> I didn't manage to construct a case where it breaks if
> tag is missing, rather than just reporting its absence.
I think this is because in case of a missing tag it is pretty unlikely
to actually crash. What it does in this case is to unwrap all
tags/inventories and then merge the complete history with the new
patches (plus their bogus context). If that doesn't already crash then
we detect the missing tag and fail cleanly.
To get a crash you need to ensure that there are common patches (in the
context of the bundle and our repo) that depend on patches that are not
in the bundle so their can't be commuted to the common trunk.
|