Misc questions about type objects and Python 3.0

David Brown david at no.westcontrol.spam.com
Tue Oct 8 06:38:01 EDT 2002


>
> 1. Is it the plan that Python 3.0 be binary compatible with Python
>    2.x?
>
> 2. If it isn't (I'm guessing it will be), is it the plan to keep
>    reference counting?  It seems to me that getting rid of reference
>    counting might be the first step to getting rid of the GIL.

First, I would think that removing reference counting would wildly change
the way Python works (correct me if I'm wrong here - I program with Python,
but I haven't studied its implementation).  I don't see how you can get
efficient garbage collection without some sort of reference counting.

Second, why do you want to get rid of the GIL?  There was a thread a while
back (I know, because I asked the question) about threading and the GIL, and
as far as I learned from the thread, there are no real disadvantages of the
GIL (and many big advantages) except on heavily cpu-instensive python code
running on SMP computers where the underlying C routines don't release the
GIL (well-behaved time-consuming C functions do release the GIL).

>
> 3. The docs say types choose their member resolution order (?), but
>    I'm wondering if that's just a temporary hack to support both
>    old-style and new-style classes.  After old-style classes are
>    dropped, will MRO be fixed for all types?  (And is it really that
>    useful for it to be customizable?)
>
> 4. What is the point of having both tp_base and tp_bases in the type
>    structure?
>
> 5. What's the deal with unification of numeric types?  What's expected
>    or planned or hoped for in 3.0?
>
>
> --
> CARL BANKS
> http://www.aerojockey.com





More information about the Python-list mailing list