gh-107431: Make multiprocessing.managers.{DictProxy,ListProxy} generic

[sobolevn]

This looks like a reasonable use-case for generic types.
In typeshed they are generic:

1. https://github.com/python/typeshed/blob/0d8a6bc2004b7b6d8335e35b017438125c305264/stdlib/multiprocessing/managers.pyi#L68
2. https://github.com/python/typeshed/blob/0d8a6bc2004b7b6d8335e35b017438125c305264/stdlib/multiprocessing/managers.pyi#L122

• Issue: multiprocessing manager classes DictProxy and ListProxy don't support typing in 3.11.4 #107431

Fixes #107431.

[AlexWaygood]

Some small points:

[AlexWaygood]

Thanks, LGTM! Will wait to see if Jelle has any objections before merging.

Previous additions of __class_getitem__ methods have been treated as features, so we probably can't backport this, unfortunately.

[AlexWaygood]

Hmm... I suppose one reason for not making these subscriptable at runtime might be that these are undocumented classes that aren't included in __all__. But I personally still lean towards making them subscriptable, nonetheless.

[JelleZijlstra]

Maybe these classes should be documented? I'm also a bit hesitant about this if these classes are undocumented.

[AlexWaygood]

Maybe these classes should be documented? I'm also a bit hesitant about this if these classes are undocumented.

@gpshead, any thoughts from you on this, as a multiprocessing expert? I think there's a decent case for documenting these classes, since instances of ListProxy and DictProxy are returned from multiprocessing.managers.SyncManager.list() and multiprocessing.managers.SyncManager.dict() respectively. Both of those are public, documented methods:

• https://docs.python.org/3/library/multiprocessing.html#multiprocessing.managers.SyncManager.list
• https://docs.python.org/3/library/multiprocessing.html#multiprocessing.managers.SyncManager.dict

[vstinner]

LGTM, but i'm not a typing expert.

In case of doubt, I prefer to not document DictProxy and ListProxy. I don't think that not documenting these types should block this enhancement.

[vstinner]

@JelleZijlstra: What do you think?

@gvanrossum, @carljm (typing experts): Do you want to review this typing change?

[AlexWaygood]

@vstinner, this needs input from multiprocessing experts, not typing experts. Two typing experts have already given their input here (Jelle and myself). From a typing perspective, the change is fine. But from a multiprocessing perspective: the classes are currently undocumented and not included in __all__. Adding __class_getitem__ might encourage their use in type annotations. Is that okay? Should we document these classes, or are they best left undocumented? What's our position on that?

[vstinner]

@pitrou: Do you have an opinion on this change?

[gpshead]

I don't think anyone wants to claim expertise for multiprocessing no matter how much we've done work on it. This change looks good to me, I made a couple edits.

[pitrou]

I have no idea what a GenericAlias is, and the doc is rather cryptic. As long as this doesn't change runtime behavior I see no reason to oppose it.

[vstinner]

Thanks for making _BaseDictProxy private.