networkx: fix subgraph and degree

[AckslD]

Two small fixes to networkx, let me know if it's better with separate PRs:

1. Graph.subgraph can also take a single node.
2. If eg degree takes a single node, the return type is different so add overloads.

[AckslD]

Hmm, degree works differently if the first argument is None vs any other Hashable, not sure how to type that.

[srittau]

Thanks, remarks below.

[srittau]

Probably overly pedantic, but this doesn't necessarily return a DiDegreeView, but could also return a sub-class if called on an instance of a sub-class. So Self is more appropriate:

Suggested change

	def __call__(self, nbunch: Iterable[_Node] | None = None, weight: None | bool | str = None) -> DiDegreeView[_Node]: ...
	def __call__(self, nbunch: Iterable[_Node] | None = None, weight: None | bool | str = None) -> Self: ...

[AckslD]

good point, fixed

[srittau]

__getitem__ returns float, so this should also return float:

Suggested change

	def __call__(self, nbunch: _Node, weight: None | bool | str = None) -> int: ...
	def __call__(self, nbunch: _Node, weight: None | bool | str = None) -> float: ...

[AckslD]

hmm shouldn't the degree always be an integer, at least this passes:

import networkx as nx
g = nx.Graph()
g.add_node('a')
assert type(g.degree('a')) is int

[srittau]

I have no experience with networkx, so I can't say either way, I just noticed that __getitem__ return float (which is equivalent to int | float, so I was going by that. If you say that this will always be int, I'll believe you.

[AckslD]

@srittau I tried something when nbunch = None but that also doesn't work, any suggestions how to distinguish that from other hashables?

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

bokeh (https://github.com/bokeh/bokeh)
+ src/bokeh/models/util/structure.py:165: note: ... from here:

[srittau]

@srittau I tried something when nbunch = None but that also doesn't work, any suggestions how to distinguish that from other hashables?

We'll need a # type: ignore[overload-overlap] on the offending line. Unfortunately, mypy's overlap check is sometimes a bit over eager.