merge networkx Graph classes from python-type-stubs and address a few recent issues #14597

Avasam · 2025-08-20T17:21:49Z

Extracted from #14038 with a few additional changes. Should reduce changes in that PR

Changes:

Removed int from degree-related returns (fixes networkx: Types for Graph properties are more general than required #14592 )
- Note: This should be fully compatible with @charmoniumQ 's changes in Correct type annotations on NetworkX DiGraphs #14595
Added back Liskov's Substitution Principle considerations from before complete networkx/digraph.pyi addresses #14499 follows #14509 #14569 and Complete networkx/graph.pyi. #14499 #14509
- See networkx: Types for Graph properties are more general than required #14592 (comment)
- I also slightly changed the wording of my comments to make them more obvious they relate to LSP
Moved comments above their related location, rather than under (pretty sure we have an unwritten convention of placing comments on the same line or above)
Removed redundant comments that just spell the annotated return type
Ordered functions to consistently match runtime
Included PR review suggestions from Correct type annotations on NetworkX DiGraphs #14595 (comment) and Correct type annotations on NetworkX DiGraphs #14595 (comment)
- Closes networkx: DiDegreeView doesn't have overload for single node as argument to __call__ #14616
Graphs were inconsistently (as in it was only done sometime) overriding Graph methods with the same signature as their parents. I went with the following consistent approach (I'm fine switching to always overriding methods if present on the specific subclass at runtime, as long as we're consistent here)

NOTE: Graph subclasses relationships are so complex we're only overriding methods that differ in signature from the base classes to use inheritance to our advantage and reduce complexity

Edit: From @Akuli : What tools and process should I use? #14499 (comment)
- You have included various things that are inherited from the base class Graph. In typeshed, we like to repeat things inherited from a base class only if we want something different than the definition in base class (such as the to_undirected method).
So we seem to be on the same track here.

There seems to have been a bit on confusion regarding these stubs recently, let's take the time to properly review these changes, and discuss any uncertainty (whilst I do think this PR is correct, I could be wrong)

CC @srittau @TomerGodinger

… recent issues

stubs/networkx/networkx/classes/multidigraph.pyi

Avasam · 2025-08-20T18:10:07Z

stubs/networkx/networkx/algorithms/planarity.pyi

+    # Overriden in __init__ to always raise
+    def add_edge(self, u_of_edge: _Node, v_of_edge: _Node, **attr: Unused) -> NoReturn: ...
+    def add_edges_from(self, ebunch_to_add: Iterable[_EdgePlus[_Node]], **attr: Unused) -> NoReturn: ...
+    def add_weighted_edges_from(
+        self, ebunch_to_add: Iterable[tuple[_Node, _Node, float]], weight: str = "weight", **attr: Unused
+    ) -> NoReturn: ...


This isn't strictly necessary and I don't mind removing it from the PR if unwanted

…networkx-Graph-classes-from-python-type-stubs

Avasam · 2025-08-20T18:43:17Z

stubs/networkx/networkx/classes/reportviews.pyi

-    def __call__(self, nbunch: None = None, weight: None | bool | str = None) -> int: ...  # type: ignore[overload-overlap]
+    def __call__(self, nbunch: None = None, weight: None | bool | str = None) -> Self: ...
    @overload
-    def __call__(self, nbunch: None | Iterable[_Node], weight: None | bool | str = None) -> Self: ...
+    def __call__(self, nbunch: Iterable[_Node], weight: None | bool | str = None) -> int: ...


#14595 (comment)

the overload should be specialized to _Node.

None and Iterable[_Node] return Self (first and last return), but _Node returns int (middle two returns)

(ignoring, like we have been, that it could actually return float in the case of graphs whose edge weights are float)

Here, Iterable[_Node] is typed as returning int, but I think it should return Self (like None), whereas _Node should return int.

They didn't design this API with typechecking in mind.

(ignoring, like we have been, that it could actually return float in the case of graphs whose edge weights are float)

Considering we typed __getitem__ to return float, it'd be more consistent to return float from self[nbunch] (unless you have a specific reason to not do that)

For reference to future readers, the __call__ method:

def __call__(self, nbunch=None, weight=None): if nbunch is None: if weight == self._weight: return self return self.__class__(self._graph, None, weight) try: if nbunch in self._nodes: if weight == self._weight: return self[nbunch] return self.__class__(self._graph, None, weight)[nbunch] except TypeError: pass return self.__class__(self._graph, nbunch, weight)

I'll also flip the order of overloads in case the hashable node is a str (because of the whole str matches Iterable[str] issue)

it'd be more consistent to return float from self[nbunch] (unless you have a specific reason to not do that)

I was thinking it might break code that does calculation on the degree of nodes, which is always an int. But we have mypy_primer to tell us if that will actually break any code or not.

because of the whole str matches Iterable[str] issue

Ugh, don't remind me /s

Good catch though.

They didn't design this API with typechecking in mind.

Tbf most libraries older than a few years haven't ^^ Especially powerful ones actively using dynamic typing.

But I think we're able to represent this as you've mentioned. My last commit should do that.

it'd be more consistent to return float from self[nbunch] (unless you have a specific reason to not do that)

I was thinking it might break code that does calculation on the degree of nodes, which is always an int. But we have mypy_primer to tell us if that will actually break any code or not.

Looking at the implementation, Pylance/pyright already infers that __getitem__ and __call__ always returns int. That's another good catch.

it could be because int is the fallback to sum with Unknown. But if a DiDegreeView is always meant to be working on degrees, which would be an int, then yeah it should return int

But we have mypy_primer to tell us if that will actually break any code or not.

