summaryrefslogtreecommitdiff
path: root/doc/bugs/directory_remote_and_case_sensitivity_on_FAT.mdwn
blob: b686304e598edfa2bdbae0e2e2d21dae0f145c7a (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
I was copying files to a directory remote with `git annex copy`.  Out of 114 files, 9 of them failed with no message, just:

    copy data/foo.dat (to usbdrive...) failed
    copy data/bar.dat (to usbdrive...) failed

According to strace:

    31338 mkdir("/media/annex/Zp/9v/SHA256-s1362999320--d650297c8cf8c2dc0575110a52d0c5cc0ff266f294a0599f85796a6b44b23492", 0777) = -1 ENOENT (No such file or directory)
    31338 mkdir("/media/annex/Zp/9v", 0777) = -1 ENOENT (No such file or directory)
    31338 mkdir("/media/annex/Zp", 0777)    = -1 EEXIST (File exists)
    31338 stat("/media/annex/Zp", 0x7f8449f170d0) = -1 ENOENT (No such file or directory)

The filesystem is FAT32 and has weird case semantics.  This was mounted by udisks with its default options:

    /dev/sdb1 on /media/annex type vfat (rw,nosuid,nodev,uhelper=udisks,uid=1000,gid=1000,shortname=mixed,dmask=0077,utf8=1,showexec)

I wonder if the directory remote should use hashDirLower instead of hashDirMixed?

> git-annex intentionally uses the same layout for directory and rsync
> special remotes as it does for the .git/annex directory. As far
> as I know it works ok on case-insensative filesystems.
> 
> Based on your strace, if you `ls /media/annex/Zp`, you will see
> "No such file or directory", but if you `mkdir /media/annex/Zp` it will
> fail with "File exists". Doesn't make much sense to me.
> 
> I cannot reproduce this problem using a vfat filesystem 
> mounted using the same options you show (linux 3.0.0). --[[Joey]]