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

Find lost files deep deep in the catacombs of Git

Last year I posted a script to find lost files deep in the .git/objects folder.
But the way I chose was not the smartest:
It scanned all tree objects for a given file name (or pattern) and searched for commits pointing to this tree.
But since only the root tree has a commit pointing to it the script could only find files in the root folder.

That was dumb from my side. So I fixed it and made it much simpler:

It searches for trees or blobs containing the given pattern (it could also search for commits or tags if you wanted to)
It it finds a blob it just logs it out
If it finds a tree it creates a commit pointing to it via 'git commit-tree' and a branch pointing to this tree

Searching for the contents of blobs may be necessary. E.g. if the file you are searching for never made it into a commit, but was part of the index (and therefore could be somewhere in the object store if the gc has not already removed it).

The script

The script has 3 parameters:

The object type to search in. Can be 'tree', 'blob', 'tag' or 'commit'
The pattern to search for
'y' if you want to create commits and branches (only when type='tree')

The script is super simple, has no bells and zero whistles. So feel free to --amend it.
You can also find it on Github (in its newest version)

Possible improvements

Galore!
Just some examples:

Make it more convenient: Use options, print a help text, ...
Search also in the pack file
To make it faster: Search at first in the reflog of the HEAD and the branches
...

I will implement the second bullet above soon. Stay tuned!

#!/opt/local/bin/perl

my $type = $ARGV[0]; # 'tree' or 'blob'?
my $pattern = $ARGV[1]; # filename to search for as regex
my $createBranches = $ARGV[2]; # 'y' if you want to create branches (only when search for 'tree')

##################################################################
# 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 or blobs containing the pattern
##################################################################

my $ctr = 0;
for my $foundSha1 (@AllSha1) {
	chomp $foundSha1;
	my $t = `git cat-file -t $foundSha1`;
	chomp $t;

	if($t eq $type) {

		my @Lines = `git cat-file -p $foundSha1`;
		for my $line (@Lines) {
			if($line =~ /$pattern/) {
				$ctr++;
				printf "found $type: %s\n", $foundSha1;
				if($type eq 'tree' && $createBranches eq 'y') {
					my $branchName = sprintf "found/%03s", $ctr;
					printf "created branch: %s\n", $branchName;
					`git commit-tree $foundSha1 -m "found $pattern" | xargs -I{} git branch $branchName {}`;
				}
				break;
			}
		}
	}
}