[Python-Dev] Python-3.0, unicode, and os.environ

[Adam Olsen]

On Fri, Dec 5, 2008 at 12:00 AM, "Martin v. Löwis" <martin at v.loewis.de> wrote:
>> Please, if you have a *new* idea that doesn't have a failure mode, by
>> all means post it.  But don't resurrect a pointless bikeshed.
>
> While I completely agree that it is pointless to reiterate the same
> arguments over and over, I disagree that the bikeshed metapher applies.
> This metapher (IIUC) describes a trivial design issue that is merely
> a matter of taste, rather than having deep technical implications.
> Using Unicode or bytes for strings is not of that kind.

That we need to support both unicode and bytes is important, but
already seems to have consensus.  However, they present two distinct
usage patterns:

* unicode text, presentable to the user, interacts with all manor of
standardized APIs
* bytes, limited to local, internal use.  Only approximated forms can
be presented to the user, only custom formats can be saved externally

None of the proposals have turned these into a single use case.  All
they do is trade off various forms of subtly switch back and forth,
which leads to failure.  Debating which subtle failure is better is a
bikeshed.

Not only that, but we already have a solution that makes the choice
explicit, avoiding the subtle failure.  This is the solution already
in use for os file & path functions.  It's the solution Guido
supports.


-- 
Adam Olsen, aka Rhamphoryncus