Import Upstream version 0.0~git20180821.e1733b1

upstream
Marcel Kapfer 3 years ago
commit eb2d41f476
  1. 22
      LICENSE
  2. 242
      README.md
  3. 628
      git-fat
  4. 53
      test-retroactive.sh
  5. 64
      test.sh

@ -0,0 +1,22 @@
Copyright (c) 2012, Jed Brown
All rights reserved.
Redistribution and use in source and binary forms, with or without modification,
are permitted provided that the following conditions are met:
* Redistributions of source code must retain the above copyright notice, this
list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright notice, this
list of conditions and the following disclaimer in the documentation and/or
other materials provided with the distribution.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

@ -0,0 +1,242 @@
# Introduction
Checking large binary files into a source repository (Git or otherwise) is a bad idea because repository size quickly becomes unreasonable.
Even if the instantaneous working tree stays manageable, preserving repository integrity requires all binary files in the entire project history, which given the typically poor compression of binary diffs, implies that the repository size will become impractically large.
Some people recommend checking binaries into different repositories or even not versioning them at all, but these are not satisfying solutions for most workflows.
## Features of `git-fat`
* clones of the source repository are small and fast because no binaries are transferred, yet fully functional with complete metadata and incremental retrieval (`git clone --depth` has limited granularity and couples metadata to content)
* `git-fat` supports the same workflow for large binaries and traditionally versioned files, but internally manages the "fat" files separately
* `git-bisect` works properly even when versions of the binary files change over time
* selective control of which large files to pull into the local store
* local fat object stores can be shared between multiple clones, even by different users
* can easily support fat object stores distributed across multiple hosts
* depends only on stock Python and rsync
## Related projects
* [git-annex](http://git-annex.branchable.com) is a far more comprehensive solution, but with less transparent workflow and with more dependencies.
* [git-media](https://github.com/schacon/git-media) adopts a similar approach to `git-fat`, but with a different synchronization philosophy and with many Ruby dependencies.
# Installation and configuration
Place `git-fat` in your `PATH`.
Edit (or create) `.gitattributes` to regard any desired extensions as fat files.
$ cd path-to-your-repository
$ cat >> .gitattributes
*.png filter=fat -crlf
*.jpg filter=fat -crlf
*.gz filter=fat -crlf
^D
Run `git fat init` to activate the extension. Now add and commit as usual.
Matched files will be transparently stored externally, but will appear
complete in the working tree.
Set a remote store for the fat objects by editing `.gitfat`.
[rsync]
remote = your.remote-host.org:/share/fat-store
This file should typically be committed to the repository so that others
will automatically have their remote set. This remote address can use
any protocol supported by rsync.
Most users will configure it to use remote ssh in a directory with shared
access. To do this, set the `sshuser` and `sshport` variables in `.gitfat`
configuration file. For example, to use rsync with ssh, with the default
port (22) and authenticate with the user "_fat_", your configuration would
look like this:
[rsync]
remote = your.remote-host.org:/share/fat-store
sshuser = fat
# A worked example
Before we start, let's turn on verbose reporting so we can see what's
happening. Without this environment variable, all the output lines
starting with `git-fat` will not be shown.
$ export GIT_FAT_VERBOSE=1
First, we create a repository and configure it for use with `git-fat`.
$ git init repo
Initialized empty Git repository in /tmp/repo/.git/
$ cd repo
$ git fat init
$ cat > .gitfat
[rsync]
remote = localhost:/tmp/fat-store
$ mkdir -p /tmp/fat-store # make sure the remote directory exists
$ echo '*.gz filter=fat -crlf' > .gitattributes
$ git add .gitfat .gitattributes
$ git commit -m'Initial repository'
[master (root-commit) eb7facb] Initial repository
2 files changed, 3 insertions(+)
create mode 100644 .gitattributes
create mode 100644 .gitfat
Now we add a binary file whose name matches the pattern we set in `.gitattributes`.
$ curl https://nodeload.github.com/jedbrown/git-fat/tar.gz/master -o master.tar.gz
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 6449 100 6449 0 0 7741 0 --:--:-- --:--:-- --:--:-- 9786
$ git add master.tar.gz
git-fat filter-clean: caching to /tmp/repo/.git/fat/objects/b3489819f81603b4c04e8ed134b80bace0810324
$ git commit -m'Added master.tar.gz'
[master b85a96f] Added master.tar.gz
git-fat filter-clean: caching to /tmp/repo/.git/fat/objects/b3489819f81603b4c04e8ed134b80bace0810324
1 file changed, 1 insertion(+)
create mode 100644 master.tar.gz
The patch itself is very simple and does not include the binary.
