[Python-checkins] CVS: python/dist/src/Misc NEWS,1.165,1.166
Tim Peters
tim_one@users.sourceforge.net
Sat, 12 May 2001 17:19:34 -0700
- Previous message: [Python-checkins] CVS: python/dist/src/Lib/test test_cookie.py,1.9,1.10 test_extcall.py,1.15,1.16 test_pyexpat.py,1.8,1.9 test_regex.py,1.9,1.10 test_support.py,1.21,1.22 test_unicode.py,1.32,1.33
- Next message: [Python-checkins] CVS: python/dist/src/Objects dictobject.c,2.86,2.87
- Messages sorted by:
[ date ]
[ thread ]
[ subject ]
[ author ]
Update of /cvsroot/python/python/dist/src/Misc
In directory usw-pr-cvs1:/tmp/cvs-serv1323/python/dist/src/Misc
Modified Files:
NEWS
Log Message:
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.
Index: NEWS
===================================================================
RCS file: /cvsroot/python/python/dist/src/Misc/NEWS,v
retrieving revision 1.165
retrieving revision 1.166
diff -C2 -r1.165 -r1.166
*** NEWS 2001/05/11 21:51:48 1.165
--- NEWS 2001/05/13 00:19:31 1.166
***************
*** 24,27 ****
--- 24,37 ----
algorithms to break.
+ - The implementation of dicts suffers fewer collisions, which has speed
+ benefits. However, the order in which dict entries appear in dict.keys(),
+ dict.values() and dict.items() may differ from previous releases for a
+ given dict. Nothing is defined about this order, so no program should
+ rely on it. Nevertheless, it's easy to write test cases that rely on the
+ order by accident, typically because of printing the str() or repr() of a
+ dict to an "expected results" file. See Lib/test/test_support.py's new
+ sortdict(dict) function for a simple way to display a dict in sorted
+ order.
+
- Dictionary objects now support the "in" operator: "x in dict" means
the same as dict.has_key(x).
- Previous message: [Python-checkins] CVS: python/dist/src/Lib/test test_cookie.py,1.9,1.10 test_extcall.py,1.15,1.16 test_pyexpat.py,1.8,1.9 test_regex.py,1.9,1.10 test_support.py,1.21,1.22 test_unicode.py,1.32,1.33
- Next message: [Python-checkins] CVS: python/dist/src/Objects dictobject.c,2.86,2.87
- Messages sorted by:
[ date ]
[ thread ]
[ subject ]
[ author ]