Alejandro Candioti

I updated the approach based on @eriknw feedback. I also updated description to mentioned other ideas such as the one we discussed in NX community call!

Friendly ping @eriknw ! What do you think of the current approach?

Happy to wait! I'm also at limited capacity this summer so let's sync back in a couple months!

> Simple data structures are helpful. But I'm not all that consistent in this choice -- sometimes it is better to cache. I think similarly, I believe it is a...

Yeah, let's gather some data on the actual overhead of the current approach. I believe we can implement it in a way that overhead is similar to the one added...

> I haven't looked closely at this yet, but if we're caching something, we should probably consider saving it in `G.__networkx_cache__` so that it can be cleared via [`nx._clear_cache`](https://github.com/networkx/networkx/blob/d1c41f886add2ab86c1ce4c1d788d3c286721351/networkx/utils/misc.py#L593-L599), which...

Let's say I pass a custom object without converting. Then I get this error without giving a chance to the backend to convert my custom object. ```python >>> import networkx...

> I think this is a good discussion to have. > > Yes, your understanding is correct that `.is_directed()` and `.is_multigraph()` are currently _required_ interfaces so that `@not_implemented_for` decorators work....

I put up this PR for the second point https://github.com/networkx/networkx/pull/8035. @eriknw would love to have your input!

!!! ```python (nx-rust-dev) amcandio@penguin:~/workspace/nx-rust$ python Python 3.13.2 | packaged by Anaconda, Inc. | (main, Feb 6 2025, 18:56:02) [GCC 11.2.0] on linux Type "help", "copyright", "credits" or "license" for more...