GeezerVison is showing github.com — struct.Struct inheritance with Python 3.12.0 · Issue #112358 · python/cpython
struct.Struct inheritance with Python 3.12.0 #112358DescriptionBug report
Bug description:
import struct
class MyStruct(struct.Struct):
def __init__(self):
super().__init__('>h')
obj = MyStruct()When I run this code I receive an error:
Traceback (most recent call last):
File "/home/user/bug.py", line 7, in <module>
obj = MyStruct()
^^^^^^^^^^
TypeError: Struct() missing required argument 'format' (pos 1)
It is rather strange error, I have passed format parameter to base class constructor, I receive this error with any value of format parameter.
There are no any problems with this code in Python 3.11 and older.
CPython versions tested on:
3.12
Operating systems tested on:
Linux
Linked PRs
ActivityEclips4 commented on Nov 24, 2023 AlexWaygood commented on Nov 24, 2023 Looks like another user also hit this problem, but reported it on the PR rather than opening an issue: #94532 (comment) mdickinson commented on Nov 26, 2023 Hmm, this isn't nice. I'm not seeing any good options here other than simply reverting the offending commit (and re-opening the issues that PR #94532 was supposed to close). @kumaraditya303? mdickinson commented on Nov 26, 2023
Hmm, this isn't nice. [...]
To elaborate: the existence of struct.Struct subclasses that extend the __init__ method means that Struct.__new__ can make essentially no assumptions about the arguments it gets. It might get no arguments; it might get just the format, it might get the format plus additional things, or it might get a collection of arguments that doesn't include the format at all. Previously all those things worked because Struct.__new__ was simply object.__new__, which accepted any combination of args and kwargs.
This makes the solution in #94532 of initialising the struct in Struct.__new__ non-viable - at least, not without backwards compatibility breakage, which would have to be managed in the usual way (deprecation period, etc.).
I've opened #112424 for reversion. Once merged, it'll need to be backported to 3.12, and issues #75960 and #78724 will need to be re-opened. added 3 commits that reference this issue on Nov 26, 2023gh-112358: Fix Python 3.12 regression with subclassing struct.Struct. (… mdickinson commented on Nov 27, 2023 serhiy-storchaka commented on Nov 27, 2023 It is more convenient to initialize the Struct instance in __new__ than in __init__, and it makes sense, since Struct instances are cached and therefore can be considered immutable like ints or tuples. But the possibility of creating subclasses and the existence of subclasses in the wild makes this a breaking change.
Can this breakage be considered a lesser evil? mdickinson commented on Nov 27, 2023 @serhiy-storchaka: Maybe. I'd failed to recognise the scope of the breakage introduced by #94532, and I think reverting in the short term is the right solution. Longer term, if we can find a properly deprecated path to moving the initialisation to __new__, I agree that that would be worth considering.
For the particular issues that #94532 was addressing, I think there are likely less disruptive fixes available. Reacted by Serhiy Storchaka
MetadataMetadataStatusShow more project fields
DevelopmentNo branches or pull requestsIssue actions
|