Inconsistent handling of edges involving non-existent nodes

[David Oleander]

Consider the following program:

module Main exposing (..)

import Html exposing (Html, text)
import Graph

x : Int
x = 0
y : Int
y = 0

graph : Graph.Graph () ()
graph =
    Graph.fromNodesAndEdges
        [{id = 0, label = ()}]
        [{from = x, to = y, label = ()}]

main : Html a
main =
    text (Graph.toString graph)

It produces the following output, which is expected:
Graph.fromNodesAndEdges [{ id = 0, label = () }] [{ to = 0, from = 0, label = () }]

However, changing the values of x and y to 0 and 1 gives:
Graph.fromNodesAndEdges [{ id = 0, label = () }] [{ to = 1, from = 0, label = () }]
while changing the values of x and y to 1 and 0 gives:
Graph.fromNodesAndEdges [{ id = 0, label = () }] []

The documentation does not describe what should happen if a node id for a non-existent node is provided for an edge, so this is not a bug in the strictest sense, but the fact that these last two scenarios provide different results strikes me as odd. I would expect that both would produce the last output.

[Sebastian Graf]

That's indeed surprising. I'd also expect the edge set for the first case to be empty.

[Sebastian Graf]

Published as v3.0.1.

[Sebastian Graf]

Not that the library code changed, but I fixed and updated the test suite and bumped to v3.0.2 in order to make CI green again.