[Python-Dev] PEP 329: Treating Builtins as Constants inthe Standard Library

[Phillip J. Eby]

At 05:23 PM 4/19/04 -0700, Guido van Rossum wrote:
> > This suggests that there should be some way of disabling
> > the optimisation from *outside* the module. Not sure what
> > that would be, though.
>
>A command line flag?

For that to work, there'd need to be some place to store the names of the 
modules to which one wished to apply this "pessimization".  That being the 
case, one might as well expose an API to Python for that, rather than 
forcing the issue up to the command line and thereby reducing modularity, 
as well as possibly being an issue for embedders, Jython, et al.

Anyway, the API would give you a function to call if you want to disable 
optimization for a module in order to monkeypatch it.  That function would 
then raise an error if the module in question was already imported with 
optimization enabled, or if the runtime environment disallows such changes 
(e.g. if Jython, IronPython, or various hypothetical 
Python-to-static-language translators want to disallow monkeypatching).

One possibly interesting use case is Pyrex, which would like to be able to 
disallow modifications to built-ins in order to translate e.g. 'len()' into 
C.  It would be nice if there's a way for C modules to "play nice" with the 
mechanisms under discussion.  Specifically, if there's a way for them to 
declare metadata of the same sort that's contemplated for indicating what 
attributes of a module are modifiable from outside.