Do a “git export” (like “svn export”)?

Question

I've been wondering whether there is a good "git export" solution that creates a copy of a tree without the .git repository directory. There are at least three methods I know of:

1. git clone followed by removing the .git repository directory.
2. git checkout-index alludes to this functionality but starts with "Just read the desired tree into the index..." which I'm not entirely sure how to do.
3. git-export is a third party script that essentially does a git clone into a temporary location followed by rsync --exclude='.git' into the final destination.

None of these solutions really strike me as being satisfactory. The closest one to svn export might be option 1, because both those require the target directory to be empty first. But option 2 seems even better, assuming I can figure out what it means to read a tree into the index.

Answer 1

I think @Aredridel's post was closest, but there's a bit more to that - so I will add this here; the thing is, in svn, if you're in a subfolder of a repo, and you do:

/media/disk/repo_svn/subdir$ svn export . /media/disk2/repo_svn_B/subdir

then svn will export all files that are under revision control (they could have also freshly Added; or Modified status) - and if you have other "junk" in that directory (and I'm not counting .svn subfolders here, but visible stuff like .o files), it will not be exported; only those files registered by the SVN repo will be exported. For me, one nice thing is that this export also includes files with local changes that have not been committed yet; and another nice thing is that the timestamps of the exported files are the same as the original ones. Or, as svn help export puts it:

2. Exports a clean directory tree from the working copy specified by PATH1, at revision REV if it is given, otherwise at WORKING, into PATH2. ... If REV is not specified, all local changes will be preserved. Files not under version control will not be copied.

To realize that git will not preserve the timestamps, compare the output of these commands (in a subfolder of a git repo of your choice):

/media/disk/git_svn/subdir$ ls -la .

... and:

/media/disk/git_svn/subdir$ git archive --format=tar --prefix=junk/ HEAD | (tar -t -v --full-time -f -)

... and I, in any case, notice that git archive causes all the timestamps of the archived file to be the same! git help archive says:

git archive behaves differently when given a tree ID versus when given a commit ID or tag ID. In the first case the current time is used as the modification time of each file in the archive. In the latter case the commit time as recorded in the referenced commit object is used instead.

... but apparently both cases set the "modification time of each file"; thereby not preserving the actual timestamps of those files!

So, in order to also preserve the timestamps, here is a bash script, which is actually a "one-liner", albeit somewhat complicated - so below it is posted in multiple lines:

/media/disk/git_svn/subdir$ git archive --format=tar master | (tar tf -) | (\
  DEST="/media/diskC/tmp/subdirB"; \
  CWD="$PWD"; \
  while read line; do \
    DN=$(dirname "$line"); BN=$(basename "$line"); \
    SRD="$CWD"; TGD="$DEST"; \
    if [ "$DN" != "." ]; then \
      SRD="$SRD/$DN" ; TGD="$TGD/$DN" ; \
      if [ ! -d "$TGD" ] ; then \
        CMD="mkdir \"$TGD\"; touch -r \"$SRD\" \"$TGD\""; \
        echo "$CMD"; \
        eval "$CMD"; \
      fi; \
    fi; \
    CMD="cp -a \"$SRD/$BN\" \"$TGD/\""; \
    echo "$CMD"; \
    eval "$CMD"; \
    done \
)

Note that it is assumed that you're exporting the contents in "current" directory (above, /media/disk/git_svn/subdir) - and the destination you're exporting into is somewhat inconveniently placed, but it is in DEST environment variable. Note that with this script; you must create the DEST directory manually yourself, before running the above script.

After the script is ran, you should be able to compare:

ls -la /media/disk/git_svn/subdir
ls -la /media/diskC/tmp/subdirB   # DEST

... and hopefully see the same timestamps (for those files that were under version control).

Hope this helps someone,
Cheers!

Answer 2

If you need submodules as well, this should do the trick: https://github.com/meitar/git-archive-all.sh/wiki

Answer 3

I have hit this page frequently when looking for a way to export a git repository. My answer to this question considers three properties that svn export has by design compared to git, since svn follows a centralized repository approach:

• It minimizes the traffic to a remote repository location by not exporting all revisions
• It does not include meta information in the export directory

• Exporting a certain branch using svn is accomplished by specifying the appropriate path

  git clone --depth 1 --branch master git://git.somewhere destination_path
  rm -rf destination_path/.git
  

When building a certain release it is useful to clone a stable branch as for example --branch stable or --branch release/0.9.