$ git show --pretty=oneline HEAD
918063043a6156172c2ad66478c6edd5c7df0217 Add master.tar.gz
diff --git a/master.tar.gz b/master.tar.gz
new file mode 100644
index 0000000..12f7d52
--- /dev/null
+++ b/master.tar.gz
@@ -0,0 +1 @@
+#$# git-fat 1f218834a137f7b185b498924e7a030008aee2ae
## Pushing fat files
Now let's push our fat files using the rsync configuration that we set up earlier.
$ git fat push
Pushing to localhost:/tmp/fat-store
building file list ...
1 file to consider
sent 61 bytes received 12 bytes 48.67 bytes/sec
total size is 6449 speedup is 88.34
We might normally set a remote now and push the git repository.
## Cloning and pulling
Now let's look at what happens when we clone.
$ cd ..
$ git clone repo repo2
Cloning into 'repo2'...
done.
$ cd repo2
$ git fat init # don't forget
$ ls -l # file is just a placeholder
total 4
-rw-r--r-- 1 jed users 53 Nov 25 22:42 master.tar.gz
$ cat master.tar.gz # holds the SHA1 of the file
#$# git-fat 1f218834a137f7b185b498924e7a030008aee2ae
We can always get a summary of what fat objects are missing in our local cache.
Orphan objects:
1f218834a137f7b185b498924e7a030008aee2ae
Now get any objects referenced by our current `HEAD`. This command also
accepts the `--all` option to pull full history, or a revision to pull
selected history.
$ git fat pull
receiving file list ...
1 file to consider
1f218834a137f7b185b498924e7a030008aee2ae
6449 100% 6.15MB/s 0:00:00 (xfer#1, to-check=0/1)
sent 30 bytes received 6558 bytes 4392.00 bytes/sec
total size is 6449 speedup is 0.98
Restoring 1f218834a137f7b185b498924e7a030008aee2ae -> master.tar.gz
git-fat filter-smudge: restoring from /tmp/repo2/.git/fat/objects/1f218834a137f7b185b498924e7a030008aee2ae
Everything is in place
$ git status
git-fat filter-clean: caching to /tmp/repo2/.git/fat/objects/1f218834a137f7b185b498924e7a030008aee2ae
# On branch master
nothing to commit, working directory clean
$ ls -l # recovered the full file
total 8
-rw-r--r-- 1 jed users 6449 Nov 25 17:10 master.tar.gz
## Summary
* Set the "fat" file types in `.gitattributes`.
* Use normal git commands to interact with the repository without
thinking about what files are fat and non-fat. The fat files will be
treated specially.
* Synchronize fat files with `git fat push` and `git fat pull`.
## Retroactive import using `git filter-branch` [Experimental]
Sometimes large objects were added to a repository by accident or for
lack of a better place to put them. _If_ you are willing to rewrite
history, forcing everyone to reclone, you can retroactively manage those
files with `git fat`. Be sure that you understand the consequences of
`git filter-branch` before attempting this. This feature is experimental
and irreversible, so be doubly careful with backups.
### Step 1: Locate the fat files
Run `git fat find THRESH_BYTES > fat-files` and inspect `fat-files` in
an editor. Lines will be sorted by the maximum object size that has been
at each path, and look like
something.big filter=fat -text # 8154677 1
where the first number after the `#` is the number of bytes and the
second number is the number of modifications that path has seen. You
will normally filter out some of these paths using grep and/or an
editor. When satisfied, remove the ends of the lines (including the `#`)
and append to `.gitattributes`. It's best to `git add .gitattributes` and commit
at this time (likely enrolling some extant files into `git fat`).
### Step 2: `filter-branch`
Copy `.gitattributes` to `/tmp/fat-filter-files` and edit to remove
everything after the file name (e.g., `sed s/ \+filter=fat.*$//`).
Currently, this may only contain exact paths relative to the root of the
repository. Finally, run
git filter-branch --index-filter \
'git fat index-filter /tmp/fat-filter-files --manage-gitattributes' \
--tag-name-filter cat -- --all
(You can remove the `--manage-gitattributes` option if you don't want to
append all the files being enrolled in `git fat` to `.gitattributes`,
however, future users would need to use `.git/info/attributes` to have
the `git fat` fileters run.)
When this finishes, inspect to see if everything is in order and follow
the
[Checklist for Shrinking a Repository](http://www.kernel.org/pub/software/scm/git/docs/git-filter-branch.html#_checklist_for_shrinking_a_repository)
in the `git filter-branch` man page, typically `git clone
file:///path/to/repo`. Be sure to `git fat push` from the original
repository.
See the script `test-retroactive.sh` for an example of cleaning.
