Fix bug when processing blank nodes with reverse properties

[lanthaler]

This fixes a bug when creating a node map for blank nodes with reverse properties. See http://lists.w3.org/Archives/Public/public-linked-json/2013Oct/0046.html for the original bug report.

I've also added two new tests (toRdf-0118 and flatten-0044) which test the fix and the EARL report for my implementation. None of the existing tests has been invalidated.

[gkellogg]

+1

[dlongley]

I solved this a little differently. When processing the items in the reverse map, I check to see if the item is a blank node, and if so, I generate the new blank node identifier for it. Then, I recurse before merging the reference node into the node in the node map related to the item. It seems to work and passes the test. This avoids any need to pass somewhat hacky parameters.

[dlongley]

So, 6.8.3.1.1 becomes a check to see if "value" is a blank node, and if so, a new blank node identifier is generated. 6.8.3.1.2 is the same but also passes the new identifier. Then 6.8.3.1.3 is the old 6.8.3.1.1 except the change is made to the entry in the node map, not "value".

[dlongley]

Well, looking at the rest of the node map generation algorithm, it seems like it has deviated a bit from how I originally wrote mine. So while my change works with my existing code, it may not work properly with the rest of the algorithm as it stands in the spec. My code always passes generated blank node IDs as a parameter so they don't get regenerated; it looks like the spec algorithm always generates them.

[lanthaler]

Since the time is critical right now, I'm going ahead and merge this changes directly instead of trying to improve the algorithm further.