Comment 5 for bug 324724

This change has been merged now, but I noticed that Gustavo noted a problem on IRC:

<niemeyer> jamesh: We've got a broken test in Landscape due to the flushing on references
<niemeyer> jamesh: The problem is that we're building certain objects in memory, and passing them around to build further objects
<niemeyer> jamesh: and the additional logic is forcing flushes in the middle of the procedure
<niemeyer> jamesh: Which means that what would be only a few CREATEs become unfinished CREATEs + UPDATEs
<niemeyer> That can potentially break up with IntegrityErrors because it's flushing mid-way through constructors :-(

It'd be good to work out what the root cause here is. My guess here is that the unset local keys here would be set to NULL by the database when the row is added, so Landscape is relying on the unset columns being treated as NULLs. Does that sound like what you're seeing, Gustavo?