gh-108901: Add inspect.Signature.from_frame

[sobolevn]

I had to go with a bigger diff, but easier code and probably more optimal one.
The main reason was that creating types.FunctionType is not trivial, for example, it required the correct amount of __closure__ vars for a code object. So, let's not create it: it will reduce the amount of possible errors.

First PR: #112639

• Issue: Add modern alternatives to inspect module, deprecate old incorrect APIs #108901

---

📚 Documentation preview 📚: https://cpython-previews--116537.org.readthedocs.build/

[vstinner]

LGTM

[AlexWaygood]

I haven't really got the bandwidth to look at this right now, sorry :)

[vstinner]

cc @serhiy-storchaka@pitrou

[sobolevn]

I agree, we should drop the defaults part.

>>>importsys, inspect>>>deff(x=1): ... print(inspect.getargvalues(sys._getframe())) ... >>>f() ArgInfo(args=['x'], varargs=None, keywords=None, locals={'x': 1})

Current API does not show defaults, only locals. But, locals are stored as .f_locals, it is a stable API. Users can use .f_locals to populate defaults, it is quite easy. Later we can add defaults support if users ask for it.

Showing defaults is also not safe in this case by default, for example we can leak things like passwords or secret keys.

[vstinner]

I agree, we should drop the defaults part.

I looked at tests where the frame is captured whereas frame locals are not modified, so "it just works" magically. But if I look at the test which modify a variable, I now understand how wrong it is. A frame local is not a function default parameter value: they are two different things. signature() is the signature, before the function is called. Frame locals is the "live state" of a frame, it's unrelated.

Moreover, frame locals can contain sensitive information (like a password) which should not be exposed in a signature.

[sobolevn]

Updated, now defaults are not shown. Right now this is very close to getargvalues:

>>>importsys, inspect>>>deff(x=1): ... print(inspect.getargvalues(sys._getframe())) ... >>>f() ArgInfo(args=['x'], varargs=None, keywords=None, locals={'x': 1})

And we are trying to replace this old function, so I guess it is fine :)

[hugovk]

(This has a merge conflict)

[vstinner]

@sobolevn: Can you please try to fix the merge conflict?

[picnixz]

Suggested change

	.. versionadded:: 3.13
	.. versionadded:: next

[picnixz]

Don't we need the "Contributed by" part? (to have the issue number)