Hmm....

• if list(dict.keys())[0] is broken
• list(dict)[0] will be broken too
• next(iter(dict)) will be broken too
• and the only hope would be to order these dicts or deprecate Python 2 - 3.5 soon

In T220099#5084934, @Dvorapa wrote:

Hmm....

• and the only hope would be to order these dicts or deprecate Python 2 soon

I just looked for the token .keys())[ but you are right there might be other problematic issues too. Unfortunately the difference is between Python 3.6/3.7+ and previous releases but I also don't know whether the results of 2.7 - 3.5 are equal.

The solution is using OrderedDict in such cases for Python < 3.7.

In T220099#5084979, @Xqt wrote:

Unfortunately the difference is between Python 3.6/3.7+ and previous releases but I also don't know whether the results of 2.7 - 3.5 are equal.

Perhaps they are (I tried py2.7 and 3.5 now), but the order is random for each dict:

>>> d={'en': 'hi', 'de': 'hallo', 'cs': 'ahoj', 'pl': 'czeszcz'}
>>> d
{'cs': 'ahoj', 'de': 'hallo', 'en': 'hi', 'pl': 'czeszcz'}
>>> d={'banana': 20, 'apple': 40, 'coconut': 60, 'apricot': 1}
>>> d
{'apricot': 1, 'coconut': 60, 'banana': 20, 'apple': 40}
>>> d={'banana': 20, 'apple': 40, 'coconut': 60, 'apricot': 1, 'air': 0, 'cherry': 55}
>>> d
{'coconut': 60, 'apple': 40, 'apricot': 1, 'air': 0, 'cherry': 55, 'banana': 20}

We do not know, what the first key is in py2 - 3.5. And it changes every time the dict is changed (e.g. new translation is added to the code)

For i18n.translate, I would avoid using first key as a fallback at all. We would need to move all the translation dicts to ordered because of this fallback. Is that fallback necessary? If fallback = False I would fail/return None if code was not found. If fallback = True and dict doesn't have _default or en, I would fail that dict is purely written.

In T220099#5085013, @Dvorapa wrote:

For i18n.translate, I would avoid using first key as a fallback at all. We would need to move all the translation dicts to ordered because of this fallback. Is that fallback necessary? If fallback = False I would fail/return None if code was not found. If fallback = True and dict doesn't have _default or en, I would fail that dict is purely written.

I agree with your proposal for this case. In i18n.translate it should not be necessary to use an unsprecific value (see the doc: If none of the options gives result, we just take the one language from xdict which may not be always the same - which is weird). Normally i18n.translate should never have a fallback because this is a L10N and not a i18n method. If a fallback is made I guess this could be moved to twn then.

Sure and for this we have T110856

Change 589557 had a related patch set uploaded (by Xqt; owner: Xqt):
[pywikibot/core@master] [bugfix] Do not return a random i18n.translation() result

https://gerrit.wikimedia.org/r/589557

Change 589557 merged by jenkins-bot:
[pywikibot/core@master] [bugfix] Do not return a random i18n.translation() result

https://gerrit.wikimedia.org/r/589557

list(dict.keys()) is not deterministic

list(dict.values()), list(dict.items()), dict.popitem(), list(set), set.pop() as well. In general, any iteration over a dict or set is non-deterministic.

Unfortunately the difference is between Python 3.6/3.7+ and previous releases but I also don't know whether the results of 2.7 - 3.5 are equal.

<3.3 and 3.3-3.5 have different results due to "hash randomization" added in 3.3 (see the second gray box).