There is a bug in the FS history tracing algorithm. I've not yet been
successful at creating a small reproduction recipe, but it has manifested itself
in the Codespeak repository. The bug can be triggered (at least for now) by
running
$ svn log http://codespeak.net/svn/pypy/trunk/src/pypy/__init__.py
The resulting log output will display (among other revisions) revision 1119, in
which the path being reported on actually did *not* change -- it changed instead
revision 1118.
Suspect in this matter is the particular use of copy ID inheritance rules to
determine which copies are relevant to the history of a node. I'm not certain
of this, but there may be a timing problem with that data, which is calculated
for immutable nodes in anticipation of them being mutable, and (I fear)
near-worthless afterwards.