Weight matrices

[gdalle]

Do they still make sense?
Should weights be defined even for edges that don't exist?

[gdalle]

Related:

• [Port] Inconsistency about weights Graphs.jl#45
• Lazy graphs Graphs.jl#115

[CameronBieganek]

I think it's probably best to leave the interface generic, so something like weight(g, e) seems good.

Should weights be defined even for edges that don't exist?

Probably not? But I'm not really an expert on graph algorithms, so I can't say for sure.

[gdalle]

Should weights be defined even for edges that don't exist?

The question arises because in Graphs.jl, the default weight matrix is basically a FillArrays.Ones, which returns 1 for every query even when the edge does not exist. I don't like that.

[gdalle]

I think it's probably best to leave the interface generic, so something like weight(g, e) seems good.

Yes, and it's consistent with a remark by @mtfishman about separating the data from the edge (as in, weight(e) is less flexible)