## Implementation notes
The actual binary files are stored in `.git/fat/objects`, leaving `.git/objects` nice and small.
$ du -bs .git/objects
2212 .git/objects/
$ ls -l .git/fat/objects # This is where the file actually goes, but that's not important
total 8
-rw------- 1 jed users 6449 Nov 25 17:01 1f218834a137f7b185b498924e7a030008aee2ae
If you have multiple clones that access the same filesystem, you can make
`.git/fat/objects` a symlink to a common location, in which case all content
will be available in all repositories without extra copies. You still need to
`git fat push` to make it available to others.
# Some refinements
* Allow pulling and pushing only select files
* Relate orphan objects to file system
* Put some more useful message in smudged (working tree) version of missing files.
* More friendly configuration for multiple fat remotes
* Make commands safer in presence of a dirty tree.
* Private setting of a different remote.
* Gracefully handle unmanaged files when the filter is called (either
legacy files or files matching the pattern that should some reason not
be treated as fat).

@ -0,0 +1,628 @@
#!/usr/bin/env python
# -*- mode:python -*-
from __future__ import print_function, with_statement
import sys
import hashlib
import tempfile
import os
import subprocess
import shlex
import shutil
import itertools
import threading
import time
import collections
if not type(sys.version_info) is tuple and sys.version_info.major > 2:
sys.stderr.write('git-fat does not support Python-3 yet. Please use python2.\n')
sys.exit(1)
try:
from subprocess import check_output
del check_output
except ImportError:
def backport_check_output(*popenargs, **kwargs):
r"""Run command with arguments and return its output as a byte string.
Backported from Python 2.7 as it's implemented as pure python on stdlib.
>>> check_output(['/usr/bin/python', '--version'])
Python 2.6.2
"""
process = subprocess.Popen(stdout=subprocess.PIPE, *popenargs, **kwargs)
output, unused_err = process.communicate()
retcode = process.poll()
if retcode:
cmd = kwargs.get("args")
if cmd is None:
cmd = popenargs[0]
error = subprocess.CalledProcessError(retcode, cmd)
error.output = output
raise error
return output
subprocess.check_output = backport_check_output
BLOCK_SIZE = 4096
def verbose_stderr(*args, **kwargs):
return print(*args, file=sys.stderr, **kwargs)
def verbose_ignore(*args, **kwargs):
pass
def mkdir_p(path):
import errno
try:
os.makedirs(path)
except OSError as exc: # Python >2.5
if exc.errno == errno.EEXIST and os.path.isdir(path):
pass
else: raise
def umask():
"""Get umask without changing it."""
old = os.umask(0)
os.umask(old)
return old
def readblocks(stream):
bytes = 0
while True:
data = stream.read(BLOCK_SIZE)
bytes += len(data)
if not data:
break
yield data
def cat_iter(initer, outstream):
for block in initer:
outstream.write(block)
def cat(instream, outstream):
return cat_iter(readblocks(instream), outstream)
def difftreez_reader(input):
"""Incremental reader for git diff-tree -z output
:oldmode newmode oldsha1 newsha1 modflag\0filename\0:oldmode newmode ...
"""
buffer = []
partial = ''
while True:
newread = input.read(BLOCK_SIZE)
if not newread:
break
partial += newread
while True:
head, sep, partial = partial.partition('\0')
if not sep:
partial = head
break
buffer.append(head)
if len(buffer) == 2:
oldmode, newmode, oldhash, newhash, modflag = buffer[0].split()
path = buffer[1]
yield (newhash, modflag, path)
buffer = []
def gitconfig_get(name, file=None):
args = ['git', 'config', '--get']
if file is not None:
args += ['--file', file]
args.append(name)
p = subprocess.Popen(args, stdout=subprocess.PIPE)
output = p.communicate()[0].strip()
if p.returncode and file is None:
return None
elif p.returncode:
return gitconfig_get(name)
else:
return output
def gitconfig_set(name, value, file=None):
args = ['git', 'config']
if file is not None:
args += ['--file', file]
args += [name, value]
p = subprocess.check_call(args)
class GitFat(object):
DecodeError = RuntimeError
def __init__(self):
self.verbose = verbose_stderr if os.environ.get('GIT_FAT_VERBOSE') else verbose_ignore
try:
self.gitroot = subprocess.check_output('git rev-parse --show-toplevel'.split()).strip()
except subprocess.CalledProcessError:
sys.exit(1)
self.gitdir = subprocess.check_output('git rev-parse --git-dir'.split()).strip()
self.objdir = os.path.join(self.gitdir, 'fat', 'objects')
if os.environ.get('GIT_FAT_VERSION') == '1':
self.encode = self.encode_v1
else:
self.encode = self.encode_v2
def magiclen(enc):
return len(enc(hashlib.sha1('dummy').hexdigest(), 5))
