[git] Git: "Corrupt loose object"

Whenever I pull from my remote, I get the following error about compression. When I run the manual compression, I get the same:

$ git gc
error: Could not read 3813783126d41a3200b35b6681357c213352ab31
fatal: bad tree object 3813783126d41a3200b35b6681357c213352ab31
error: failed to run repack

Does anyone know, what to do about that?

From cat-file I get this:

$ git cat-file -t 3813783126d41a3200b35b6681357c213352ab31
error: unable to find 3813783126d41a3200b35b6681357c213352ab31
fatal: git cat-file 3813783126d41a3200b35b6681357c213352ab31: bad file

And from git fsck I get this ( don't know if it's actually related):

$ git fsck
error: inflate: data stream error (invalid distance too far back)
error: corrupt loose object '45ba4ceb93bc812ef20a6630bb27e9e0b33a012a'
fatal: loose object 45ba4ceb93bc812ef20a6630bb27e9e0b33a012a (stored in .git/objects/45/ba4ceb93bc812ef20a6630bb27e9e0b33a012a) is corrupted

Can anyone help me decipher this?

I just experienced this - my machine crashed whilst writing to the Git repo, and it became corrupted. I fixed it as follows.

I started with looking at how many commits I had not pushed to the remote repo, thus:

gitk &

If you don't use this tool it is very handy - available on all operating systems as far as I know. This indicated that my remote was missing two commits. I therefore clicked on the label indicating the latest remote commit (usually this will be /remotes/origin/master) to get the hash (the hash is 40 chars long, but for brevity I am using 10 here - this usually works anyway).

Here it is:

14c0fcc9b3

I then click on the following commit (i.e. the first one that the remote does not have) and get the hash there:

04d44c3298

I then use both of these to make a patch for this commit:

git diff 14c0fcc9b3 04d44c3298 > 1.patch

I then did likewise with the other missing commit, i.e. I used the hash of the commit before and the hash of the commit itself:

git diff 04d44c3298 fc1d4b0df7 > 2.patch

I then moved to a new directory, cloned the repo from the remote:

git clone [email protected]:username/repo.git

I then moved the patch files into the new folder, and applied them and committed them with their exact commit messages (these can be pasted from git log or the gitk window):

patch -p1 < 1.patch
git commit

patch -p1 < 2.patch
git commit

This restored things for me (and note there's probably a faster way to do it for a large number of commits). However I was keen to see if the tree in the corrupted repo can be repaired, and the answer is it can. With a repaired repo available as above, run this command in the broken folder:

git fsck 

You will get something like this:

error: object file .git/objects/ca/539ed815fefdbbbfae6e8d0c0b3dbbe093390d is empty
error: unable to find ca539ed815fefdbbbfae6e8d0c0b3dbbe093390d
error: sha1 mismatch ca539ed815fefdbbbfae6e8d0c0b3dbbe093390d

To do the repair, I would do this in the broken folder:

rm .git/objects/ca/539ed815fefdbbbfae6e8d0c0b3dbbe093390d
cp ../good-repo/.git/objects/ca/539ed815fefdbbbfae6e8d0c0b3dbbe093390d .git/objects/ca/539ed815fefdbbbfae6e8d0c0b3dbbe093390d

i.e. remove the corrupted file and replace it with a good one. You may have to do this several times. Finally there will be a point where you can run fsck without errors. You will probably have "dangling commit" and "dangling blob" lines in the report, these are a consequence of your rebases and amends in this folder, and are OK. The garbage collector will remove them in due course.

Thus (at least in my case) a corrupted tree does not mean unpushed commits are lost.

I encountered this once my system crashed. What I did is this:

(Please note your corrupt commits are lost, but changes are retained. You might have to recreate those commits at the end of this procedure)

• Backup your code.
• Go to your working directory and delete the .git folder.
• Now clone the remote in another location and copy the .git folder in it.
• Paste it in your working directory.
• Commit as you wanted.

My computer crashed while I was writing a commit message. After rebooting, the working tree was as I had left it and I was able to successfully commit my changes.

However, when I tried to run git status I got

error: object file .git/objects/xx/12345 is empty
fatal: loose object xx12345 (stored in .git/objects/xx/12345 is corrupt

Unlike most of the other answers, I wasn't trying to recover any data. I just needed Git to stop complaining about the empty object file.

