Given the behavior, the documentation is gratifyingly correct. Given that the syntax is legal, though, the behavior is not what one would intuitively expect, and is therefore unPythonic by (rather dramatically) violating the principle of least surprise.
It's also, to me, understandable why it's difficult for the language design to avoid this behavior. This little discovery of mine sheds some considerable light on the awkwardness of what you guys will deign to call "declarations". This being the case, I can understand the resistance to "declarations" in Python. I had thought, until the current conversation and this experiment, that the globals statement, er, declaration was just another executable, especially given all the stress on Python's being purely executable. I still see "global" and "@" as expressions of the same fundamental problem, even though decorators are not implemented as declarations. They both take effect in a non-intuitive sequence and they both affect the reference rather than the referent. This points out the problem that Python has in qualifying references rather than referents. Since BDFL is contemplating some optional typing, does this also imply qualifying the references? Maybe you wizard types can agree that there is a useful abstraction that I'm talking about here, whether you wish to call it "declarations" or not, and try to factor out some sort of consistent strategy for dealing with it, perhaps in P3K. (I will try to be in a position to help someday, but I have a long way to go.) Language features that modify references rather than referents appear to be problematic. Python clearly chafes at these. Yet there are at least a few compelling reasons to want them. -- mt -- http://mail.python.org/mailman/listinfo/python-list
