The __signature__ attribute is an override for the inspect.signature function, which returns the value of that attribute if present instead of computing it.
That behavior is not documented as part of the inspect.signature doc, but it is alluded to in inspect.unwrap, and it is described in the implementation section of the signature PEP.
I think it’s a useful feature that’s not really hard to maintain (unlike the __text_signature__ counterpart), and that it deserves to be documented.
Furthermore, the signature computation is expensive. The PEP states that " the Signature object is created in a lazy manner, and is not automatically cached. However, the user can manually cache a Signature by storing it in the __signature__ attribute." But that’s not true : unless __signature__ is documented, the user is not at liberty to do that.
Seems reasonable. Is there any reason you’ve posted this here, given that you’ve already raised an issue and PR? Unless there’s pushback on the tracker and it’s worth demonstrating community support for the proposal, it’s usually fine just to do the issue/PR.
I consider not documenting a special name a doc bug. I requested reviews from the Signature PEP authors. Hopefully, they will at least suggest someone else. If that does not get a response, use git blame on the inspect doc to see who wrote the mentions therein.
It’s been 10 days, so I checked the git blame : this commit introducing the __signature__ mention in unwrap’s doc was authored by @ncoghlan (hi !), but apparently written by Daniel Urban and Aaron Iles (whom I don’t know the handles, if any).