self.magiclen = magiclen(self.encode) # Current version
self.magiclens = [magiclen(enc) for enc in [self.encode_v1, self.encode_v2]] # All prior versions
def setup(self):
mkdir_p(self.objdir)
def is_init_done(self):
return gitconfig_get('filter.fat.clean') or gitconfig_get('filter.fat.smudge')
def assert_init_done(self):
if not self.is_init_done():
sys.stderr.write('fatal: git-fat is not yet configured in this repository.\n')
sys.stderr.write('Run "git fat init" to configure.\n')
sys.exit(1)
def get_rsync(self):
cfgpath = os.path.join(self.gitroot,'.gitfat')
remote = gitconfig_get('rsync.remote', file=cfgpath)
ssh_port = gitconfig_get('rsync.sshport', file=cfgpath)
ssh_user = gitconfig_get('rsync.sshuser', file=cfgpath)
options = gitconfig_get('rsync.options', file=cfgpath)
if remote is None:
raise RuntimeError('No rsync.remote in %s' % cfgpath)
return remote, ssh_port, ssh_user, options
def get_rsync_command(self,push):
(remote, ssh_port, ssh_user, options) = self.get_rsync()
if push:
self.verbose('Pushing to %s' % (remote))
else:
self.verbose('Pulling from %s' % (remote))
cmd = ['rsync', '--progress', '--ignore-existing', '--from0', '--files-from=-']
rshopts = ''
if ssh_user:
rshopts += ' -l ' + ssh_user
if ssh_port:
rshopts += ' -p ' + ssh_port
if rshopts:
cmd.append('--rsh=ssh' + rshopts)
if options:
cmd += options.split(' ')
if push:
cmd += [self.objdir + '/', remote + '/']
else:
cmd += [remote + '/', self.objdir + '/']
return cmd
def revparse(self, revname):
return subprocess.check_output(['git', 'rev-parse', revname]).strip()
def encode_v1(self, digest, bytes):
'Produce legacy representation of file to be stored in repository.'
return '#$# git-fat %s\n' % (digest,)
def encode_v2(self, digest, bytes):
'Produce representation of file to be stored in repository. 20 characters can hold 64-bit integers.'
return '#$# git-fat %s %20d\n' % (digest, bytes)
def decode(self, string, noraise=False):
cookie = '#$# git-fat '
if string.startswith(cookie):
parts = string[len(cookie):].split()
digest = parts[0]
bytes = int(parts[1]) if len(parts) > 1 else None
return digest, bytes
elif noraise:
return None, None
else:
raise GitFat.DecodeError('Could not decode %s' % (string))
def decode_stream(self, stream):
'Return digest if git-fat cache, otherwise return iterator over entire file contents'
preamble = stream.read(self.magiclen)
try:
return self.decode(preamble)
except GitFat.DecodeError:
# Not sure if this is the right behavior
return itertools.chain([preamble], readblocks(stream)), None
def decode_file(self, fname):
# Fast check
try:
stat = os.lstat(fname)
except OSError:
return False, None
if stat.st_size != self.magiclen:
return False, None
# read file
try:
digest, bytes = self.decode_stream(open(fname))
except IOError:
return False, None
if isinstance(digest, str):
return digest, bytes
else:
return None, bytes
def decode_clean(self, body):
'''
Attempt to decode version in working tree. The tree version could be changed to have a more
useful message than the machine-readable copy that goes into the repository. If the tree
version decodes successfully, it indicates that the fat data is not currently available in
this repository.
'''
digest, bytes = self.decode(body, noraise=True)
return digest
def filter_clean(self, instream, outstreamclean):
h = hashlib.new('sha1')
bytes = 0
fd, tmpname = tempfile.mkstemp(dir=self.objdir)
try:
ishanging = False
cached = False # changes to True when file is cached
with os.fdopen(fd, 'w') as cache:
outstream = cache
firstblock = True
for block in readblocks(instream):
if firstblock:
if len(block) == self.magiclen and self.decode_clean(block[0:self.magiclen]):
ishanging = True # Working tree version is verbatim from repository (not smudged)
outstream = outstreamclean
firstblock = False
h.update(block)
bytes += len(block)
outstream.write(block)
outstream.flush()
digest = h.hexdigest()
objfile = os.path.join(self.objdir, digest)
if not ishanging:
if os.path.exists(objfile):
self.verbose('git-fat filter-clean: cache already exists %s' % objfile)
os.remove(tmpname)
else:
# Set permissions for the new file using the current umask
os.chmod(tmpname, int('444', 8) & ~umask())
os.rename(tmpname, objfile)
self.verbose('git-fat filter-clean: caching to %s' % objfile)
cached = True
outstreamclean.write(self.encode(digest, bytes))
finally:
if not cached:
os.remove(tmpname)
def cmd_filter_clean(self):
'''
The clean filter runs when a file is added to the index. It gets the "smudged" (tree)
version of the file on stdin and produces the "clean" (repository) version on stdout.