Answer 4

git archive also works with remote repository.

git archive --format=tar \
--remote=ssh://remote_server/remote_repository master | tar -xf -

To export particular path inside the repo add as many paths as you wish as last argument to git, e.g.:

git archive --format=tar \
--remote=ssh://remote_server/remote_repository master path1/ path2/ | tar -xv

Answer 5

Bash-implementation of git-export.

I have segmented the .empty file creation and removal processes on their own function, with the purpose of re-using them in the 'git-archive' implementation (will be posted later on).

I have also added the '.gitattributes' file to the process in order to remove un-wanted files from the target export folder. Included verbosity to the process while making the 'git-export' function more efficient.

EMPTY_FILE=".empty";

function create_empty () {
## Processing path (target-dir):
    TRG_PATH="${1}";
## Component(s):
    EXCLUDE_DIR=".git";
echo -en "\nAdding '${EMPTY_FILE}' files to empty folder(s): ...";
    find ${TRG_PATH} -not -path "*/${EXCLUDE_DIR}/*" -type d -empty -exec touch {}/${EMPTY_FILE} \;
#echo "done.";
## Purging SRC/TRG_DIRs variable(s):
    unset TRG_PATH EMPTY_FILE EXCLUDE_DIR;
    return 0;
  }

declare -a GIT_EXCLUDE;
function load_exclude () {
    SRC_PATH="${1}";
    ITEMS=0; while read LINE; do
#      echo -e "Line [${ITEMS}]: '${LINE%%\ *}'";
      GIT_EXCLUDE[((ITEMS++))]=${LINE%%\ *};
    done < ${SRC_PATH}/.gitattributes;
    GIT_EXCLUDE[${ITEMS}]="${EMPTY_FILE}";
## Purging variable(s):
    unset SRC_PATH ITEMS;
    return 0;
  }

function purge_empty () {
## Processing path (Source/Target-dir):
    SRC_PATH="${1}";
    TRG_PATH="${2}";
echo -e "\nPurging Git-Specific component(s): ... ";
    find ${SRC_PATH} -type f -name ${EMPTY_FILE} -exec /bin/rm '{}' \;
    for xRULE in ${GIT_EXCLUDE[@]}; do
echo -en "    '${TRG_PATH}/{${xRULE}}' files ... ";
      find ${TRG_PATH} -type f -name "${xRULE}" -exec /bin/rm -rf '{}' \;
echo "done.'";
    done;
echo -e "done.\n"
## Purging SRC/TRG_PATHs variable(s):
    unset SRC_PATH; unset TRG_PATH;
    return 0;
  }

function git-export () {
    TRG_DIR="${1}"; SRC_DIR="${2}";
    if [ -z "${SRC_DIR}" ]; then SRC_DIR="${PWD}"; fi
    load_exclude "${SRC_DIR}";
## Dynamically added '.empty' files to the Git-Structure:
    create_empty "${SRC_DIR}";
    GIT_COMMIT="Including '${EMPTY_FILE}' files into Git-Index container."; #echo -e "\n${GIT_COMMIT}";
    git add .; git commit --quiet --all --verbose --message "${GIT_COMMIT}";
    if [ "${?}" -eq 0 ]; then echo " done."; fi
    /bin/rm -rf ${TRG_DIR} && mkdir -p "${TRG_DIR}";
echo -en "\nChecking-Out Index component(s): ... ";
    git checkout-index --prefix=${TRG_DIR}/ -q -f -a
## Reset: --mixed = reset HEAD and index:
    if [ "${?}" -eq 0 ]; then
echo "done."; echo -en "Resetting HEAD and Index: ... ";
        git reset --soft HEAD^;
        if [ "${?}" -eq 0 ]; then
echo "done.";
## Purging Git-specific components and '.empty' files from Target-Dir:
            purge_empty "${SRC_DIR}" "${TRG_DIR}"
          else echo "failed.";
        fi
## Archiving exported-content:
echo -en "Archiving Checked-Out component(s): ... ";
        if [ -f "${TRG_DIR}.tgz" ]; then /bin/rm ${TRG_DIR}.tgz; fi
        cd ${TRG_DIR} && tar -czf ${TRG_DIR}.tgz ./; cd ${SRC_DIR}
echo "done.";
## Listing *.tgz file attributes:
## Warning: Un-TAR this file to a specific directory:
        ls -al ${TRG_DIR}.tgz
      else echo "failed.";
    fi
## Purgin all references to Un-Staged File(s):
   git reset HEAD;
## Purging SRC/TRG_DIRs variable(s):
    unset SRC_DIR; unset TRG_DIR;
    echo "";
    return 0;
  }