Overview

The "object file" is git's hashed representation of a real file that you care about. Git thinks it should have a hashed version of some/file.whatever stored in .git/object/xx/12345, and fixing the error turned out to be mostly a matter of figuring out which file the "loose object" was supposed to represent.

Details

Possible options seemed to be

1. Delete the empty file
2. Get the file into a state acceptable to Git

Approach 1: Remove the object file

The first thing I tried was just moving the object file

mv .git/objects/xx/12345 ..

That didn't work - git began complaining about a broken link. On to Approach 2.

Approach 2: Fix the file

Linus Torvalds has a great writeup of how to recover an object file that solved the problem for me. Key steps are summarized here.

$> # Find out which file the blob object refers to
$> git fsck
broken link from    tree 2d9263c6d23595e7cb2a21e5ebbb53655278dff8
           to    blob xx12345
missing blob xx12345

$> git ls-tree 2d926
...
10064 blob xx12345  your_file.whatever

This tells you what file the empty object is supposed to be a hash of. Now you can repair it.

$> git hash-object -w path/to/your_file.whatever

After doing this I checked .git/objects/xx/12345, it was no longer empty, and git stopped complaining.

Runnning git stash; git stash pop fixed my problem

I just had a problem like this. My particular problem was caused by a system crash that corrupted the most recent commit (and hence also the master branch). I hadn't pushed, and wanted to re-make that commit. In my particular case, I was able to deal with it like this:

1. Make a backup of .git/: rsync -a .git/ git-bak/
2. Check .git/logs/HEAD, and find the last line with a valid commit ID. For me, this was the second most recent commit. This was good, because I still had the working directory versions of the file, and so the every version I wanted.
3. Make a branch at that commit: git branch temp <commit-id>
4. re-do the broken commit with the files in the working directory.
5. git reset master temp to move the master branch to the new commit you made in step 2.
6. git checkout master and check that it looks right with git log.
7. git branch -d temp.
8. git fsck --full, and it should now be safe to delete any corrupted objects that fsck finds.
9. If it all looks good, try pushing. If that works,

That worked for for me. I suspect that this is a reasonably common scenario, since the most recent commit is the most likely one to be corrupted, but if you lose one further back, you can probably still use a method like this, with careful use of git cherrypick, and the reflog in .git/logs/HEAD.

I had this same problem in my bare remote git repo. After much troubleshooting, I figured out one of my coworkers had made a commit in which some files in .git/objects had permissions of 440 (r--r-----) instead of 444 (r--r--r--). After asking the coworker to change the permissions with "chmod 444 -R objects" inside the bare git repo, the problem was fixed.

In answer of @user1055643 missing the last step:

$ rm -fr .git
$ git init
$ git remote add origin your-git-remote-url
$ git fetch
$ git reset --hard origin/master
$ git branch --set-upstream-to=origin/master master  

Try

git stash

This worked for me. It stashes anything you haven't committed and that got around the problem.

I was getting a corrupt loose object error as well.

./objects/x/x

I successfully fixed it by going into the directory of the corrupt object. I saw that the users assigned to that object was not my git user's. I don't know how it happened, but I ran a chown git:git on that file and then it worked again.

This may be a potential fix for some peoples' issues but not necessary all of them.

We just had the case here. It happened that the problem was the ownership of the corrupt file was root instead of our normal user. This was caused by a commit done on the server after someone has done a "sudo su --".

First, identify your corrupt file with:

$> git fsck --full

You should receive a answer like this one:

fatal: loose object 11b25a9d10b4144711bf616590e171a76a35c1f9 (stored in .git/objects/11/b25a9d10b4144711bf616590e171a76a35c1f9) is corrupt

Go in the folder where the corrupt file is and do a:

$> ls -la

Check the ownership of the corrupt file. If that's different, just go back to the root of your repo and do a:

$> sudo chown -R YOURCORRECTUSER:www-data .git/

Hope it helps!

I had the same problem (don't know why).

This fix requires access to an uncorrupted remote copy of the repository, and will keep your locally working copy intact.

But it has some drawbacks:

• You will lose the record of any commits that were not pushed, and will have to recommit them.
• You will lose any stashes.

