Throw a BoundsError when setting an attribute for a vertex that doesn't exist

[CameronBieganek]

The current behavior is that g[:a] = "hello" sets the vertex attribute for vertex :a to "hello" and it creates a new vertex with the index :a if it doesn't exist already. This syntax looks convenient, but I think it's a bit dangerous. Normally in Julia if you try to set a value at an index that doesn't exist, you get a BoundsError. I think it would be best to stick with that convention here. So if you want to add a vertex you have to explicitly add it via add_vertex!(g, :a, "hello").

[bramtayl]

I modeled this behavior on Dict, which does create new indexes.

[bramtayl]

Given that Dict does the same thing, I'm not too concerned, but feel free to reopen if you disagree.

[CameronBieganek]

One thing that I like about the SimpleGraphs API is that the data structure used to store the graph is abstracted away from the user. In other words, the user uses add_vertex! and add_edge! without any hint of how the graph is stored under the hood. But with the MetaGraphsNext API, it's pretty clear that the graph is stored as a dictionary.

Also, I find it a little weird that if I want to add a new vertex with no metadata, I have to do this:

g[:a] = nothing

(and likewise for adding edges with no metadata).

EDIT:
It's intuitive for access to meta data to follow the dictionary API, but it's not very intuitive for manipulations of the graph structure to follow the dictionary API.

[bramtayl]

I'm not a big fan of g[:a] = nothing either... I could just add a method

add_vertex!(g::MetaGraph, label) = add_vertex!(g, label, nothing)

?

[CameronBieganek]

I could just add a method

Yeah, that seems reasonable.