[Python-Dev] Drop support for ones' complement machines?

[Mark Dickinson]

[Mark]
> I'm not sure how to decide which particular consequences
> should be listed, but those for &, |, ^ and ~ could certainly
> be included.

[Martin]
> It should give the CPython contributors an indication what kind
> of code would be ok, and which would not. Perhaps it should include
> both a black list and a white list: some may assume that two's
> complement already provides guarantees on left-shift, when it
> actually does not (**).

Okay.  I'll have to think about this a bit;  I'll try to come up with
some suitable wording.

> (*) I wonder why you are not talking about padding bits
> (6.2.6.2p1)

Good point.  Mostly because I haven't recently encountered any
code where it matters, I suppose.  But there's certainly CPython
source that assumes no padding bits:  long_hash in longobject.c
is one example that comes to mind:  it assumes that the number
of value bits in an unsigned long is 8*SIZEOF_LONG.

> (**) I also wonder why C fails to make left-shift
> implementation-defined, perhaps with an even stronger binding
> to the options for the integer representation.

I wonder too.  The C rationale document doesn't
have anything to say on this subject.

Mark