Found the bug. What was happening was:

1) γ contains a range, say a to b, where a is a segment boundary but b-a < segmentsize
2) γ -> β happens, so all of γ gets loaded into q3 as betas. γ is empty
3) then push β to γ happens. There was a logic failure in there, which meant that with γ empty, _all_ of q3 would be pushed to γ, but incorrectly accounted for, making the vq think there is less in γ than there really is
4) now γ -> β happens again, vq is amazed to find more in the segment in γ than it was expecting. cue explosion
The bug was simply not ensuring that if γ was empty, only push out enough to leave q3 with a maximum of one segment, ending on a segment boundary

0 files changed, 0 insertions, 0 deletions