I ran into the same kind of bug that I thought we'd eliminated with a recent commit.
It turns out that the range extension code which checked for sequence starts as a reason to break the range extension was only doing so correctly for the reverse strand. This makes it fully stranded and correct. It also pushes one of the tests back to its previous md5, which matched the (much simpler, and likely correct) version that I'd established before implementing parallel union find.
I ran into the same kind of bug that I thought we'd eliminated with a recent commit.
It turns out that the range extension code which checked for sequence starts as a reason to break the range extension was only doing so correctly for the reverse strand. This makes it fully stranded and correct. It also pushes one of the tests back to its previous md5, which matched the (much simpler, and likely correct) version that I'd established before implementing parallel union find.