'''
self.setup()
self.filter_clean(sys.stdin, sys.stdout)
def cmd_filter_smudge(self):
self.setup()
result, bytes = self.decode_stream(sys.stdin)
if isinstance(result, str): # We got a digest
objfile = os.path.join(self.objdir, result)
try:
cat(open(objfile), sys.stdout)
self.verbose('git-fat filter-smudge: restoring from %s' % objfile)
except IOError: # file not found
self.verbose('git-fat filter-smudge: fat object missing %s' % objfile)
sys.stdout.write(self.encode(result, bytes)) # could leave a better notice about how to recover this file
else: # We have an iterable over the original input.
self.verbose('git-fat filter-smudge: not a managed file')
cat_iter(result, sys.stdout)
def catalog_objects(self):
return set(os.listdir(self.objdir))
def referenced_objects(self, rev=None, all=False):
referenced = set()
if all:
rev = '--all'
elif rev is None:
rev = self.revparse('HEAD')
# Revision list gives us object names to inspect with cat-file...
p1 = subprocess.Popen(['git','rev-list','--objects',rev], stdout=subprocess.PIPE)
def cut_sha1hash(input, output):
for line in input:
output.write(line.split()[0] + '\n')
output.close()
# ...`cat-file --batch-check` filters for git-fat object candidates in bulk...
p2 = subprocess.Popen(['git','cat-file','--batch-check'], stdin=subprocess.PIPE, stdout=subprocess.PIPE)
def filter_gitfat_candidates(input, output):
for line in input:
objhash, objtype, size = line.split()
if objtype == 'blob' and int(size) in self.magiclens:
output.write(objhash + '\n')
output.close()
# ...`cat-file --batch` provides full contents of git-fat candidates in bulk
p3 = subprocess.Popen(['git','cat-file','--batch'], stdin=subprocess.PIPE, stdout=subprocess.PIPE)
# Stream data: p1 | cut_thread | p2 | filter_thread | p3
cut_thread = threading.Thread(target=cut_sha1hash, args=(p1.stdout, p2.stdin))
filter_thread = threading.Thread(target=filter_gitfat_candidates, args=(p2.stdout, p3.stdin))
cut_thread.start()
filter_thread.start()
# Process metadata + content format provided by `cat-file --batch`
while True:
metadata_line = p3.stdout.readline()
if not metadata_line:
break # EOF
objhash, objtype, size_str = metadata_line.split()
size, bytes_read = int(size_str), 0
# We know from filter that item is a candidate git-fat object and
# is small enough to read into memory and process
content = ''
while bytes_read < size:
data = p3.stdout.read(size - bytes_read)
if not data:
break # EOF
content += data
bytes_read += len(data)
try:
fathash = self.decode(content)[0]
referenced.add(fathash)
except GitFat.DecodeError:
pass
# Consume LF record delimiter in `cat-file --batch` output
bytes_read = 0
while bytes_read < 1:
data = p3.stdout.read(1)
if not data:
break # EOF
bytes_read += len(data)
# Ensure everything is cleaned up
cut_thread.join()
filter_thread.join()
p1.wait()
p2.wait()
p3.wait()
return referenced
def orphan_files(self, patterns=[]):
'generator for all orphan placeholders in the working tree'
if not patterns or patterns == ['']:
patterns = ['.']
for fname in subprocess.check_output(['git', 'ls-files', '-z'] + patterns).split('\x00')[:-1]:
digest = self.decode_file(fname)[0]
if digest:
yield (digest, fname)
def cmd_status(self, args):
self.setup()
catalog = self.catalog_objects()
refargs = dict()
if '--all' in args:
refargs['all'] = True
referenced = self.referenced_objects(**refargs)
garbage = catalog - referenced
orphans = referenced - catalog
if '--all' in args:
for obj in referenced:
print(obj)
if orphans:
print('Orphan objects:')
for orph in orphans:
print(' ' + orph)
if garbage:
print('Garbage objects:')
for g in garbage:
print(' ' + g)
def is_dirty(self):
return subprocess.call(['git', 'diff-index', '--quiet', 'HEAD']) == 0
def cmd_push(self, args):
'Push anything that I have stored and referenced'
self.setup()
