[Python-3000] Is reference counting still needed?

[Thomas Heller]

Greg Ewing wrote:
> Tim Peters wrote:
> 
>> Yes, but I don't think anyone's offering to build a P3K chip for us ;-)
> 
> Since you mention it, one of the things that's been
> occupying my mind in idle moments recently is trying to
> imagine what a Python chip *would* be like, and maybe
> even designing it and prototyping it with an FPGA.

Since MyHDL now exists, it may even be possible to write and test
the prototype in Python.

> Whether the project will ever get that far is highly
> uncertain yet, but it's an interesting thing to think
> about. Some of the ideas I've been considering include
> keeping all the object headers, including refcount,
> size, type pointer and maybe some flags, in a separate
> specialised memory with its own cache and maybe its own
> bus.

Or even in the FPGA's own memory.

> Another idea is just keeping the low-order byte of the
> refcount in a special high-speed memory, so that you
> only need to touch main memory for about 1 in 256
> inc/decrefs.
> 
> This is getting off-topic for Py3k, though. Any hardware
> buffs interested in discussing these ideas further are
> welcome to contact me by email.

Yes.  I'll shut up now.


Thomas