[[!comment format=mdwn
 username="joey"
 subject="""comment 2"""
 date="2016-05-23T17:30:07Z"
 content="""
I've now got a copy of the repo, in
~/lib/big/git-annex-test-repos/ssl.zerodogg.org__zerodogg_private_tmp_privateDocs.zerodogg.tar.xz.gpg

Looking at commit 77c7d149646655fb5851c3db584fe70d64707d04, it merges in
0b4896bc205696630c81cf557959a4aaa24906f0 which has an empty tree. 

0b4896bc205696630c81cf557959a4aaa24906f0 is itself a merge commit. 
Both of the commits it merges themselves have empty trees.
And so it goes down quite a way, with empty merge commits including
418367b, 7bab5cf, b651554, cf5de84, c5905f7, 928040e, a590245, 5b53fc9, 
6d9f5da, 5f2623d. The freqency of these might indeed indicate some kind
of feedback loop, but I don't think whatever is causing that is the core problem.

fc6670a37fd9d3984a112a80d9bbaec5c041c005 is the crucial merge it seems. Its
parents are 71b6c8a and f8dfc21. Both of those parents have the same tree,
5f18bed323c29fb77add3a84abcf8b1fb6b646d7, and that tree is populated with all
the files. But somehow this merge deleted everything.

	tree 4b825dc642cb6eb9a060e54bf8d69288fbee4904
	parent 71b6c8ad3fc44926c9be2bbb1bd308592b6eb05c
	parent f8dfc219a40b2871baed3192ea5806bb4ac551e7
	author xxx 1463570165 +0200
	committer xxx 1463570165 +0200
	
	Merge branch 'refs/heads/synced/master' into HEAD

(There are empty trees earlier where the same thing happened that you
reverted, but it seems best to focus on the most recent occurance.)

So, can you find fc6670a in .git/annex/daemon.log* in any of the
clones of this repository? It would be good to narrow down on which
machine(s) the bad merge is happening. (Maybe you've already narrowed it down?)

One of the two parent commits (71b6c8ad3fc44926c9be2bbb1bd308592b6eb05c)
is a manual revert you did, the other commit looks to have been done by
the assistant. 
I'm guessing that refs/heads/synced/master was f8dfc219a40b2871baed3192ea5806bb4ac551e7
when the bad merge was generated. So this bad merge probably happened in
the repository where you did that manual reversion.

As far as I can tell this was a regular git merge that somehow decided to empty
the tree. It was not a case of git-annex auto-resolving a merge conflict.

Are you using adjusted branches in any of the clones of this repository?

What version(s) of git are being used?

(I noticed that despite using v6 mode, every file in the repository
seems to be locked, so the smudge filters etc should not be involved in the
problem unless using an adjusted branch.)
"""]]