] = i
...
>>> d
{0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9}
>>>
Unfortunately. people may latch on to that in small examples and draw a
bogus conclusion.
test_support.py
Moved test_extcall's sortdict() into test_support, made it stronger,
and imported sortdict into other std tests that needed it.
test_unicode.py
Excluced cp875 from the "roundtrip over range(128)" test, because
cp875 doesn't have a well-defined inverse for unicode("?", "cp875").
See Python-Dev for excruciating details.
Cookie.py
Chaged various output functions to sort dicts before building
strings from them.
test_extcall
Fiddled the expected-result file. This remains sensitive to native
dict ordering, because, e.g., if there are multiple errors in a
keyword-arg dict (and test_extcall sets up many cases like that), the
specific error Python complains about first depends on native dict
ordering.
2001-05-12 21:19:31 -03:00
|
|
|
zade (1, 2) {'a': 'aa', 'd': 'dd'} -> zade() got multiple values for keyword argument 'a'
|
|
|
|
|
zade (1, 2) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zade() got multiple values for keyword argument 'a'
|
2001-01-15 18:14:16 -04:00
|
|
|
zade (1, 2, 3, 4, 5) {} -> zade() takes at most 3 arguments (5 given)
|
|
|
|
|
zade (1, 2, 3, 4, 5) {'a': 'aa'} -> zade() takes at most 3 non-keyword arguments (5 given)
|
|
|
|
|
zade (1, 2, 3, 4, 5) {'d': 'dd'} -> zade() takes at most 3 non-keyword arguments (5 given)
|
2001-01-21 14:52:02 -04:00
|
|
|
zade (1, 2, 3, 4, 5) {'a': 'aa', 'd': 'dd'} -> zade() takes at most 3 non-keyword arguments (5 given)
|
|
|
|
|
zade (1, 2, 3, 4, 5) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zade() takes at most 3 non-keyword arguments (5 given)
|
2001-01-15 18:14:16 -04:00
|
|
|
zabk () {} -> zabk() takes exactly 2 arguments (0 given)
|
|
|
|
|
zabk () {'a': 'aa'} -> zabk() takes exactly 2 non-keyword arguments (1 given)
|
|
|
|
|
zabk () {'d': 'dd'} -> zabk() takes exactly 2 non-keyword arguments (0 given)
|
2001-01-21 14:52:02 -04:00
|
|
|
zabk () {'a': 'aa', 'd': 'dd'} -> zabk() takes exactly 2 non-keyword arguments (1 given)
|
|
|
|
|
zabk () {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> ok zabk aa bb D E V {'d': 'dd', 'e': 'ee'}
|
2001-01-15 18:14:16 -04:00
|
|
|
zabk (1, 2) {} -> ok zabk 1 2 D E V {}
|
|
|
|
|
zabk (1, 2) {'a': 'aa'} -> zabk() got multiple values for keyword argument 'a'
|
|
|
|
|
zabk (1, 2) {'d': 'dd'} -> ok zabk 1 2 D E V {'d': 'dd'}
|
2001-01-21 14:52:02 -04:00
|
|
|
zabk (1, 2) {'a': 'aa', 'd': 'dd'} -> zabk() got multiple values for keyword argument 'a'
|
Get rid of the superstitious "~" in dict hashing's "i = (~hash) & mask".
The comment following used to say:
/* We use ~hash instead of hash, as degenerate hash functions, such
as for ints <sigh>, can have lots of leading zeros. It's not
really a performance risk, but better safe than sorry.
12-Dec-00 tim: so ~hash produces lots of leading ones instead --
what's the gain? */
That is, there was never a good reason for doing it. And to the contrary,
as explained on Python-Dev last December, it tended to make the *sum*
(i + incr) & mask (which is the first table index examined in case of
collison) the same "too often" across distinct hashes.
Changing to the simpler "i = hash & mask" reduced the number of string-dict
collisions (== # number of times we go around the lookup for-loop) from about
6 million to 5 million during a full run of the test suite (these are
approximate because the test suite does some random stuff from run to run).
The number of collisions in non-string dicts also decreased, but not as
dramatically.
Note that this may, for a given dict, change the order (wrt previous
releases) of entries exposed by .keys(), .values() and .items(). A number
of std tests suffered bogus failures as a result. For dicts keyed by
small ints, or (less so) by characters, the order is much more likely to be
in increasing order of key now; e.g.,
>>> d = {}
>>> for i in range(10):
... d[i] = i
...
>>> d
{0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9}
>>>
Unfortunately. people may latch on to that in small examples and draw a
bogus conclusion.
test_support.py
Moved test_extcall's sortdict() into test_support, made it stronger,
and imported sortdict into other std tests that needed it.
test_unicode.py
Excluced cp875 from the "roundtrip over range(128)" test, because
cp875 doesn't have a well-defined inverse for unicode("?", "cp875").
