Comment 4 for bug 256409

This is critical, having looked at it it is a data loss bug:

here is what I think is happening:
 - we get a bad delta for some reason coming *from commit*
 - a bad commit is recorded - we have dangling references
 - the tree update catches the fail

What we need to do urgently:
 - ensure commit rolls back the branch pointer if this happens so that the problem commit doesn't get referenced
 - help the users affected to date recover
 - get the workaround into distros using 1.3.1 (which seems to be common version for this, could be 1.3.1 specific or more likely just that its the version in hardy

Less urgently we should get a root cause diagnosis and fix the deltas being generated to be correct; also we need to ensure a bad delta passed to commit won't cause nonsense data to be inserted.