UPDATE: There is a new post about a much improved version of this script: Lost in the Git Orcus (Part 3)

The Problem

Recently I lost a file in the Git orcus (aka './git/objects').
I knew it had been there some time ago.
But now it was gone and the Git on-board tools ('rev-list', 'fsck', 'reflog', ...) didn't help.
They showed a lot of dangling commits, trees and blobs but not the file I was searching for.

The Solution (maybe not the best)

I was doomed to dive into the .git/object-slough.
Based on a shell script on stackoverflow.com from willkil
I created a small Perl script:

It has one command line argument: the file to search for (as a regex).
In the first phase it looks into each tree object and checks whether it contains the file.
In the second phase it looks into each commit object and checks whether it points to one of the tree objects found in phase 1.
At last it creates a branch ('found/0', 'found/1', ...) for each commit found in phase 2.
It does not look into packed objects.

Here is the script:


#!/opt/local/bin/perl

my $pattern = $ARGV[0]; # filename to search for as regex

##################################################################
# find all objects (blobs, trees, commits, tags) in .git/objects
##################################################################

my @AllFiles = `find .git/objects/`;
my @AllSha1 = ();

for my $object (@AllFiles) {
	if($object =~ /([0-9a-f][0-9a-f])\/([0-9a-f]{38})/) {
		my $sha1 = $1 . $2;
		push(@AllSha1, $sha1);
	}
}

##################################################################
# find all trees containing the file
##################################################################

my @FoundTrees = ();

for my $treeSha1 (@AllSha1) {
	chomp $treeSha1;
	my $type = `git cat-file -t $treeSha1`;
	chomp $type;

	if($type eq "tree") {

		my @Lines = `git cat-file -p $treeSha1`;
		for my $line (@Lines) {
			if($line =~ /$pattern/) {
				printf "found tree: %s\n", $treeSha1;
				push(@FoundTrees, $treeSha1);
				break;
			}
		}
	}
}

##################################################################
# find all commits pointing to one of the trees found above
##################################################################

my @FoundCommits = ();

for my $commitSha1 (@AllSha1) {
	chomp $commitSha1;
	my $type = `git cat-file -t $commitSha1`;
	chomp $type;

	if($type eq "commit") {

		my @Lines = `git cat-file -p $commitSha1`;
		for my $line (@Lines) {
			for my $foundTreeSha1 (@FoundTrees) {
				if($line =~ /$foundTreeSha1/) {
					printf "found commit: %s\n", $commitSha1;
					push(@FoundCommits, $commitSha1);
					break;
				}
			}
		}
	}
}

##################################################################
# create a branch for each commit found
##################################################################

my $i = 0;
for my $foundSha1 (@FoundCommits) {
	my $branchName = "found/" . $i;
	`git branch $branchName $foundSha1`;
	printf "branch created: %s pointing to %s\n", $branchName, $foundSha1;
	$i++;
}

Be aware! This script creates a branch for each commit to a tree with the file you're searching for. Could be some!!!
Possible improvement: Only create a branch for the most recent commit (or track the chain of commits and search for the head of it)

The Question

Isn't there an easier way? Let me know!