GeezerVison

Unless the case of the list being empty is handled elsewhere, this could result in a raise None. Maybe that None should instead be something like Exception("Empty MultiSignature list") ?

It is checked in the constructor. I think even about removing this assignment, NameError is clearer indication if something goes wrong.

I agree.

I think this should be

and the addition to Signature's eq should be removed, above.

No, it is not correct, it would make a 1-element MultiSignature equal to any Signature.

You perhaps mistyped, I wrote more correct fix.

I did. That's good. (I would suggest removing the line about MultiSignature in Signature's eq as I believe it's dead code now, but up to you).

It is not dead. It is called in SignatureSubclass() == MultiSignature(). Although it is not covered by tests yet.

Given that the original Signature.from_callable is not changed and does the exact same call to _from_callable, that means both signature and Signature.from_callable may return a MultiSignature object, right ?

Strictly speaking this is not a violation of backwards compatibility as MultiSignature is a subclass of Signature, but I find it weird and it may break some strict-type behaviors. Is there a way to have those only return pure Signature objects, without breaking everything this update brings to the table ?

Yes, it is. The idea was that inspect.signature() will start to support previously not supported cases and return a MultiSignature which should be compatible with Signature. But now I am no longer sure. They are not so well compatible as I expected, and in most cases I needed to rewrite the code to explicitly handle MultiSignature. This is why I added inspect.signatures(). Now the code is in the intermediate state -- either MultiSignature will be made more compatible with Signature and inspect.signatures() may disappear, or they will became completely different things.

It is checked in the constructor. I think even about removing this assignment, NameError is clearer indication if something goes wrong.

No, it is not correct, it would make a 1-element MultiSignature equal to any Signature.

You perhaps mistyped, I wrote more correct fix.

Yes, it is. The idea was that inspect.signature() will start to support previously not supported cases and return a MultiSignature which should be compatible with Signature. But now I am no longer sure. They are not so well compatible as I expected, and in most cases I needed to rewrite the code to explicitly handle MultiSignature. This is why I added inspect.signatures(). Now the code is in the intermediate state -- either MultiSignature will be made more compatible with Signature and inspect.signatures() may disappear, or they will became completely different things.