Fix setting of back flag when splitting streets

[evansiroky]

This fix includes most of the recommended suggestions made by @jwoyame. It corrects the setting of the back flag for StreetEdges in order for the geometry to be deserialized properly.

This should fix ibi-group/trimet-mod-otp#243 and also fix ibi-group/trimet-mod-otp#244.

[landonreed]

I don't have a great understanding of what exactly is going on here. I think it would be helpful if you could quote/extract bits of the relevant conversation in this PR. The link provided in the current description doesn't point to a particular comment.

Is there some case that demonstrates that something has been fixed with this change? I'm not necessarily requesting a test here, just curious if there are repeatable steps I could follow to demonstrate that a bug has been fixed.

Also, are you able to identify here why this value was set to false in the first place? Would it be useful to amend the javadoc for the constructor to be more descriptive about why it uses its parent edge's value?

[evansiroky]

Yeah, I suppose I meant to link to the comment specifically made by @jwoyame. Basically the thing going on is that the way StreetEdge geometry was serialized and deserialized from internal OTP data structures became broken and needed fixing. When the StreetEdge geometry is stored, it is dependent on the back flag which is meant to store information about whether a StreetEdge represents the data about the corresponding OSM way in the backward direction.

The code I made for creating PartialStreetEdges was incorrect. At first, it was incorrectly setting the back flag to false for every PartialStreetEdge. This was an artifact of the previous version of code which might have been intended mainly for TemporaryFreeEdges which don't represent OSM ways and therefore wouldn't ever be a back representation of an OSM way, but I'm not sure how it ended up in PartialStreetEdge. However, that alone didn't cause the issue because the geometry was serialized and deserialized correctly because the back flag didn't change. The problem occurred after in StreetEdge.java when the back flag was set again to the parent edge's back flag which did result in changing the back flag after the geometry was already serialized. I mistakenly added this along with every other StreetEdge attribute that I thought needed copying to split edges. Therefore whenever the geometry was deserialized and the back flag wasn't false for the parent edge, the geometry would look reversed.

[landonreed]

Thanks for the clarification!