Open clbarnes opened 10 months ago
I fixed the issue in geodesic_matrix
with a5d4b72b439b0df6eb75e865167963287e15ad23 but your point is still well taken.
We could probably just rip out the alternative networkx
-based implementations that exist in some of the functions and nobody would notice. I'm trying to find a reason to keep them but the only thing I can come up with is a future regression in igraph
(or an improvement in networkx
) that would cause us to favour a networkx
-based implementation over igraph
. Probably not a strong argument against dropping the code that's currently effectively dead.
Oh I guess you mentioned yourself that we could test that results from igraph
- and networkx
-based implementations have the same result?
As igraph is now a non-optional dependency, would it be possible to remove a lot of the infrastructure involved in switching it on and off?
At present, we run a "no-igraph" CI job but it actually does use igraph, so it's not testing anything. Case in point: using scipy's sparse array for geodesic distance calculation is currently broken, but not tested in navis, so it just breaks in pymaid (which switches off igraph in a different way).