# Default to push only those objects referenced by current HEAD
# (includes history). Finer-grained pushing would be useful.
pushall = '--all' in args
files = self.referenced_objects(all=pushall) & self.catalog_objects()
cmd = self.get_rsync_command(push=True)
self.verbose('Executing: %s' % ' '.join(cmd))
p = subprocess.Popen(cmd, stdin=subprocess.PIPE)
p.communicate(input='\x00'.join(files))
if p.returncode:
sys.exit(p.returncode)
def checkout(self, show_orphans=False):
'Update any stale files in the present working tree'
self.assert_init_done()
for digest, fname in self.orphan_files():
objpath = os.path.join(self.objdir, digest)
if os.access(objpath, os.R_OK):
print('Restoring %s -> %s' % (digest, fname))
# The output of our smudge filter depends on the existence of
# the file in .git/fat/objects, but git caches the file stat
# from the previous time the file was smudged, therefore it
# won't try to re-smudge. I don't know a git command that
# specifically invalidates that cache, but changing the mtime
# on the file will invalidate the cache.
# Here we set the mtime to mtime + 1. This is an improvement
# over touching the file as it catches the edgecase where a
# git-checkout happens within the same second as a git fat
# checkout.
stat = os.lstat(fname)
os.utime(fname, (stat.st_atime, stat.st_mtime + 1))
# This re-smudge is essentially a copy that restores
# permissions.
subprocess.check_call(
['git', 'checkout-index', '--index', '--force', fname])
elif show_orphans:
print('Data unavailable: %s %s' % (digest,fname))
def cmd_pull(self, args):
'Pull anything that I have referenced, but not stored'
self.setup()
refargs = dict()
if '--all' in args:
refargs['all'] = True
for arg in args:
if arg.startswith('-') or len(arg) != 40:
continue
rev = self.revparse(arg)
if rev:
refargs['rev'] = rev
files = self.filter_objects(refargs, self.parse_pull_patterns(args))
cmd = self.get_rsync_command(push=False)
self.verbose('Executing: %s' % ' '.join(cmd))
p = subprocess.Popen(cmd, stdin=subprocess.PIPE)
p.communicate(input='\x00'.join(files))
if p.returncode:
sys.exit(p.returncode)
self.checkout()
def parse_pull_patterns(self, args):
if '--' not in args:
return ['']
else:
idx = args.index('--')
patterns = args[idx+1:] #we don't care about '--'
return patterns
def filter_objects(self, refargs, patterns):
files = self.referenced_objects(**refargs) - self.catalog_objects()
if refargs.get('all'): # Currently ignores patterns; can we efficiently do both?
return files
orphans_matched = list(self.orphan_files(patterns))
orphans_objects = set(map(lambda x: x[0], orphans_matched))
return files & orphans_objects
def cmd_checkout(self, args):
self.checkout(show_orphans=True)
def cmd_gc(self):
garbage = self.catalog_objects() - self.referenced_objects()
print('Unreferenced objects to remove: %d' % len(garbage))
for obj in garbage:
fname = os.path.join(self.objdir, obj)
print('%10d %s' % (os.stat(fname).st_size, obj))
os.remove(fname)
def cmd_verify(self):
"""Print details of git-fat objects with incorrect data hash"""
corrupted_objects = []
for obj in self.catalog_objects():
fname = os.path.join(self.objdir, obj)
h = hashlib.new('sha1')
for block in readblocks(open(fname)):
h.update(block)
data_hash = h.hexdigest()
if obj != data_hash:
corrupted_objects.append((obj, data_hash))
if corrupted_objects:
print('Corrupted objects: %d' % len(corrupted_objects))
for obj, data_hash in corrupted_objects:
print('%s data hash is %s' % (obj, data_hash))
sys.exit(1)
def cmd_init(self):
self.setup()
if self.is_init_done():
print('Git fat already configured, check configuration in .git/config')
else:
gitconfig_set('filter.fat.clean', 'git-fat filter-clean')
gitconfig_set('filter.fat.smudge', 'git-fat filter-smudge')
print('Initialized git fat')
def gen_large_blobs(self, revs, threshsize):
"""Build dict of all blobs"""
time0 = time.time()
def hash_only(input, output):
"""The output of git rev-list --objects shows extra info for blobs, subdirectory trees, and tags.
This truncates to one hash per line.
