cPickle cannot be imported. This was necessary because my last mass
checkin broke cPickle and I don't feel like debugging it right now;
but it seems a good idea in general not to require cPickle when
pickle.py is also there. A few unrelated fixes for issues while
debigging various test failures.
setup.py: disable building of cPickle until I've fixed it
Objects/...
genobject.c: disallow raising string exceptions
Lib/...
Cookie.py: fix doctest not to fail if cPickle is missing
ctypes/macholib/dyld.py: fix relative imports
sqlite3/__init__.py: fix relative import
xml/dom/__init__.py: fix relative import
Lib/test/...
regrtest.py: reduce list of skipped items on darwin
test_generators.py: don't test string exceptions; test throw() errors
test_traceback.py: don't test string exceptions
pickletester.py: don't fail if cPickle is missing
test_datetime.py: don't fail if cPickle is missing
test_descr.py: don't fail if cPickle is missing (still some other failures)
test_exceptions.py: don't fail if cPickle is missing
test_re.py: don't fail if cPickle is missing
test_array.py: use pickle, not cPickle
test_bool.py: don't fail if cPickle is missing
test_deque.py: use pickle, not cPickle
test_logging.py: use pickle, not cPickle
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.
add a self-test using doctest. Results:
- The docstring needs to be a raw string because it uses \"...\".
- The oreo example was broken: the Set-Cookie output doesn't add
quotes around "doublestuff".
- I had to change the example that prints the class of a Cookie.Cookie
instance to avoid incorporating an arbitrary object address in the
test output.
Pretty good score for both doctest and the doc string, I'd say!
cookies that contain '=' as part of the value. This patch modifies
Cookie.py to allow '=' as a legal character, and to make the key
search nongreedy so it stops at the first '='.
added test script and expected output file as well
this closes patch 103297.
__all__ attributes will be added to other modules without first submitting
a patch, just adding the necessary line to the test script to verify
more-or-less correct implementation.
Guido van Rossum
Georg Brandl
Walter Dörwald
Andrew M. Kuchling
Raymond Hettinger
Fred Drake
Martin v. Löwis
Tim Peters
Skip Montanaro