The following code succeeds "silently", which seems undesirable:

    from inspect import signature
    def two():
        return 2
    bound = signature(two).bind()
    bound.arguments["does_not_exist"] = 12
    two(*bound.args, **bound.kwargs)

where the mechanism for finally calling `two` is the recommended way shown in the docs for `inspect.BoundArguments`: https://docs.python.org/3/library/inspect.html#inspect.BoundArguments.apply_defaults

What's happened there is that:

    print(b.args, b.kwargs)

"silently" ignored the non-existent argument.

Somewhere along the line here it seems like something should complain. I don't see previous discussion of this from quickly searching on the bug tracker, but obviously if I've missed something let me know.

I'm also not really sure what the desirable solution is. To me, it's possibly that BoundArguments should have a fully-managed way to invoke a callable rather than asking a user to unpack *args and *kwargs, and that that mechanism, say arguments.be_passed_to(callable) should do the error checking). Having `.arguments` full-on reject unknown parameters seems like another possibility but there may be reasons that's not a good idea (e.g. if you were to for some reason use a bound arguments object to call some other function that *did* take that additional argument).

As a secondary behavior here, which is actually the one that matters more for my use case, the following seems surprising as well:

import inspect
s = inspect.signature(lambda **kwargs: kwargs).bind()
s.arguments["foo"] = 12

will similarly silently drop "foo" when attempting to call the function (i.e. it will not appear in bound.kwargs).

This behavior I suspect is the intention of the docs saying "Contains only explicitly bound arguments." but still seems easily misused when .arguments is being used as a frontend for adding additional arguments.

To me just brainstorming again it seems it'd be useful to provide a harder-to-misuse frontend, e.g. `BoundArguments.with_arguments(*args, **kwargs)` that returns a bound arguments with some additional to-be-bound arguments which may end up in `kwargs`.

Bound is created with 5 public attributes:
>>> dir(bound)
[..., 'apply_defaults', 'args', 'arguments', 'kwargs', 'signature']
>>> bound.args
()
>>> bound.arguments
{}
>>> bound.kwargs
{}
msg376578: I don't understand 'non-existent' arguments,  Nor 'what happened...print... ignored' as there is no previous print.

msg376590: Given " Changes in arguments will reflect in args and kwargs.", I agree that changes to 'arguments' *apparently* not being reflected in 'args' and 'kwargs' is initally a bit puzzling
.
>>> bound.kwargs == bound.arguments
True
>>> bound.arguments['something'] = 'guess'
>>> bound.kwargs
{}
>>> bound.arguments
{'something': 'guess'}

However, your 'two' function takes no arguments, so valid values of args and kwargs must be empty for them to be used in a call.  In all cases, args() and kwargs() must look at the signature to see which key-value pairs they should extract from arguments.

>>> def f(a): pass

>>> signature(f).bind()  # Must pass value arguments
Traceback (most recent call last):
...
TypeError: missing a required argument: 'a'
>>> b = signature(f).bind(3)
>>> b.arguments
{'a': 3}
>>> b.args
(3,)  # Because 'a' is positional.
>>> b.kwargs
{}  # Because 'a' is not keyword only.
>>> b.arguments['a']=5
>>> b.args
(5,)  # Legitimate change reflected here.

Perhaps the doc could be improved, but I have no particular suggestion.

Not sure I agree with it being just a doc issue -- happy to clarify if something was unclear, not sure from your message if it was or if you disagree, but e.g.:

> However, your 'two' function takes no arguments, so valid values of args and kwargs must be empty for them to be used in a call.  In all cases, args() and kwargs() must look at the signature to see which key-value pairs they should extract from arguments.

Right, and that's "dangerous" behavior to me -- BoundArguments.arguments will happily let you add arguments to it, but does no error checking at that point, so does not raise an exception if you typo an argument that isn't in the signature of the callable, and then when you try to call the callable (via the only way possible, namely using .args and .kwargs), that argument that was never added is just dropped.

Does that make sense? Or were you disagreeing with that being undesirable behavior?

I use signature for IDLE call tips, but have never used .bind and subsequent calls.  After understanding the behavior, my question as initial reviewer is whether it is an *implementation* bug.  Perhaps, but not obviously so.

Whether or not it is a design wart serious enough to warrant an design change is a different question, which I am not going to judge.  Maybe there is a reason for the current design we have not thought of.  Or maybe it is an accident too painful to change.

I am convinced that the doc could be better.

Totally fair! Sorry, was just making sure the label change wasn't intended to say it *wasn't* enough to warrant a design change :) (which I agree should be discussed with folks who do use that functionality, of which I only recently had to).

If you want to pursue, you might try python-list to find other users.  I am not sure if this change really qualifies for python-ideas.