[Python-3000] Making more effective use of slice objects in Py3k

[Guido van Rossum]

On 8/29/06, Jim Jewett <jimjjewett at gmail.com> wrote:
> On 8/29/06, Guido van Rossum <guido at python.org> wrote:
> > On 8/29/06, Josiah Carlson <jcarlson at uci.edu> wrote:
> > > "Guido van Rossum" <guido at python.org> wrote:
>
> > The type consistency and predictability is more important to me.
>
> Why is it essential that string views be a different type, rather than
> an internal implementation detail, like long vs int?  Today's strings
> can already return a new object or an existing one which happens to be
> equal.
>
> Is this just a matter of efficiency, or are you making a fundamental
> distinction?

Sigh. Josiah just said he wouldn't dream of proposing that all string
ops should return string views. You're not helping by questioning even
that.

The short answer is, if you don't have control over when a view on an
existing string is returned and when a copy, there are easy to see
worst-case behaviors that are worse than the problem they are trying
to fix. For example, you'd get a whole series of problems like this
one:

res = []
for i in range(1000):
  s = " "*1000000 # a new 1MB string
  res.append(s[:1]) # a one-character string that is a view on s and
hence keeps s alive

if s[:1] were to return a view on s unconditionally the above loop
would accumumate roughly 1 GB in wasted space.

To fix this you'll have to add heuristics and all sorts of other
things and that will complicate the string implementation and hence
slow it down.

-- 
--Guido van Rossum (home page: http://www.python.org/~guido/)