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.

are including Carbon/Carbon.h in stead of the old headers (unless WITHOUT_FRAMEWORKS
is defined, as it will be for classic MacPython) and selectively disabling all the
stuff that is unneeded in a unix-Python (event handling, etc).

Be more sensible about when to use TARGET_API_MAC_OS8 in stead of !TARGET_API_MAC_CARBON. This should greatly facilitate porting stuff to OSX in its MachO/BSD incarnation.

switching from tp_getattr to tp_getattro.

This now uses the new gotofileline() method defined in FileList.py.

rather than the idle.py script.  This has advantages and
disadvantages; the biggest advantage being that we can more easily
have an alternative main program.

Add gotofileline(), a convenience method which I intend to use in a
variant.

Rename test() to _test().

Allow module getattr and setattr to exploit string interning, via the
previously null module object tp_getattro and tp_setattro slots.   Yields
a very nice speedup for things like random.random and os.path etc.

When getting a string buffer for a string we just created, use
PyString_AS_STRING() instead of PyString_AsString() to avoid the
call overhead and extra type check.

catch that instead of using a bare except clause.

catch that instead of using a bare except clause.

exception (for compatibility with old versions of Python).

Add a comment elsewhere making clear an assumption in the code.

a bare except clause.

so only catch that specific exception.

some stuff around.

constants used by other macros from the headers.

Conditionalize VREPRINT and VDISCARD; these are not available on HP-UX.

This closes bug #417418.

Finish the last set of changes to these files so the conversion does not
break.

For rich comparisons, use instance_getattr2() when possible to avoid
the expense of setting an AttributeError.  Also intern the name_op[]
table and use the interned strings rather than creating a new string
and interning it each time through.

Remove unused import of "sys".

If the file TESTFN exists before we start, try to remove it.

Add spaces around the = in some assignments.

"\begin{classdesc*}{SomeClass}" -- the rendering of \unspecified was
identical to \moreargs, so this helps clarify things just a little.

excclassdesc environment.

class without providing any information about the constructor.  This
should be used for classes which only exist to act as containers rather
than as factories for instances.

useful documentation on the Scrap Manager.

doesn't know how to do LE, LT, GE, GT.  dict_richcompare can't do the
latter any faster than dict_compare can.  More importantly, for
cmp(dict1, dict2), Python *first* tries rich compares with EQ, LT, and
GT one at a time, even if the tp_compare slot is defined, and
dict_richcompare called dict_compare for the latter two because
it couldn't do them itself.  The result was a lot of wasted calls to
dict_compare.  Now dict_richcompare gives up at once the times Python
calls it with LT and GT from try_rich_to_3way_compare(), and dict_compare
is called only once (when Python gets around to trying the tp_compare
slot).
Continued mystery:  despite that this cut the number of calls to
dict_compare approximately in half in test_mutants.py, the latter still
runs amazingly slowly.  Running under the debugger doesn't show excessive
activity in the dict comparison code anymore, so I'm guessing the culprit
is somewhere else -- but where?  Perhaps in the element (key/value)
comparison code?  We clearly spend a lot of time figuring out how to
compare things.

A Mystery:  test_mutants ran amazingly slowly even before dictobject.c
"got fixed".  I don't have a clue as to why.  dict comparison was and
remains linear-time in the size of the dicts, and test_mutants only tries
100 dict pairs, of size averaging just 50.  So "it should" run in less than
an eyeblink; but it takes at least a second on this 800MHz box.

and wrap the body in try/finally to ensure TESTFN gets cleaned up no
matter what.