See Python-Dev for excruciating details.
Cookie.py
Chaged various output functions to sort dicts before building
strings from them.
test_extcall
Fiddled the expected-result file. This remains sensitive to native
dict ordering, because, e.g., if there are multiple errors in a
keyword-arg dict (and test_extcall sets up many cases like that), the
specific error Python complains about first depends on native dict
ordering.
2001-05-12 21:19:31 -03:00
|
|
|
zabk (1, 2) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zabk() got multiple values for keyword argument 'a'
|
2001-01-15 18:14:16 -04:00
|
|
|
zabk (1, 2, 3, 4, 5) {} -> zabk() takes exactly 2 arguments (5 given)
|
|
|
|
|
zabk (1, 2, 3, 4, 5) {'a': 'aa'} -> zabk() takes exactly 2 non-keyword arguments (5 given)
|
|
|
|
|
zabk (1, 2, 3, 4, 5) {'d': 'dd'} -> zabk() takes exactly 2 non-keyword arguments (5 given)
|
2001-01-21 14:52:02 -04:00
|
|
|
zabk (1, 2, 3, 4, 5) {'a': 'aa', 'd': 'dd'} -> zabk() takes exactly 2 non-keyword arguments (5 given)
|
|
|
|
|
zabk (1, 2, 3, 4, 5) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zabk() takes exactly 2 non-keyword arguments (5 given)
|
2001-01-15 18:14:16 -04:00
|
|
|
zabdv () {} -> zabdv() takes at least 2 arguments (0 given)
|
|
|
|
|
zabdv () {'a': 'aa'} -> zabdv() takes at least 2 non-keyword arguments (1 given)
|
|
|
|
|
zabdv () {'d': 'dd'} -> zabdv() takes at least 2 non-keyword arguments (0 given)
|
2001-01-21 14:52:02 -04:00
|
|
|
zabdv () {'a': 'aa', 'd': 'dd'} -> zabdv() takes at least 2 non-keyword arguments (1 given)
|
|
|
|
|
zabdv () {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zabdv() got an unexpected keyword argument 'e'
|
2001-01-15 18:14:16 -04:00
|
|
|
zabdv (1, 2) {} -> ok zabdv 1 2 d E () e
|
|
|
|
|
zabdv (1, 2) {'a': 'aa'} -> zabdv() got multiple values for keyword argument 'a'
|
|
|
|
|
zabdv (1, 2) {'d': 'dd'} -> ok zabdv 1 2 dd E () d
|
2001-01-21 14:52:02 -04:00
|
|
|
zabdv (1, 2) {'a': 'aa', 'd': 'dd'} -> zabdv() got multiple values for keyword argument 'a'
|
Get rid of the superstitious "~" in dict hashing's "i = (~hash) & mask".
The comment following used to say:
/* We use ~hash instead of hash, as degenerate hash functions, such
as for ints <sigh>, can have lots of leading zeros. It's not
really a performance risk, but better safe than sorry.
12-Dec-00 tim: so ~hash produces lots of leading ones instead --
what's the gain? */
That is, there was never a good reason for doing it. And to the contrary,
as explained on Python-Dev last December, it tended to make the *sum*
(i + incr) & mask (which is the first table index examined in case of
collison) the same "too often" across distinct hashes.
Changing to the simpler "i = hash & mask" reduced the number of string-dict
collisions (== # number of times we go around the lookup for-loop) from about
6 million to 5 million during a full run of the test suite (these are
approximate because the test suite does some random stuff from run to run).
The number of collisions in non-string dicts also decreased, but not as
dramatically.
Note that this may, for a given dict, change the order (wrt previous
releases) of entries exposed by .keys(), .values() and .items(). A number
of std tests suffered bogus failures as a result. For dicts keyed by
small ints, or (less so) by characters, the order is much more likely to be
in increasing order of key now; e.g.,
>>> d = {}
>>> for i in range(10):
... d[i] = i
...
>>> d
{0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9}
>>>
Unfortunately. people may latch on to that in small examples and draw a
bogus conclusion.
test_support.py
Moved test_extcall's sortdict() into test_support, made it stronger,
and imported sortdict into other std tests that needed it.
test_unicode.py
Excluced cp875 from the "roundtrip over range(128)" test, because
cp875 doesn't have a well-defined inverse for unicode("?", "cp875").
See Python-Dev for excruciating details.
Cookie.py
Chaged various output functions to sort dicts before building
strings from them.
test_extcall
Fiddled the expected-result file. This remains sensitive to native
dict ordering, because, e.g., if there are multiple errors in a
keyword-arg dict (and test_extcall sets up many cases like that), the
specific error Python complains about first depends on native dict
ordering.
