[19:53]	_FrnchFrgg_	esr: I have succeeded in producing a really small
                        testcase for my assert node.blobmark bug
[19:55]	_FrnchFrgg_	The problematic revision does a copyfrom to create a
                        file "a" then changes it in the same revision
[19:56]	_FrnchFrgg_	Which means that the filemaps[previous][node.path]
                        won't find any ancestor, because here the ancestor
                        is in the filemap indexed by node.revision...
[19:58]	_FrnchFrgg_	The testcase is 5 revisions long. IIRC I didn't
                        succeed in reproducing without the delete action
                        before (so that there is really no ancestor) but I
                        don't see why it should matter.
[19:58]	_FrnchFrgg_	Perhaps a mistake on my part.
[19:59]	_FrnchFrgg_	Anyway, the linked testcase asserts.