[Python-Dev] redefining is

[Tim Peters]

[Guido]
>> Sorry, if you're usign *any* immutable value there and expecting it
>> to be a unique object, you're cruisin' for a bruisin', so to speak.
>> The language spec explicitly *allows* but does not *require* the
>> implementation to cache and reuse immutable values.

[Andrew]
> Ay, there's the rub.
>
> Aren't you saying that using "is" to compare immutables is always
> broken, unless you know that the immutable values are singletons?

No, because it's still <wink> the purpose of "is" to expose the
implementation.  Examples of system-level uses were given earlier; for
another, if I'm doing a crawl over gc.get_referrers(x) trying to account for
all the references to the object currently bound to x (this is something I
do often when trying to track suspected memory leaks), object identity is
the only notion of equivalence that's relevant, and it doesn't matter
whether that object is a singleton, or mutable, neither any other property
of x's type or behavior -- the only thing about x that matters in this
context is its indenty, and locating which other objects refer to exactly
it.  If I know that references to the string "<oid at 00000001>" are growing
over time, I'll in fact add an "is" test against that string to the guts of
Zope's test.py leak-detection scaffolding.  It's necessary to be able to
spell this in Python -- although it's not necessary (IMO) to be able to
spell it with two letters.