> All in all, I fail to see what gains would be expected by making Python > into a single-assignment or single-binding language, even on a module by > module basis, to repay this kind of awkwardness.
Just to be clear, if anyone was suggesting that, it wasn't me. It would be helpful on occasion in a numarray development context to have *specific* refererences be bound only once, and I wouldn't be surprised if the compiler couldn't use that information to good advantage too. However, this subthread is about whether rebinding is completely avoidable. Others including you have come up with better reasons than I did that it's not. If rebinding is normal, I think the 'epselon' bug can't be dismissed as completely avoidable. This is something that I gather you disagree with on the presumption that everyone who writes Python is sufficently talented that they can use their skills to avoid getting too far into this trap. Since I'm very much a believer in Python as a beginner's language, that doesn't satisfy me. "Declarations are impractical" would satisfy me, but so far I'm not completely convinced of that. mt -- http://mail.python.org/mailman/listinfo/python-list
