Prior to this change, Injector.sub() instances would provide an 'injector' service if one had been created by instantiating Injector or one of it's bases with provide_self=True. This was the wrong injector, but if it wasn't closed, code may have executed using the other injector without breaking. Now Injector.sub() instances will have no injector, because Injector.sub() uses it's arguments to customize the subinjector.

We need another way to allow access to the injector in this case. Here are some ideas:

1. Revise the Injector.sub() API to allow provide_self to be specified in some way (maybe just a reserved kwarg).
2. Allow the provide_self bit to be configured on the injector class.
3. Stop pretending we can provide a sandbox in python and provide the injector always.

My preference is for 3. It might seem radical, but I really don't think it is.

Thanks for this. We can go with 3 given how subinjectors work.

I'll come up with a patch.

I see the docs need to be updated. I'm working on that now.

So now we don't support provide_self=False with subinjectors. Should we just deprecate provide_self=False?

Yes.

I've merely omitted any discussion of Injector(provide_self=False) in the Injector docstring.

Should we document the deprecation or add something like this?

    if not provide_self:
        warnings.warn(DeprecationWarning('provide_self=False is not supported'))

DeprecationWarning should be enough.