Output:

$ git-export /tmp/rel-1.0.0

Adding '.empty' files to empty folder(s): ... done.

Checking-Out Index component(s): ... done.

Resetting HEAD and Index: ... done.

Purging Git-Specific component(s): ...

'/tmp/rel-1.0.0/{.buildpath}' files ... done.'

'/tmp/rel-1.0.0/{.project}' files ... done.'

'/tmp/rel-1.0.0/{.gitignore}' files ... done.'

'/tmp/rel-1.0.0/{.git}' files ... done.'

'/tmp/rel-1.0.0/{.gitattributes}' files ... done.'

'/tmp/rel-1.0.0/{*.mno}' files ... done.'

'/tmp/rel-1.0.0/{*~}' files ... done.'

'/tmp/rel-1.0.0/{.*~}' files ... done.'

'/tmp/rel-1.0.0/{*.swp}' files ... done.'

'/tmp/rel-1.0.0/{*.swo}' files ... done.'

'/tmp/rel-1.0.0/{.DS_Store}' files ... done.'

'/tmp/rel-1.0.0/{.settings}' files ... done.'

'/tmp/rel-1.0.0/{.empty}' files ... done.'

done.

Archiving Checked-Out component(s): ... done.

-rw-r--r-- 1 admin wheel 25445901 3 Nov 12:57 /tmp/rel-1.0.0.tgz

I have now incorporated the 'git archive' functionality into a single process that makes use of 'create_empty' function and other features.

function git-archive () {
    PREFIX="${1}"; ## sudo mkdir -p ${PREFIX}
    REPO_PATH="`echo "${2}"|awk -F: '{print $1}'`";
    RELEASE="`echo "${2}"|awk -F: '{print $2}'`";
    USER_PATH="${PWD}";
echo "$PREFIX $REPO_PATH $RELEASE $USER_PATH";
## Dynamically added '.empty' files to the Git-Structure:
    cd "${REPO_PATH}"; populate_empty .; echo -en "\n";
#    git archive --prefix=git-1.4.0/ -o git-1.4.0.tar.gz v1.4.0
# e.g.: git-archive /var/www/htdocs /repos/domain.name/website:rel-1.0.0 --explode
    OUTPUT_FILE="${USER_PATH}/${RELEASE}.tar.gz";
    git archive --verbose --prefix=${PREFIX}/ -o ${OUTPUT_FILE} ${RELEASE}
    cd "${USER_PATH}";
    if [[ "${3}" =~ [--explode] ]]; then
      if [ -d "./${RELEASE}" ]; then /bin/rm -rf "./${RELEASE}"; fi
      mkdir -p ./${RELEASE}; tar -xzf "${OUTPUT_FILE}" -C ./${RELEASE}
    fi
## Purging SRC/TRG_DIRs variable(s):
    unset PREFIX REPO_PATH RELEASE USER_PATH OUTPUT_FILE;
    return 0;
  }

Answer 6

My preference would actually be to have a dist target in your Makefile (or other build system) that exports a distributable archive of your code (.tar.bz2, .zip, .jar, or whatever is appropriate). If you happen to be using GNU autotools or Perl's MakeMaker systems, I think this exists for you automatically. If not, I highly recommend adding it.

ETA (2012-09-06): Wow, harsh downvotes. I still believe it is better to build your distributions with your build tools rather than your source code control tool. I believe in building artifacts with build tools. In my current job, our main product is built with an ant target. We are in the midst of switching source code control systems, and the presence of this ant target means one less hassle in migration.