René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Unpickled Git object has broken version_info","articleBody":"A `Git` instance produced by pickle deserialization always has a `version_info` of `None`. This prevents version information from being accessed through `version_info`. It also causes the `version_info` property not to satisfy its type annotation as a tuple of `int`s (independently of #1830). This is the case whether or not `version_info` has been accessed on the original instance before it was pickled:\r\n\r\n```python\r\n\u003e\u003e\u003e import pickle, git\r\n\u003e\u003e\u003e g = git.Git()\r\n\u003e\u003e\u003e g.version_info\r\n(2, 34, 1)\r\n\u003e\u003e\u003e g2 = pickle.loads(pickle.dumps(g))\r\n\u003e\u003e\u003e g2.version_info\r\n\u003e\u003e\u003e\r\n```\r\n\r\n```python\r\n\u003e\u003e\u003e import pickle, git\r\n\u003e\u003e\u003e g = git.Git()\r\n\u003e\u003e\u003e g2 = pickle.loads(pickle.dumps(g))\r\n\u003e\u003e\u003e g2.version_info\r\n\u003e\u003e\u003e\r\n```\r\n\r\nThis is due to an inconsistency in how the state of not yet having computed `version_info` is represented in the backing attribute `_version_info`:\r\n\r\n- `Git` uses `LazyMixin` caching for `_version_info`, which treats the attribute as uncomputed when absent and computed when present.\r\n- `Git` defines `__getstate__` and `__setstate__` for pickling and unpickling to represent state pickling does not preserve by setting attributes to `None`.\r\n\r\nThe `_version_info` attribute is set to `None` on a `Git` object created by unpickling, so when accessing the `version_info` property delegates to the `_version_info` attribute, its value of `None` is returned and the `if attr == \"_version_info\"` logic in the `_set_cache_` method (inherited from `LazyMixin` and overridden) is never run.\r\n\r\nThe intention is that an unpickled `Git` object have `version_info` in the uncomputed state. This works for the other two excluded attributes, which use the same `None` convention as in the `__getstate__` and `__setstate__`  implementations, and which do not use any `LazyMixin` functionality for their caching:\r\n\r\nhttps://github.com/gitpython-developers/GitPython/blob/afa575454f85b34800460881cc9c3cd7fe78c8e3/git/cmd.py#L315\r\n\r\nhttps://github.com/gitpython-developers/GitPython/blob/afa575454f85b34800460881cc9c3cd7fe78c8e3/git/cmd.py#L319-L323\r\n\r\nUnlike `_version_info`, which is managed by `_set_cache_`, the `cat_file_all` and `cat_file_header` attributes are initialized to `None` in `Git.__init__`:\r\n\r\nhttps://github.com/gitpython-developers/GitPython/blob/afa575454f85b34800460881cc9c3cd7fe78c8e3/git/cmd.py#L786-L788\r\n\r\nThis is natural to fix at the same time as #1829, because a fix for that should change the way caching works. `LazyMixin` caching supports attributes that are computed once and never become stale, so `version_info`/`_version_info` should implement caching in a different way to fix #1829, and that can fix this too if consistent with `__getstate__`/`__setstate__`.","author":{"url":"https://github.com/EliahKagan","@type":"Person","name":"EliahKagan"},"datePublished":"2024-02-22T03:45:45.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":0},"url":"https://github.com/1836/GitPython/issues/1836"}