hi fred,
we tend to let that branch of the graph crash, which will automatically prevent things that depend on the crashed node from executing, while allowing things that do not depend on the crashed node to continue executing.
if you would like a more reasonable error message/send a notification, you can insert a function node in between those two nodes that sends does it's notification job and then crashes.
we currently don't do any analysis on the data and then operate on the graph as we've tried to maintain a dissociation between data and code. similarly mapnodes only expand at the point of execution not earlier, because it cannot determine ahead of time how many data elements it will get.