Python-Dev

Download

python-dev@python.org

February 2014

106 participants
77 discussions

Negative times behaviour in itertools.repeat for Python maintenance releases (2.7, 3.3 and maybe 3.4)
by Vajrasky Kok 01 Feb '14

01 Feb '14

Dear comrades, I would like to bring to your attention my disagreement with Larry Hastings in this ticket: http://bugs.python.org/issue19145 (Inconsistent behaviour in itertools.repeat when using negative times). Let me give you the context: >>> from itertools import repeat >>> repeat('a') repeat('a') >>> repeat('a', times=-1) repeat('a') >>> repeat('a', -1) repeat('a', 0) >>> repeat('a', times=-4) repeat('a', -4) >>> repeat('a', -4) repeat('a', 0) Right now, the only way you can tell repeat to do endless repetitions is to omit the `times` argument or by setting `times` argument to -1 via keyword. Larry intends to fix this in Python 3.5 by making None value to `times` argument as a representative of unlimited repetitions and negative `times` argument (either via keyword or positional) ALWAYS means 0 repetitions. This will ensure repeat has the appropriate signature. I have no qualms about it. All is well. My disagreement is related to Larry's decision not to fix this bug in Python 2.7, 3.3, and 3.4. Both of us agree that we should not let Python 2.7, 3.3, and 3.4 happily accepts None value because that is more than bug fix. What we don't agree is whether we should make negative `times` argument via keyword behaviour needs to be changed or not. He prefer let it be. I prefer we change the behaviour so that negative `times` argument in Python 2.7, 3.3, and 3.4 ALWAYS means 0 repetitions. My argument is that, on all circumstances, argument sent to function via positional or keyword must mean the same thing. Let's consider this hypothetical code: # 0 means draw, positive int means win, negative int means lose result = goals_result_of_the_match() # For every goal of the winning match, we donate money to charity. # Every donation consists of 10 $. import itertools itertools.repeat(donate_money_to_charity(), result) Later programmer B refactor this code: # 0 means draw, positive int means win, negative int means lose result = goals_result_of_the_match() # For every goal of the winning match, we donate money to charity. # Every donation consists of 10 $ from itertools import repeat repeat(object=donate_money_to_charity(), times=result) They use Python 2.7 (remember Python 2.7 is here to stay for a long long time). And imagine the match is lost 0-1 or 1-2 or 2-3 (so the goal difference is negative one / -1). It means they donate money to charity endlessly!!! They can go bankrupt. So I hope my argument is convincing enough. We need to fix this bug in Python 2.7, 3.3, and 3.4, by making `times` argument sent via positional or keyword in itertools.repeat ALWAYS means the same thing, which is 0 repetitions. If this is not possible, at the very least, we need to warn this behaviour in the doc. Whatever decision that comes up from this discussion, I will make peace with it. Vajrasky

12 43

Add PyType_GetSlot
by "Martin v. Löwis" 01 Feb '14

01 Feb '14

I'd like to resolve a long-standing issue of the stable ABI in 3.4: http://bugs.python.org/issue17162 The issue is that, since PyTypeObject is opaque, module authors cannot get at tp_free, which they may need to in order to implement tp_dealloc properly. Rather than providing the proposed specific wrapper for tp_dealloc, I propose to add a generic PyType_GetSlot function. From a stability point of view, exposing slot values is uncritical - it's just that the layout of the type object is hidden. Any objection to adding this before RC1? Regards, Martin

2 2