Though slightly slower, has better clarity and teaching value.

superficial errors and one deep one that aren't currently caught.  I'm
headed for bed after this checkin.

- Fixed several typos introduced by Raymond Hettinger (through
  cut-n-paste from my template): it's _as_temporarily_immutable, not
  _as_temporary_immutable, and moreover when the element is added, we
  should use _as_immutable.

- Made the seq argument to ImmutableSet.__init__ optional, so we can
  write ImmutableSet() to create an immutable empty set.

- Rename the seq argument to Set and ImmutableSet to iterable.

- Add a Set.__hash__ method that raises a TypeError.  We inherit a
  default __hash__ implementation from object, and we don't want that.
  We can then catch this in update(), so that
  e.g. s.update([Set([1])]) will transform the Set([1]) to
  ImmutableSet([1]).

- Added the dance to catch TypeError and try _as_immutable in the
  constructors too (by calling _update()).  This is needed so that
  Set([Set([1])]) is correctly interpreted as
  Set([ImmutableSet([1])]).  (I was puzzled by a side effect of this
  and the inherited __hash__ when comparing two sets of sets while
  testing different powerset implementations: the Set element passed
  to a Set constructor wasn't transformed to an ImmutableSet, and then
  the dictionary didn't believe the Set found in one dict it was the
  same as ImmutableSet in the other, because the hashes were
  different.)

- Refactored Set.update() and both __init__() methods; moved the body
  of update() into BaseSet as _update(), and call this from __init__()
  and update().

- Changed the NotImplementedError in BaseSet.__init__ to TypeError,
  both for consistency with basestring() and because we have to use
  TypeError when denying Set.__hash__.  Together those provide
  sufficient evidence that an unimplemented method needs to raise
  TypeError.

possible API improvements.

Gains a 5:1 speed-up for membership testing by
handling the most common case first (the case
where the element is hashable).

Closes SF Patch 597444.

tiny amount of code duplication, but makes it possible to give BaseSet
an __init__ that raises an exception.

received feedback that was based in the misunderstanding that sets
were sequences.

This patch causes CGIHTTPServer to update os.environ regardless of how
it tries to handle calls (fork, popen*, etc.).

Backport bugfix candidate.

SHIFT and MASK, and widen digit.  One problem is that code of the form

    digit << small_integer

implicitly assumes that the result fits in an int or unsigned int
(platform-dependent, but "int sized" in any case), since digit is
promoted "just" to int or unsigned via the usual integer promotions.
But if digit is typedef'ed as unsigned int, this loses information.
The cure for this is just to cast digit to twodigits first.

These were reported and fixed by Inyeol Lee in SF bug 595350.  The
endswith() bug was already fixed in 2.3, but this adds some more test
cases.

2045, section 5.2 states that if the Content-Type: header is
syntactically invalid, the default type should be text/plain.
Implement minimal sanity checking of the header -- it must have
exactly one slash in it.  This closes SF patch #597593 by Skip, but in
a different way.

Note that these methods used to raise ValueError for invalid ctypes,
but now they won't.

get the MIME main and sub types, instead of getting the whole ctype
and splitting it here.   The two more specific methods now correctly
implement RFC 2045, section 5.2.

signed vs unsigned).

Closes SF bug #597177.

email 2.2 but fails in email 1.0.

(There's a link to PEP218; has PEP218 been updated to match the actual
module implementation?)

   the list is getting awfully long
Mention Karatsuba multiplication and some other items

dict.items/keys/values/iteritems/iterkeys/itervalues().

interning.  I modified Oren's patch significantly, but the basic idea
and most of the implementation is unchanged.  Interned strings created
with PyString_InternInPlace() are now mortal, and you must keep a
reference to the resulting string around; use the new function
PyString_InternImmortal() to create immortal interned strings.

in LOAD_GLOBAL.  Besides saving a C function call, it saves checks
whether f_globals and f_builtins are dicts, and extracting and testing
the string object's hash code is done only once.  We bail out of the
inlining if the name is not exactly a string, or when its hash is -1;
because of interning, neither should ever happen.  I believe interning
guarantees that the hash code is set, and I believe that the 'names'
tuple of a code object always contains interned strings, but I'm not
assuming that -- I'm simply testing hash != -1.

On my home machine, this makes a pystone variant with new-style
classes and slots run at the same speed as classic pystone!  (With
new-style classes but without slots, it is still a lot slower.)

comments everywhere that bugged me: /* Foo is inlined */ instead of
/* Inline Foo */.  Somehow the "is inlined" phrase always confused me
for half a second (thinking, "No it isn't" until I added the missing
"here").  The new phrase is hopefully unambiguous.

to _PyType_Lookup().

Should save 4% on slot lookups.

Move some debugging checks inside Py_DEBUG.

They were causing cache misses according to cachegrind.

This causes a modest speedup.

convert the doc strings to LaTeX, be my guest.)

expensive and overly general PyObject_IsInstance(), call
PyObject_TypeCheck() which is a macro that often avoids a call, and if
it does make a call, calls the much more efficient PyType_IsSubtype().
This saved 6% on a benchmark for slot lookups.