mirror of https://github.com/python/cpython
2f228e75e4
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.
|
||
|---|---|---|
| .. | ||
| RPM | ||
| ACKS | ||
| AIX-NOTES | ||
| BLURB | ||
| BLURB.LUTZ | ||
| BLURB.WINDOWS | ||
| BeOS-NOTES | ||
| BeOS-setup.py | ||
| HISTORY | ||
| HPUX-NOTES | ||
| HYPE | ||
| Makefile.pre.in | ||
| NEWS | ||
| NEXT-NOTES | ||
| PURIFY.README | ||
| Porting | ||
| README | ||
| RFD | ||
| cheatsheet | ||
| comparisons | ||
| editline-fix | ||
| faq2html.py | ||
| find_recursionlimit.py | ||
| fixfuncptrs.sh | ||
| gdbinit | ||
| indent.pro | ||
| python-mode.el | ||
| python.man | ||
| renumber.py | ||
| setuid-prog.c | ||
| unicode.txt | ||
| vgrindefs | ||
README
Python Misc subdirectory ======================== This directory contains files that wouldn't fit in elsewhere, in particular the UNIX manual page, an Emacs mode for Python source code, and a list of Frequently Asked Questions (and their answers). Some documents are only of historic importance. Files found here ---------------- ACKS Acknowledgements AIX-NOTES Notes for building Python on AIX (all new!) BLURB A quick description of Python for newcomers BLURB.LUTZ A testimonial from a converted Tcl/Perl hacker :-) BLURB.WINDOWS First announcement of Python for Windows HISTORY News from previous releases -- oldest last HPUX-NOTES Notes about dynamic loading under HP-UX HYPE More hype about Python Makefile Used for administrative chores like cleaning up Makefile.pre.in Generic Makefile template for building extensions NEWS News for this release NEXT-NOTES Porting notes for NeXT PURIFY.README Information for Purify users Porting Mini-FAQ on porting to new platforms README The file you're reading now RFD Request For Discussion about a Python newsgroup cheatsheet Quick summary of Python by Ken Manheimer comparisons (Old) comparisons to some other languages editline-fix A news article on how to fix R$'s editline for Python faq2html FAQ to HTML converter by Ka-Ping Yee find_recursionlimit.py Script to find a value for sys.maxrecursionlimit fixfuncptrs.sh Shell script to fix function pointer initializers indent.pro GNU indent profile approximating my C style python-mode.el Emacs mode for editing Python programs python.man UNIX man page for the python interpreter renumber.py Script to renumber the sections in the FAQ setuid-prog.c C helper program for set-uid Python scripts unicode.txt Marc-Andre Lemburg's specification of the Unicode API vgrindefs Python configuration for vgrind (a generic pretty printer)