[Tutor] Objects, object references, object values and memory addresses

[eryksun]

On Thu, Oct 18, 2012 at 1:16 AM, Dave Angel <d at davea.name> wrote:
>
> may decide to reuse existing objects.  This includes, ints, floats,
> strings, byte strings, tuples, etc.  In the particular case of CPython,
> small integers are cached in this way, and so are short strings with no
> whitespace.  How small, and exactly which strings is irrelevant to me,
> and hopefully to you.  The point is you cannot be sure whether equal
> immutable objects are really just a single one, or not.

In case anyone is curious about implementation trivia, CPython (2.7.3
and 3.2.3) caches integers in the closed range -5 to 256. This doesn't
apply to the long() type in 2.x.

Strings for attributes, and various other cases, are interned in a
dict. code objects intern a limited set of string constants
(alphanumeric/underscore ASCII) and also all referenced names
(co_names, co_varnames, co_freevars, and co_cellvars). The built-in
function intern() (moved to sys.intern in 3.x) manually interns a
string, e.g. for a small efficiency gain with dict lookups:

http://docs.python.org/py3k/library/sys#sys.intern
http://docs.python.org/library/functions.html#intern

To give a feel for how much string interning is used, here's the
summary after importing ctypes in a fresh interpreter and releasing
the interned strings:

3.2.3:

    >>> import ctypes
    >>> ctypes.pythonapi._Py_ReleaseInternedUnicodeStrings()
    releasing 3996 interned strings
    total size of all interned strings: 34193/0 mortal/immortal

2.7.3:

    >>> import ctypes
    >>> ctypes.pythonapi._Py_ReleaseInternedStrings()
    releasing 2875 interned strings
    total size of all interned strings: 26389/0 mortal/immortal