[Python-Dev] gc ideas -- sparse memory

[Dima Tisnek]

On 3 December 2010 16:45, "Martin v. Löwis" <martin at v.loewis.de> wrote:
>> Oh my bad, I must've confused python with some research paper.
>> Unique id is not so hard to make without an address.
>>
>> While on this topic, what is the real need for unique ids?
>
> They are absolutely needed for mutable objects. For immutable ones,
> it would be ok to claim that they are identical if they are equal
> (assuming they support equality - which is tricky for things like NaN).

Indeed, but do ids really need to be unique and fixed at the same time?

a is b # (if atomic) needs unique ids, but doesn't really need fixed ids
a[b]   # needs fixed hash, but not strictly a globally unique id

I can imagine an implementaion of pickle for example that uses unique
and fixed as a given to detect cycles, etc; but that would be
implementation detail.

It seems to me unique and fixed id implies that it is stored somewhere
(with incref beforehands and decref afterwards), however a proper
reference to an object could be used just as well.

Am I still missing something?

>
> Of course, the C API has lots of assumptions that identity and address
> are really the same thing.
>
> Regards,
> Martin
>