2000-08-19 12:21:12 -03:00
|
|
|
test_cookie
|
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
|
|
|
<SimpleCookie: chips='ahoy' vienna='finger'>
|
2000-08-19 12:21:12 -03:00
|
|
|
Set-Cookie: chips=ahoy;
|
2000-08-24 08:56:19 -03:00
|
|
|
Set-Cookie: vienna=finger;
|
2000-08-19 12:21:12 -03:00
|
|
|
chips 'ahoy' 'ahoy'
|
2000-08-24 08:56:19 -03:00
|
|
|
Set-Cookie: chips=ahoy;
|
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
|
|
|
vienna 'finger' 'finger'
|
|
|
|
|
Set-Cookie: vienna=finger;
|
2001-01-24 13:19:08 -04:00
|
|
|
<SimpleCookie: keebler='E=mc2; L="Loves"; fudge=\n;'>
|
2000-08-19 12:21:12 -03:00
|
|
|
Set-Cookie: keebler="E=mc2; L=\"Loves\"; fudge=\012;";
|
2001-01-24 13:19:08 -04:00
|
|
|
keebler 'E=mc2; L="Loves"; fudge=\n;' 'E=mc2; L="Loves"; fudge=\n;'
|
2000-08-24 08:56:19 -03:00
|
|
|
Set-Cookie: keebler="E=mc2; L=\"Loves\"; fudge=\012;";
|
2001-02-20 21:17:54 -04:00
|
|
|
<SimpleCookie: keebler='E=mc2'>
|
|
|
|
|
Set-Cookie: keebler=E=mc2;
|
|
|
|
|
keebler 'E=mc2' 'E=mc2'
|
|
|
|
|
Set-Cookie: keebler=E=mc2;
|
2000-08-19 12:57:33 -03:00
|
|
|
Set-Cookie: Customer="WILE_E_COYOTE"; Path=/acme;
|
|
|
|
|
|
|
|
|
|
<SCRIPT LANGUAGE="JavaScript">
|
|
|
|
|
<!-- begin hiding
|
|
|
|
|
document.cookie = "Customer="WILE_E_COYOTE"; Path=/acme; Version=1;"
|
|
|
|
|
// end hiding -->
|
|
|
|
|
</script>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<SCRIPT LANGUAGE="JavaScript">
|
|
|
|
|
<!-- begin hiding
|
|
|
|
|
document.cookie = "Customer="WILE_E_COYOTE"; Path=/acme;"
|
|
|
|
|
// end hiding -->
|
|
|
|
|
</script>
|
|
|
|
|
|
2001-04-06 18:20:58 -03:00
|
|
|
If anything blows up after this line, it's from Cookie's doctest.
|