2001-05-12 21:19:31 -03:00
|
|
|
zabdv (1, 2) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zabdv() got multiple values for keyword argument 'a'
|
2001-01-15 18:14:16 -04:00
|
|
|
zabdv (1, 2, 3, 4, 5) {} -> ok zabdv 1 2 3 E (4, 5) e
|
|
|
|
|
zabdv (1, 2, 3, 4, 5) {'a': 'aa'} -> zabdv() got multiple values for keyword argument 'a'
|
|
|
|
|
zabdv (1, 2, 3, 4, 5) {'d': 'dd'} -> zabdv() got multiple values for keyword argument 'd'
|
Get rid of the superstitious "~" in dict hashing's "i = (~hash) & mask".
The comment following used to say:
/* We use ~hash instead of hash, as degenerate hash functions, such
as for ints <sigh>, can have lots of leading zeros. It's not
really a performance risk, but better safe than sorry.
12-Dec-00 tim: so ~hash produces lots of leading ones instead --
what's the gain? */
That is, there was never a good reason for doing it. And to the contrary,
as explained on Python-Dev last December, it tended to make the *sum*
(i + incr) & mask (which is the first table index examined in case of
collison) the same "too often" across distinct hashes.
Changing to the simpler "i = hash & mask" reduced the number of string-dict
collisions (== # number of times we go around the lookup for-loop) from about
6 million to 5 million during a full run of the test suite (these are
approximate because the test suite does some random stuff from run to run).
The number of collisions in non-string dicts also decreased, but not as
dramatically.
Note that this may, for a given dict, change the order (wrt previous
releases) of entries exposed by .keys(), .values() and .items(). A number
of std tests suffered bogus failures as a result. For dicts keyed by
small ints, or (less so) by characters, the order is much more likely to be
in increasing order of key now; e.g.,
>>> d = {}
>>> for i in range(10):
... d[i] = i
...
>>> d
{0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9}
>>>
Unfortunately. people may latch on to that in small examples and draw a
bogus conclusion.
test_support.py
Moved test_extcall's sortdict() into test_support, made it stronger,
and imported sortdict into other std tests that needed it.
test_unicode.py
Excluced cp875 from the "roundtrip over range(128)" test, because
cp875 doesn't have a well-defined inverse for unicode("?", "cp875").
See Python-Dev for excruciating details.
Cookie.py
Chaged various output functions to sort dicts before building
strings from them.
test_extcall
Fiddled the expected-result file. This remains sensitive to native
dict ordering, because, e.g., if there are multiple errors in a
keyword-arg dict (and test_extcall sets up many cases like that), the
specific error Python complains about first depends on native dict
ordering.
2001-05-12 21:19:31 -03:00
|
|
|
zabdv (1, 2, 3, 4, 5) {'a': 'aa', 'd': 'dd'} -> zabdv() got multiple values for keyword argument 'a'
|
|
|
|
|
zabdv (1, 2, 3, 4, 5) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zabdv() got multiple values for keyword argument 'a'
|
2001-01-15 18:14:16 -04:00
|
|
|
zabdevk () {} -> zabdevk() takes at least 2 arguments (0 given)
|
|
|
|
|
zabdevk () {'a': 'aa'} -> zabdevk() takes at least 2 non-keyword arguments (1 given)
|
|
|
|
|
zabdevk () {'d': 'dd'} -> zabdevk() takes at least 2 non-keyword arguments (0 given)
|
2001-01-21 14:52:02 -04:00
|
|
|
zabdevk () {'a': 'aa', 'd': 'dd'} -> zabdevk() takes at least 2 non-keyword arguments (1 given)
|
|
|
|
|
zabdevk () {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> ok zabdevk aa bb dd ee () {}
|
2001-01-15 18:14:16 -04:00
|
|
|
zabdevk (1, 2) {} -> ok zabdevk 1 2 d e () {}
|
|
|
|
|
zabdevk (1, 2) {'a': 'aa'} -> zabdevk() got multiple values for keyword argument 'a'
|
|
|
|
|
zabdevk (1, 2) {'d': 'dd'} -> ok zabdevk 1 2 dd e () {}
|
2001-01-21 14:52:02 -04:00
|
|
|
zabdevk (1, 2) {'a': 'aa', 'd': 'dd'} -> zabdevk() got multiple values for keyword argument 'a'
|
Get rid of the superstitious "~" in dict hashing's "i = (~hash) & mask".
The comment following used to say:
/* We use ~hash instead of hash, as degenerate hash functions, such
as for ints <sigh>, can have lots of leading zeros. It's not
really a performance risk, but better safe than sorry.