"""
for line in input:
output.write(line[:40] + '\n')
output.close()
revlist = subprocess.Popen(['git', 'rev-list', '--all', '--objects'], stdout=subprocess.PIPE, bufsize=-1)
objcheck = subprocess.Popen(['git', 'cat-file', '--batch-check'], stdin=subprocess.PIPE, stdout=subprocess.PIPE, bufsize=-1)
hashonly = threading.Thread(target=hash_only, args=(revlist.stdout, objcheck.stdin))
hashonly.start()
numblobs = 0; numlarge = 1
# Build dict with the sizes of all large blobs
for line in objcheck.stdout:
objhash, blob, size = line.split()
if blob != 'blob':
continue
size = int(size)
numblobs += 1
if size > threshsize:
numlarge += 1
yield objhash, size
revlist.wait()
objcheck.wait()
hashonly.join()
time1 = time.time()
self.verbose('%d of %d blobs are >= %d bytes [elapsed %.3fs]' % (numlarge, numblobs, threshsize, time1-time0))
def cmd_find(self, args):
maxsize = int(args[0])
blobsizes = dict(self.gen_large_blobs('--all', maxsize))
time0 = time.time()
# Find all names assumed by large blobs (those in blobsizes)
pathsizes = collections.defaultdict(lambda:set())
revlist = subprocess.Popen(['git', 'rev-list', '--all'], stdout=subprocess.PIPE, bufsize=-1)
difftree = subprocess.Popen(['git', 'diff-tree', '--root', '--no-renames', '--no-commit-id', '--diff-filter=AMCR', '-r', '--stdin', '-z'],
stdin=revlist.stdout, stdout=subprocess.PIPE)
for newblob, modflag, path in difftreez_reader(difftree.stdout):
bsize = blobsizes.get(newblob)
if bsize: # We care about this blob
pathsizes[path].add(bsize)
time1 = time.time()
self.verbose('Found %d paths in %.3f s' % (len(pathsizes), time1-time0))
maxlen = max(map(len,pathsizes)) if pathsizes else 0
for path, sizes in sorted(pathsizes.items(), key=lambda ps: max(ps[1]), reverse=True):
print('%-*s filter=fat -text # %10d %d' % (maxlen, path,max(sizes),len(sizes)))
revlist.wait()
difftree.wait()
def cmd_index_filter(self, args):
manage_gitattributes = '--manage-gitattributes' in args
filelist = set(f.strip() for f in open(args[0]).readlines())
lsfiles = subprocess.Popen(['git', 'ls-files', '-s'], stdout=subprocess.PIPE)
updateindex = subprocess.Popen(['git', 'update-index', '--index-info'], stdin=subprocess.PIPE)
for line in lsfiles.stdout:
mode, sep, tail = line.partition(' ')
blobhash, sep, tail = tail.partition(' ')
stageno, sep, tail = tail.partition('\t')
filename = tail.strip()
if filename not in filelist:
continue
if mode == "120000":
# skip symbolic links
continue
# This file will contain the hash of the cleaned object
hashfile = os.path.join(self.gitdir, 'fat', 'index-filter', blobhash)
try:
cleanedobj = open(hashfile).read().rstrip()
except IOError:
catfile = subprocess.Popen(['git', 'cat-file', 'blob', blobhash], stdout=subprocess.PIPE)
hashobject = subprocess.Popen(['git', 'hash-object', '-w', '--stdin'], stdin=subprocess.PIPE, stdout=subprocess.PIPE)
def dofilter():
self.filter_clean(catfile.stdout, hashobject.stdin)
hashobject.stdin.close()
filterclean = threading.Thread(target=dofilter)
filterclean.start()
cleanedobj = hashobject.stdout.read().rstrip()
catfile.wait()
hashobject.wait()
filterclean.join()
mkdir_p(os.path.dirname(hashfile))
open(hashfile, 'w').write(cleanedobj + '\n')
updateindex.stdin.write('%s %s %s\t%s\n' % (mode, cleanedobj, stageno, filename))
if manage_gitattributes:
try:
mode, blobsha1, stageno, filename = subprocess.check_output(['git', 'ls-files', '-s', '.gitattributes']).split()
gitattributes_lines = subprocess.check_output(['git', 'cat-file', 'blob', blobsha1]).splitlines()
except ValueError: # Nothing to unpack, thus no file
mode, stageno = '100644', '0'
gitattributes_lines = []
gitattributes_extra = ['%s filter=fat -text' % line.split()[0] for line in filelist]
hashobject = subprocess.Popen(['git', 'hash-object', '-w', '--stdin'], stdin=subprocess.PIPE, stdout=subprocess.PIPE)
stdout, stderr = hashobject.communicate('\n'.join(gitattributes_lines + gitattributes_extra) + '\n')
