One last shot at the Augmented Assignment PEP

[Alex Martelli]

"Fredrik Lundh" <effbot at telia.com> wrote in message
news:TQbw5.367$n4.28461 at newsc.telia.net...
> warren wrote:
> > I've always been tempted to tell people to think of "=" as binding, not
> > assignment, but even in versions <= 2.0, it didn't always mean that:
> >
> > a = [0]
> > a[0] = 1
> >
> > No symbol rebinding here, although we have copied a reference.
>
> well, it's only binding if the assignment target is
> a simple name.

For me, at least, my mental model of what's going on works better if
I say that it's binding anyway -- i.e., that you're still binding (or
re-binding) a *slot*, whether that slot is "simple-name"'d or not.


> all other assignments are just nice ways to write
> certain method calls:
>
>     a[0] = 1
>     a.__setitem__(0, 1)
>
>     a.x = 1
>     a.__setattr__("x", 1)

And (although this is not guaranteed to work in the future...):

a = 1
globals().__setitem__('a', 1)

No major difference.  "Simple-names" are clearly more subject
to present or future optimization (and thus, a little magic
behind the covers) -- so locals() is readonly (and writing to
it may not really DO anything...), and globals() may one day
be somewhat optimized too, etc.  But I do not see these (quite
worthwhile) optimization-possibilities as breaking the chance
to reason about a mental model of runtime execution in terms
of slot being bound/unbound/re-bound to objects, and names just
being one (handy) way to navigate to certain of these slots.


Alex