Before I start on what's gone wrong, many thanks for a great program: finally a way of finding out where I've put all those files, and I enjoyed your talk in Australia. Not quite to New Zealand, but a good start :-)

To play with git-annex, I decided to convert my ~/Downloads directory to a git-annex repository, as there were a wide variety of files in it, mostly easily replaceable, and also handy to have on multiple machines. In hindsight, probably wasn't a great idea, as I'd regularly forget it was a git-annex repo and move files and directories out manually, which caused all sorts of fun trying to sort out the dead symlinks. Then I after one update, the repository format changed, from WORM to xxx256 format which meant there were now both sorts in the GA object store.

More recently I'd tried converting the repo to direct mode, you get the idea: lots of playing with git-annex commands, and now that I think about it, possibly some git commands too trying to repair missing files.

Anyway I've ended up with a 27GB git-annex repo that now manages to kill git-annex whenever I try to check it using "git-annex fsck". 

Not only does the fsck subcommand cause it to die, but also "find", "whereis" and "status". It dies on the same file (for find/whereis/fsck).

e.g.

    ... lots of stuff deleted ...
    whereis 1wolf14.zip (2 copies) 
            051f0b00-e265-11e1-894e-3b0b3f3844f2 -- Laptop
            2c4e11e0-a1b4-11e1-9a02-73e17b04c00f -- here (myPC - Downloads)
    ok
    git-annex: out of memory (requested 2097152 bytes)

Now "git status" for some repo data:

    myPC:~/Downloads$ git annex status
    supported backends: SHA256E SHA1E SHA512E SHA224E SHA384E SHA256 SHA1 SHA512 SHA224 SHA384 WORM URL
    supported remote types: git S3 bup directory rsync web webdav glacier hook
    repository mode: direct
    trusted repositories: 0
    semitrusted repositories: 4
            00000000-0000-0000-0000-000000000001 -- web
            051f0b00-e265-11e1-894e-3b0b3f3844f2 -- Laptop
            2c4e11e0-a1b4-11e1-9a02-73e17b04c00f -- here (myPC - Downloads)
            48fbe52a-a1b3-11e1-bb80-ebc15118871d -- netbk
    untrusted repositories: 0
    dead repositories: 0
    transfers in progress: none
    available local disk space: 100 gigabytes (+1 megabyte reserved)
    local annex keys: 1719
    local annex size: 27 gigabytes
    known annex keys: git-annex: out of memory (requested 2097152 bytes)

It always seems to die at about 3.5GB memory usage. This is running on Ubuntu 12.04, using the latest GA release built using cabal:

    git-annex version: 4.20130227
    local repository version: 3
    default repository version: 3
    supported repository versions: 3 4
    upgrade supported from repository versions: 0 1 2

There are also dead symlinks that point to directories that have meta-data but not the symlink target (manually line-wrapped):

    myPC:~/Downloads$ ls -l precise*
    lrwxrwxrwx 1 nino nino 194 Oct 21 12:48 precise-dvd-i386.iso -> 
                 .git/annex/objects/0x/Xz/SHA256-s3590631424--
                 b08ecdd4846948ec076b23afae7f87be9cfba5218fb9ba4160f26c0b8d4b5dd0/
                 SHA256-s3590631424--b08ecdd4846948ec076b23afae7f87be9cfba5218fb9ba4160f26c0b8d4b5dd0

But looking in the symlink destination directory, there's no corresponding object, only metadata:

    myPC:~/Downloads/.git/annex/objects/0x/Xz/SHA256-s3590631424--b08ecdd4846948ec076b23afae7f87be9cfba5218fb9ba4160f26c0b8d4b5dd0$ ls -l
    total 8
    -rw-rw-r-- 1 nino nino 30 Jan  8 23:15 SHA256-s3590631424--b08ecdd4846948ec076b23afae7f87be9cfba5218fb9ba4160f26c0b8d4b5dd0.cache
    -rw-rw-r-- 1 nino nino 49 Jan  8 23:15 SHA256-s3590631424--b08ecdd4846948ec076b23afae7f87be9cfba5218fb9ba4160f26c0b8d4b5dd0.map

But there is another version somewhere else.

    -r--r--r-- 1 nino nino 3590631424 Mar 18  2012 ./.git/annex/objects/Kj/wM/WORM-s3590631424-m1331991509
                        --precise-dvd-i386.iso/WORM-s3590631424-m1331991509--precise-dvd-i386.iso

This actual file does exist in the "Used" directory:

    -rw-r--r-- 1 nino nino 3.4G May 27  2012 precise-dvd-i386.iso

I'm not so worried about the mangled repo - it's quite possibly because of clueless git/git-annex command usage - but the inability to use the fsck command is concerning

I could just uninit everything, but as it dies prematurely, I'm not certain that all the contents would be restored.
Any thoughts on how I can get git-annex (esp. fsck) to complete would be appreciated.

Thanks
Giovanni

> [[fixed|done]]. However, if you saw this behavior,
> you have large files checked directly into git. You may
> want to examine your repository and use git filter-branch to clean
> it up.
> --[[Joey]]