updateindex.stdin.write('%s %s %s\t%s\n' % (mode, stdout.strip(), stageno, '.gitattributes'))
updateindex.stdin.close()
lsfiles.wait()
updateindex.wait()
if __name__ == '__main__':
fat = GitFat()
cmd = sys.argv[1] if len(sys.argv) > 1 else ''
if cmd == 'filter-clean':
fat.cmd_filter_clean()
elif cmd == 'filter-smudge':
fat.cmd_filter_smudge()
elif cmd == 'init':
fat.cmd_init()
elif cmd == 'status':
fat.cmd_status(sys.argv[2:])
elif cmd == 'push':
fat.cmd_push(sys.argv[2:])
elif cmd == 'pull':
fat.cmd_pull(sys.argv[2:])
elif cmd == 'gc':
fat.cmd_gc()
elif cmd == 'verify':
fat.cmd_verify()
elif cmd == 'checkout':
fat.cmd_checkout(sys.argv[2:])
elif cmd == 'find':
fat.cmd_find(sys.argv[2:])
elif cmd == 'index-filter':
fat.cmd_index_filter(sys.argv[2:])
else:
print('Usage: git fat [init|status|push|pull|gc|verify|checkout|find|index-filter]', file=sys.stderr)

@ -0,0 +1,53 @@
#!/bin/sh -ex
fullpath() { echo "`pwd`/$1"; }
git init retro
cd retro
cp /usr/share/dict/words words.big
chmod u+w words.big
git add words.big
git commit -m'Add big file without using git-fat'
sort words.big > sorted.big
git add sorted.big
git commit -m'Add sorted file without using git-fat'
cat > .gitattributes <<EOF
original-attributes -text
EOF
git add .gitattributes
echo 'truncated' > words.big
git commit -am'Truncated words.big and add .gitattributes'
git fat init
cat > .gitattributes <<EOF
*.big filter=fat -text
EOF
git add .gitattributes
git checkout .
git commit -am'Import big files into git-fat'
git log --stat
git fat find 10000 | awk '{print $1}' > fat-files
git filter-branch --index-filter "git fat index-filter $(fullpath fat-files) --manage-gitattributes" --tag-name-filter cat -- --all
git log --stat
git checkout HEAD^
rm *
git checkout .
ls -al
# Set up place to push
git checkout master
cat > .gitfat <<EOF
[rsync]
remote = $(fullpath ../retro-store)
EOF
git add .gitfat
git commit -m'Add .gitfat for local push'
git fat push
cd ..
git clone file:///$(fullpath retro) retro-clone
cd retro-clone
git fat init
git fat pull

@ -0,0 +1,64 @@
#!/bin/bash -ex
# Any copyright is dedicated to the Public Domain.
# http://creativecommons.org/publicdomain/zero/1.0/
# Clear out repos and fat store from prior test runs
rm -fR fat-test fat-test2 /tmp/fat-store
git init fat-test
cd fat-test
git fat init
cat - >> .gitfat <<EOF
[rsync]
remote = localhost:/tmp/fat-store
EOF
echo '*.fat filter=fat -crlf' > .gitattributes
git add .gitattributes .gitfat
git commit -m'Initial fat repository'
ln -s /oe/dss-oe/dss-add-ons-testing-build/deploy/licenses/common-licenses/GPL-3 c
git add c
git commit -m'add broken symlink'
echo 'fat content a' > a.fat
git add a.fat
git commit -m'add a.fat'
echo 'fat content b' > b.fat
git add b.fat
git commit -m'add b.fat'
echo 'revise fat content a' > a.fat
git commit -am'revise a.fat'
git fat push
cd ..
git clone fat-test fat-test2
cd fat-test2
# checkout and pull should fail in repo not yet init'ed for git-fat
git fat checkout && true
if [ $? -eq 0 ]
then
echo 'ERROR: "git fat checkout" in uninitialised repo should fail'
exit 1
fi
git fat pull -- 'a.fa*' && true
if [ $? -eq 0 ]
then
echo 'ERROR: "git fat pull" in uninitialised repo should fail'
exit 1
fi
git fat init
git fat pull -- 'a.fa*'
cat a.fat
echo 'file which is committed and removed afterwards' > d
git add d
git commit -m'add d with normal content'
rm d
git fat pull
# Check verify command finds corrupt object
mv .git/fat/objects/6ecec2e21d3033e7ba53e2db63f69dbd3a011fa8 \
.git/fat/objects/6ecec2e21d3033e7ba53e2db63f69dbd3a011fa8.bak
echo "Not the right data" > .git/fat/objects/6ecec2e21d3033e7ba53e2db63f69dbd3a011fa8
git fat verify && true
if [ $? -eq 0 ]; then echo "Verify did not detect invalid object"; exit 1; fi
mv .git/fat/objects/6ecec2e21d3033e7ba53e2db63f69dbd3a011fa8.bak \
.git/fat/objects/6ecec2e21d3033e7ba53e2db63f69dbd3a011fa8