The fix

Execute these commands from the parent directory above your repo (replace 'foo' with the name of your project folder):

1. Create a backup of the corrupt directory:
   cp -R foo foo-backup
2. Make a new clone of the remote repository to a new directory:
   git clone [email protected]:foo foo-newclone
3. Delete the corrupt .git subdirectory:
   rm -rf foo/.git
4. Move the newly cloned .git subdirectory into foo:
   mv foo-newclone/.git foo
5. Delete the rest of the temporary new clone:
   rm -rf foo-newclone

On Windows you will need to use:

• copy instead of cp -R
• rmdir /S instead of rm -rf
• move instead of mv

Now foo has its original .git subdirectory back, but all the local changes are still there. git status, commit, pull, push, etc. work again as they should.

Just remove .git folder and add it again. This simple solution worked for me.

simply running a git prune fixed this issue for me

Your best bet is probably to simply re-clone from the remote repo (ie. Github or other). Unfortunately you will lose any unpushed commits and stashed changes, however your working copy should remain intact.

First make a backup copy of your local files. Then do this from the root of your working tree:

rm -fr .git
git init
git remote add origin [your-git-remote-url]
git fetch
git reset --mixed origin/master
git branch --set-upstream-to=origin/master master  

Then commit any changed files as necessary.

I solved this way: I decided to simply copy the uncorrupted object file from the backup's clone to my original repository. This worked just as well. (By the way: If you can't find the object in .git/objects/ by its name, it probably has been [packed][pack] to conserve space.)

Working on a VM, in my notebook, battery died, got this error;

error: object file .git/objects/ce/theRef is empty error: object file .git/objects/ce/theRef is empty fatal: loose object theRef (stored in .git/objects/ce/theRef) is corrupt

I managed to get the repo working again with only 2 commands and without losing my work (modified files/uncommitted changes)

find .git/objects/ -size 0 -exec rm -f {} \;
git fetch origin

After that I ran a git status, the repo was fine and there were my changes (waiting to be committed, do it now..).

git version 1.9.1

Remember to backup all changes you remember, just in case this solution doesn't works and a more radical approach is needed.

A garbage collection fixed my problem:

git gc --aggressive --prune=now

Takes a while to complete, but every loose object and/or corrupted index was fixed.

I got this error after my (windows) machine decided to reboot itself. Thankfully my remote repo was up to date so I just did a fresh git-clone..

To me this happened due to a power failure while doing a git push.

The messages looked like this:

$ git status
error: object file .git/objects/c2/38824eb3fb602edc2c49fccb535f9e53951c74 is empty
error: object file .git/objects/c2/38824eb3fb602edc2c49fccb535f9e53951c74 is empty
fatal: loose object c238824eb3fb602edc2c49fccb535f9e53951c74 (stored in .git/objects/c2/38824eb3fb602edc2c49fccb535f9e53951c74) is corrupt

I tried things like git fsck but that didn't help. Since the crash happened during a git push, it obviously happened during rewrite on the client side which happens after the server is updated. I looked around and figured that c2388 in my case was a commit object, because it was referred to by entries in .git/refs. So I knew that I would be able to find c2388 when I look at the history (through a web interface or second clone).

On the second clone I did a git log -n 2 c2388 to identify the predecessor of c2388. Then I manually modified .git/refs/heads/master and .git/refs/remotes/origin/master to be the predecessor of c2388 instead of c2388. Then I could do a git fetch. The git fetch failed a few times for conflicts on empty objects. I removed each of these empty objects until git fetch succeeded. That has healed the repository.

When I had this issue I backed up my recent changes (as I knew what I had changed) then deleted that file it was complaining about in .git/location. Then I did a git pull. Take care though, this might not work for you.

Create a backup and clone the repository into a fresh directory

cp -R foo foo-backup
git clone git@url:foo foo-new

(optional) If you are working on a different branch than master, switch it.

cd foo-new
git checkout -b branch-name origin/branch-name

Sync changes excluding the .git directory

rsync -aP --exclude=.git foo-backup/ foo-new

I followed many of the other steps here; Linus' description of how to look at the git tree/objects and find what's missing was especially helpful. git-git recover corrupted blob

But in the end, for me, I had loose/corrupt tree objects caused by a partial disk failure, and tree objects are not so easily recovered/not covered by that doc.

In the end, I moved the conflicting objects/<ha>/<hash> out of the way, and used git unpack-objects with a pack file from a reasonably up to date clone. It was able to restore the missing tree objects.

Still left me with a lot of dangling blobs, which can be a side effect of unpacking previously archived stuff, and addressed in other questions here