Bad error message for weighted adjacency matrix

[kcrisman]

H=Graph({0:[1,2,3], 4:[0,2], 6:[1,2,3,4,5]})
H.weighted(True)
H.weighted_adjacency_matrix()

should give something useful, but instead is an inexplicable error message

TypeError: NoneType takes no arguments

This happens whether or not this graph is said to be weighted. There should be an error catch for a useful message.

Component: graph theory

Author: David Coudert

Branch/Commit: fe8b4d5

Reviewer: Dima Pasechnik

Issue created by migration from https://trac.sagemath.org/ticket/33562

[dcoudert]

comment:1

It's not the only undesirable behavior of methods using weighted.

sage: G = Graph([(0, 1)])
sage: G.weighted_adjacency_matrix()
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
...
TypeError: NoneType takes no arguments
sage: G.weighted_adjacency_matrix(base_ring=ZZ)
[0 0]
[0 0]
sage: G = Graph([(0, 1, '12')])
sage: G.weighted_adjacency_matrix(base_ring=ZZ)
[ 0 12]
[12  0]

This said, I'm not sure which is the best solution here. Should we first check edge weights (with/without using a base ring) ? force to pass a weight function ? Something else ?

[dcoudert]

Commit: aaabe1d

[dcoudert]

Author: David Coudert

[dcoudert]

comment:3

I added a check of edge weights. Let me know if this is what you expect.

---

New commits:

aaabe1d

trac #33562: better error message

[dcoudert]

Branch: public/graphs/33562

[dcoudert]

comment:5

Green bot, please review.

[dimpase]

comment:6

Why can't 'a' be a weight?

IMHO whatever values we have on edges may be weights.

[dcoudert]

comment:7

'a' can be a label, but not a weight. So far we always consider weights as numbers that you can use for instance to compute a shortest path or a minimum weight spanning tree.

[dimpase]

comment:8

Replying to David Coudert:

'a' can be a label, but not a weight. So far we always consider weights as numbers that you can use for instance to compute a shortest path or a minimum weight spanning tree.

sure, but we're talking about the adjacency matrix with the edge labels put at their respective places. These labels may be e.g. polynomials, or complex numbers - objects on which there is no natural order to speak about. Why would one want to forbid such matrices? E.g. in linear algebra one naturally may represent sparse matrices as graphs.

[sagetrac-git]

Changed commit from aaabe1d to b6abee8