Thanks for the reply and analysis, Mark.

My motivation was that as a "clueless user", I shouldn't worry about how Decimals are created. Given two equal numbers, I would expect their behavior (e.g., result of a hash) to be the same as well. In this example, the behavior differs simply based on whether the Decimal was created from a string vs a float.

However, since there are bound to be collisions, and the performance overhead is negligible (in case of a collision), and Python 3 solves this problem I agree this can be closed as "won't-fix".