git-annex confused by a case-preserving filesystem?

Hello,

I use git annex to sync files between my home linux server and my mac os laptop (macbook). This is a report about some troubles I had recovering git-annexed files from the macbook hard-drive.

The laptop had some hardware failure, so I bought another laptop (this time running linux), and plugged the macbook hard drive on yet another linux computer to recover the data (some files which were added to the macbook annex but never synchronized to the server):

mount the hfsplus FS as read-only
cp -a (as root) the annex to /home/seb/macbook
use git annex as usual (from the linux computer) to push the files to the linux server.

Later on, I noticed that some files were missing (for instance "Le retour du printemps.mp4"). I eventually understood that it was related to HFS+ being case-insensitive (case-preserving. actually), and that by copying the annex to the linux partition I lost that property.

here are commands which I run from the linux computer to reproduce the issue:

$ cd /home/seb/macbook
$ git annex version
git-annex version: 5.20140412ubuntu1
$ git annex whereis  "public/Videos/Documentaries/Nature/Le retour du printemps.mp4"
whereis public/Videos/Documentaries/Nature/Le retour du printemps.mp4 (1 copy) 
        18b9c2cc-e3c1-4002-a0a2-7e07782fae2c -- here (navimac/annex)

Trying to copy the file to the linux server

$ git annex copy  "public/Videos/Documentaries/Nature/Le retour du printemps.mp4" --to origin ; echo $?
0

Here git-annex did not copy anything, but reported success, maybe a bug. Fsck does account the issue:

$ git annex fsck  "public/Videos/Documentaries/Nature/Le retour du printemps.mp4" 
fsck public/Videos/Documentaries/Nature/Le retour du printemps.mp4 (fixing location log) 
  ** Based on the location log, public/Videos/Documentaries/Nature/Le retour du printemps.mp4
  ** was expected to be present, but its content is missing.
  ** No known copies exist of public/Videos/Documentaries/Nature/Le retour du printemps.mp4
failed
git-annex: fsck: 1 failed

It turn out the file is there, but in gm/29 instead of gM/29:

$ ls -l  "public/Videos/Documentaries/Nature/Le retour du printemps.mp4"
lrwxrwxrwx 1 seb root 214 Apr 17  2017 public/Videos/Documentaries/Nature/Le retour du printemps.mp4 -> ../../../../.git/annex/objects/gM/29/SHA256E-s915126032--79eb20cad1b56adf4831b8a665b02c7de68aaaa4c9ca424a5a90e288abcec2aa.mp4/SHA256E-s915126032--79eb20cad1b56adf4831b8a665b02c7de68aaaa4c9ca424a5a90e288abcec2aa.mp4
$ ls -l .git/annex/objects/g{M,m}/29/SHA256E-s915126032--79eb20cad1b56adf4831b8a665b02c7de68aaaa4c9ca424a5a90e288abcec2aa.mp4/SHA256E-s915126032--79eb20cad1b56adf4831b8a665b02c7de68aaaa4c9ca424a5a90e288abcec2aa.mp4
ls: cannot access .git/annex/objects/gM/29/SHA256E-s915126032--79eb20cad1b56adf4831b8a665b02c7de68aaaa4c9ca424a5a90e288abcec2aa.mp4/SHA256E-s915126032--79eb20cad1b56adf4831b8a665b02c7de68aaaa4c9ca424a5a90e288abcec2aa.mp4: No such file or directory
-r--r--r-- 1 seb root 915126032 Apr 17  2017 .git/annex/objects/gm/29/SHA256E-s915126032--79eb20cad1b56adf4831b8a665b02c7de68aaaa4c9ca424a5a90e288abcec2aa.mp4/SHA256E-s915126032--79eb20cad1b56adf4831b8a665b02c7de68aaaa4c9ca424a5a90e288abcec2aa.mp4

I think that if ran the git annex from the hfsplus partition directly (ie. not copying the annex to /home/seb/macbook), it would have worked.

Moving the file where git annex expects to find it solves it all:

$ mkdir .git/annex/objects/gM
$ mv  .git/annex/objects/gm/29 .git/annex/objects/gM/
$ git annex fsck  "public/Videos/Documentaries/Nature/Le retour du printemps.mp4" 
fsck public/Videos/Documentaries/Nature/Le retour du printemps.mp4 (fixing location log) (checksum...)
ok
(Recording state in git...)

$ git annex copy  "public/Videos/Documentaries/Nature/Le retour du printemps.mp4" --to origin ; echo $?                                                                                                                                                 
copy public/Videos/Documentaries/Nature/Le retour du printemps.mp4 (checking origin...)
seb@192.168.0.1's password: 
(to origin...) 
SHA256E-s915126032--79eb20cad1b56adf4831b8a665b02c7de68aaaa4c9ca424a5a90e288abcec2aa.mp4
    915,126,032 100%    3.44MB/s    0:04:13 (xfr#1, to-chk=0/1)
ok
(Recording state in git...)
0

Maybe I could instead have used cpiofs to avoid renaming files. (But I have less than 10 files to rename anyway).

I learned a lot about git annex today! Thank you again fot git-annex.