Zane Bitter <za...@fedoraproject.org> added the comment:
> It feels wrong to do this as a band-aid to help out people who didn't do the > right thing. I mean... are we here to punish people for making mistakes by sending them to schroedinbug purgatory, or are we here to help them upgrade to the latest Python without feeling like it's a minefield? I genuinely don't know what Python's stance on this is. Certainly the Linux kernel and Windows both have a philosophy of not breaking working userspace code no matter how wrong it might be, so it's not unheard-of to choose the latter. I suggested on the PR that we could deprecate the setter so that people who are Doing It Wrong will get a warning that they can much more easily track down and fix, and we won't encourage any wrong new code. Would that change your opinion? > > but the good news is that the values are now cached > That's not relevant to the performance numbers I posted above, is it? Not directly - your performance numbers show that accessing the attribute is ~300ns slower when it's a property. But the thing that gets called all the time that we want to be fast is isEnabledFor(), and the results of that are cached so it only calls getEffectiveLevel() (which is the thing accessing the attribute) once. Note that populating the cache is already considerably more expensive than the previous path (it requires handling an exception and acquiring a lock), but it's still worth it because the cost is typically amortised across many log calls. The addition of an extra property lookup in this path is not going to be noticeable. ---------- _______________________________________ Python tracker <rep...@bugs.python.org> <https://bugs.python.org/issue37857> _______________________________________ _______________________________________________ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
