summaryrefslogtreecommitdiff
path: root/doc/devblog/day_193-194__ugly_bug.mdwn
blob: 668b7b2ab9c5dd04fe494d2a6c3e79397798211a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
**Important** A bug [[caused the assistant to sometimes remove all files|bugs/bad_merge_commit_deleting_all_files]] from the git repository.
You should check if your repository is ok. If the bug hit you, it should be
possible to revert the bad commit and recover your files with no data loss.
See the bug report for details. 

This affected git-annex versions since 5.20140613, and only when using the
assistant in direct mode. It should be fixed in today's release,
5.20140709.

I'm available <urgent2014@joeyh.name> to help anyone hit by this
unfortunate bug.

This is another bug in the direct mode merge code. I'm not happy about it.
It's particularly annoying that I can't fix up after it automatically
(because there's no way to know if any given commit in the git history that
deletes all the files is the result of this bug, or a legitimate deletion
of all files).

The only good thing is that the design of git-annex is pretty robust, and
in this case, despite stupidly committing the deletion of all the files in
the repository, git-annex did take care to preserve all their contents and
so the problem should be able to be resolved without data loss.

Unfortunately, the main autobuilder is down and I've had to spin up
autobuilders on a different machine (thank goodness that's very automated
now!), and so I have not been able to build the fixed git-annex for android
yet. I hope to get that done later this evening.

---

Yesterday, I fixed a few (much less bad) bugs, and did some thinking about
plans for this month. The [[design/roadmap]] suggests working on some of
[[!traillink design/assistant/chunks]], [[!traillink design/assistant/deltas]] or [[!traillink design/assistant/gpgkeys]].
I don't know how to do deltas yet really. Chunks is pretty easily done.
The gpg keys stuff is pretty open ended and needs some more work to define
some use cases. But, after today, I am more inclined to want to spend
time on better testing and other means of avoiding this kind of situation.