René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Revert  __signature__ to being a cache, remove string and callable support","articleBody":"# Feature or enhancement\r\n\r\n### Proposal:\r\n\r\n#100039 modified the behavior of `__signature__` as used by the inspect.signature function :\r\n- if the fields contains a Signature object, it is returned by the function (that doesn't change)\r\n- if it contains None, it is ignored (that doesn't change)\r\n- if it contains a string, it is converted to a signature (new)\r\n- if it contains a callable, it is called with no parameters and expected to return either a signature object or a string which will be treated as the above (new)\r\n- if it contains any other value, a TypeError is raised (this used to include strings and callables, it doesn't anymore)\r\n\r\n**As for strings :**\r\nThe `__text_signature__` attribute's purpose is to pass a string to inform the signature. It's also only used for C-written functions, and iirc there's no way to use it in pure python : I understand that, but if the ability to specify a text override of inspect.signature must happen, then it should go through `__text_signature__` and not `__signature__`.\r\n\r\n**As for callables :**\r\nThe Signature object qualifies, and the signature function inspects, callables. So, if when asked for the signature of a given callable you return another callable, it would make more sense to consider the second callable's signature as the signature of the first callable, than considering the second callable to _return_ the signature of the first. And why no parameters ? Why not pass the first callable as parameter ? Or the actual signature of the first callable ? That implementation raises a lot of questions of arbitrariness, which makes it a bad design in my opinion.\r\n\r\n**As for the purpose :**\r\nThe use case, which was adding signature overrides to Enums, does not require any change to the inspect module, in fact a perfectly good alternative implementation can be found in #115937. It should also make for a considerably faster execution.\r\n\r\nI suggest reverting these changes.\r\n\r\n### Links to previous discussion of this feature:\r\n\r\nThis was discussed in #115937 and #116086 in which it was advised to separate the question of the implementation from that of the documentation.\n\n\u003c!-- gh-linked-prs --\u003e\n### Linked PRs\n* gh-116234\n\u003c!-- /gh-linked-prs --\u003e\n","author":{"url":"https://github.com/Gouvernathor","@type":"Person","name":"Gouvernathor"},"datePublished":"2024-02-29T10:49:56.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":10},"url":"https://github.com/116110/cpython/issues/116110"}