12-Dec-00 tim: so ~hash produces lots of leading ones instead --
what's the gain? */
That is, there was never a good reason for doing it. And to the contrary,
as explained on Python-Dev last December, it tended to make the *sum*
(i + incr) & mask (which is the first table index examined in case of
collison) the same "too often" across distinct hashes.
Changing to the simpler "i = hash & mask" reduced the number of string-dict
collisions (== # number of times we go around the lookup for-loop) from about
6 million to 5 million during a full run of the test suite (these are
approximate because the test suite does some random stuff from run to run).
The number of collisions in non-string dicts also decreased, but not as
dramatically.
Note that this may, for a given dict, change the order (wrt previous
releases) of entries exposed by .keys(), .values() and .items(). A number
of std tests suffered bogus failures as a result. For dicts keyed by
small ints, or (less so) by characters, the order is much more likely to be
in increasing order of key now; e.g.,
>>> d = {}
>>> for i in range(10):
... d[i] = i
...
>>> d
{0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9}
>>>
Unfortunately. people may latch on to that in small examples and draw a
bogus conclusion.
test_support.py
Moved test_extcall's sortdict() into test_support, made it stronger,
and imported sortdict into other std tests that needed it.
test_unicode.py
Excluced cp875 from the "roundtrip over range(128)" test, because
cp875 doesn't have a well-defined inverse for unicode("?", "cp875").
See Python-Dev for excruciating details.
Cookie.py
Chaged various output functions to sort dicts before building
strings from them.
test_extcall
Fiddled the expected-result file. This remains sensitive to native
dict ordering, because, e.g., if there are multiple errors in a
keyword-arg dict (and test_extcall sets up many cases like that), the
specific error Python complains about first depends on native dict
ordering.
2001-05-12 21:19:31 -03:00
|
|
|
zabdevk (1, 2) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zabdevk() got multiple values for keyword argument 'a'
|
2001-01-15 18:14:16 -04:00
|
|
|
zabdevk (1, 2, 3, 4, 5) {} -> ok zabdevk 1 2 3 4 (5,) {}
|
|
|
|
|
zabdevk (1, 2, 3, 4, 5) {'a': 'aa'} -> zabdevk() got multiple values for keyword argument 'a'
|
|
|
|
|
zabdevk (1, 2, 3, 4, 5) {'d': 'dd'} -> zabdevk() got multiple values for keyword argument 'd'
|
Get rid of the superstitious "~" in dict hashing's "i = (~hash) & mask".
The comment following used to say:
/* We use ~hash instead of hash, as degenerate hash functions, such
as for ints <sigh>, can have lots of leading zeros. It's not
really a performance risk, but better safe than sorry.
12-Dec-00 tim: so ~hash produces lots of leading ones instead --
what's the gain? */
That is, there was never a good reason for doing it. And to the contrary,
as explained on Python-Dev last December, it tended to make the *sum*
(i + incr) & mask (which is the first table index examined in case of
collison) the same "too often" across distinct hashes.
Changing to the simpler "i = hash & mask" reduced the number of string-dict
collisions (== # number of times we go around the lookup for-loop) from about
6 million to 5 million during a full run of the test suite (these are
approximate because the test suite does some random stuff from run to run).
The number of collisions in non-string dicts also decreased, but not as
dramatically.
Note that this may, for a given dict, change the order (wrt previous
releases) of entries exposed by .keys(), .values() and .items(). A number
of std tests suffered bogus failures as a result. For dicts keyed by
small ints, or (less so) by characters, the order is much more likely to be
in increasing order of key now; e.g.,
>>> d = {}
>>> for i in range(10):
... d[i] = i
...
>>> d
{0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9}
>>>
Unfortunately. people may latch on to that in small examples and draw a
bogus conclusion.
test_support.py
Moved test_extcall's sortdict() into test_support, made it stronger,
and imported sortdict into other std tests that needed it.
test_unicode.py
Excluced cp875 from the "roundtrip over range(128)" test, because
cp875 doesn't have a well-defined inverse for unicode("?", "cp875").
See Python-Dev for excruciating details.
Cookie.py
Chaged various output functions to sort dicts before building
strings from them.
test_extcall
Fiddled the expected-result file. This remains sensitive to native
dict ordering, because, e.g., if there are multiple errors in a
keyword-arg dict (and test_extcall sets up many cases like that), the
specific error Python complains about first depends on native dict
ordering.
2001-05-12 21:19:31 -03:00
|
|
|
zabdevk (1, 2, 3, 4, 5) {'a': 'aa', 'd': 'dd'} -> zabdevk() got multiple values for keyword argument 'a'
|
|
|
|
|
zabdevk (1, 2, 3, 4, 5) {'a': 'aa', 'b': 'bb', 'd': 'dd', 'e': 'ee'} -> zabdevk() got multiple values for keyword argument 'a'
|