Unfortunately mypy_primer has a ton of false-negative, because it doesn't force re-enable anything (last I checked).
So if a project excludes files, disables codes, doesn't explicitly checks untyped methods, or doesn't run in strict mode, then the primer won't catch any of those diffs since mypy didn't run on that code (or with the disabled rules)

For instance I've been fixing tons of issues in pywin32 and setuptools 's own projects referencing the stubs from typeshed, and I almost never see the primer telling me I've fixed stuff, because the violations I'm fixing are disabled in said projects due to too many false-positives.

github-actions · 2025-08-20T19:35:06Z

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

discord.py (https://github.com/Rapptz/discord.py)
- discord/ext/commands/hybrid.py:508: error: Overlap between argument names and ** TypedDict items: "description", "name"  [misc]
+ discord/ext/commands/hybrid.py:508: error: Overlap between argument names and ** TypedDict items: "name", "description"  [misc]
- discord/ext/commands/hybrid.py:629: error: Overlap between argument names and ** TypedDict items: "description", "name"  [misc]
+ discord/ext/commands/hybrid.py:629: error: Overlap between argument names and ** TypedDict items: "name", "description"  [misc]

Avasam · 2025-08-20T20:02:35Z

stubs/networkx/networkx/classes/reportviews.pyi

+    @overload  # Use this overload first in case _Node=str, since `str` matches `Iterable[str]`
+    def __call__(self, nbunch: _Node, weight: None | bool | str = None) -> int: ...  # type: ignore[overload-overlap]
    @overload
-    def __call__(self, nbunch: None = None, weight: None | bool | str = None) -> int: ...  # type: ignore[overload-overlap]
-    @overload
-    def __call__(self, nbunch: None | Iterable[_Node], weight: None | bool | str = None) -> Self: ...
-    def __getitem__(self, n: _Node) -> float: ...
-    def __iter__(self) -> Iterator[tuple[_Node, float]]: ...
+    def __call__(self, nbunch: Iterable[_Node] | None = None, weight: None | bool | str = None) -> Self: ...
+    def __getitem__(self, n: _Node) -> int: ...
+    def __iter__(self) -> Iterator[tuple[_Node, int]]: ...


I've included this file because of #14595 (comment)

But if reviewer is uncertain about these changes, I can split it off

hamdanal · 2025-08-20T20:09:36Z

stubs/networkx/networkx/classes/reportviews.pyi

@@ -55,12 +58,12 @@ class NodeDataView(AbstractSet[_Node]):

 class DiDegreeView(Generic[_Node]):
    def __init__(self, G: Graph[_Node], nbunch: _NBunch[_Node] = None, weight: None | bool | str = None) -> None: ...
+    @overload  # Use this overload first in case _Node=str, since `str` matches `Iterable[str]`


Maybe this is worth a test in check_tricky_function_params.py?

Maybe. I was also pondering if I should add a test or if a comment is enough.

I guess both don't hurt.

merge networkx Graph classes from python-type-stubs and address a few…

d43edd1

… recent issues

This comment has been minimized.

Sign in to view

charmoniumQ reviewed Aug 20, 2025

View reviewed changes

stubs/networkx/networkx/classes/multidigraph.pyi Outdated Show resolved Hide resolved

in_edges and stubtest

cb4b103

Avasam commented Aug 20, 2025

View reviewed changes

This comment has been minimized.

Sign in to view

Avasam mentioned this pull request Aug 20, 2025

Correct type annotations on NetworkX DiGraphs #14595

Merged

Avasam added 2 commits August 20, 2025 14:31

Merge branch 'main' of https://github.com/python/typeshed into merge-…

f8bf5e1

…networkx-Graph-classes-from-python-type-stubs

Adjust NodeView and DiDegreeView overloads

9369176

Avasam requested a review from charmoniumQ August 20, 2025 18:40

Avasam commented Aug 20, 2025

View reviewed changes

This comment has been minimized.

Sign in to view

Better call overload on DiDegreeView

c3e9fbc

This comment has been minimized.

Sign in to view

degrees are integer

2c5d1d8

Avasam commented Aug 20, 2025

View reviewed changes

hamdanal reviewed Aug 20, 2025

View reviewed changes

charmoniumQ approved these changes Aug 20, 2025

View reviewed changes

jenshnielsen mentioned this pull request Aug 21, 2025

Bump types-networkx from 3.5.0.20250812 to 3.5.0.20250821 microsoft/Qcodes#7409

Closed

brianschubert mentioned this pull request Aug 21, 2025

networkx: DiDegreeView doesn't have overload for single node as argument to __call__ #14616

Open

Uh oh!

merge networkx Graph classes from python-type-stubs and address a few recent issues #14597

Are you sure you want to change the base?

merge networkx Graph classes from python-type-stubs and address a few recent issues #14597

Conversation

Avasam commented Aug 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

Uh oh!

Avasam Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

Avasam Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

charmoniumQ Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

Avasam Aug 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Avasam Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

charmoniumQ Aug 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Avasam Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

Avasam Aug 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Avasam Aug 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

This comment has been minimized.

github-actions bot commented Aug 20, 2025

Uh oh!

Avasam Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

hamdanal Aug 20, 2025

Choose a reason for hiding this comment

Uh oh!

Avasam Aug 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Avasam commented Aug 20, 2025 •

edited

Loading

Avasam Aug 20, 2025 •

edited

Loading

charmoniumQ Aug 20, 2025 •

edited

Loading

Avasam Aug 20, 2025 •

edited

Loading

Avasam Aug 20, 2025 •

edited

Loading

Avasam Aug 20, 2025 •

edited

Loading