How to make shallow git submodules?

Question

Is it possible to have shallow submodules? I have a superproject with several submodules, each with a long history, so it gets unnecessarily big dragging all that history.

All I have found is this unanswered thread.

Should I just hack git-submodule to implement this?

Answer 1

Are the canonical locations for your submodules remote? If so, are you OK with cloning them once? In other words, do you want the shallow clones just because you are suffering the wasted bandwidth of frequent submodule (re)clones?

If you want shallow clones to save local diskspace, then Ryan Graham's answer seems like a good way to go. Manually clone the repositories so that they are shallow. If you think it would be useful, adapt git submodule to support it. Send an email to the list asking about it (advice for implementing it, suggestions on the interface, etc.). In my opinion, the folks there are quite supportive of potential contributors that earnestly want to enhance Git in constructive ways.

If you are OK with doing one full clone of each submodule (plus later fetches to keep them up to date), you might try using the --reference option of git submodule update (it is in Git 1.6.4 and later) to refer to local object stores (e.g. make --mirror clones of the canonical submodule repositories, then use --reference in your submodules to point to these local clones). Just be sure to read about git clone --reference/git clone --shared before using --reference. The only likely problem with referencing mirrors would be if they ever end up fetching non-fast-forward updates (though you could enable reflogs and expand their expiration windows to help retain any abandoned commits that might cause a problem). You should not have any problems as long as

• you do not make any local submodule commits, or
• any commits that are left dangling by non-fast-forwards that the canonical repositories might publish are not ancestors to your local submodule commits, or
• you are diligent about keeping your local submodule commits rebased on top of whatever non-fast-forwards might be published in the canonical submodule repositories.

If you go with something like this and there is any chance that you might carry local submodule commits in your working trees, it would probably be a good idea to create an automated system that makes sure critical objects referenced by the checked-out submodules are not left dangling in the mirror repositories (and if any are found, copies them to the repositories that need them).

And, like the git clone manpage says, do not use --reference if you do not understand these implications.

# Full clone (mirror), done once.
git clone --mirror $sub1_url $path_to_mirrors/$sub1_name.git
git clone --mirror $sub2_url $path_to_mirrors/$sub2_name.git

# Reference the full clones any time you initialize a submodule
git clone $super_url super
cd super
git submodule update --init --reference $path_to_mirrors/$sub1_name.git $sub1_path_in_super
git submodule update --init --reference $path_to_mirrors/$sub2_name.git $sub2_path_in_super

# To avoid extra packs in each of the superprojects' submodules,
#   update the mirror clones before any pull/merge in super-projects.
for p in $path_to_mirrors/*.git; do GIT_DIR="$p" git fetch; done

cd super
git pull             # merges in new versions of submodules
git submodule update # update sub refs, checkout new versions,
                     #   but no download since they reference the updated mirrors

---

Alternatively, instead of --reference, you could use the mirror clones in combination with the default hardlinking functionality of git clone by using local mirrors as the source for your submodules. In new super-project clones, do git submodule init, edit the submodule URLs in .git/config to point to the local mirrors, then do git submodule update. You would need to reclone any existing checked-out submodules to get the hardlinks. You would save bandwidth by only downloading once into the mirrors, then fetching locally from those into your checked-out submodules. The hard linking would save disk space (although fetches would tend to accumulate and be duplicated across multiple instances of the checked-out submodules' object stores; you could periodically reclone the checked-out submodules from the mirrors to regain the disk space saving provided by hardlinking).

Answer 2

New in the upcoming git1.8.4 (July 2013):

"git submodule update" can optionally clone the submodule repositories shallowly.

(And git 2.10 Q3 2016 allows to record that with git config -f .gitmodules submodule.<name>.shallow true.
See the end of this answer)

See commit 275cd184d52b5b81cb89e4ec33e540fb2ae61c1f:

Add the --depth option to the add and update commands of "git submodule", which is then passed on to the clone command. This is useful when the submodule(s) are huge and you're not really interested in anything but the latest commit.

Tests are added and some indention adjustments were made to conform to the rest of the testfile on "submodule update can handle symbolic links in pwd".

Signed-off-by: Fredrik Gustafsson <[e-mail]>
Acked-by: Jens Lehmann <[e-mail]>

That means this works:

git submodule add --depth 1 -- repository path
git submodule update --depth -- [<path>...]

With:

--depth::

This option is valid for add and update commands.
Create a 'shallow' clone with a history truncated to the specified number of revisions.

---

atwyman adds in the comments:

As far as I can tell this option isn't usable for submodules which don't track master very closely. If you set depth 1, then submodule update can only ever succeed if the submodule commit you want is the latest master. Otherwise you get "fatal: reference is not a tree".

That is true.
That is, until git 2.8 (March 2016). With 2.8, the submodule update --depth has one more chance to succeed, even if the SHA1 is directly reachable from one of the remote repo HEADs.

See commit fb43e31 (24 Feb 2016) by Stefan Beller (stefanbeller).
Helped-by: Junio C Hamano (gitster).
^{(Merged by Junio C Hamano -- gitster -- in commit 9671a76, 26 Feb 2016)}

submodule: try harder to fetch needed sha1 by direct fetching sha1

When reviewing a change that also updates a submodule in Gerrit, a common review practice is to download and cherry-pick the patch locally to test it.
However when testing it locally, the 'git submodule update' may fail fetching the correct submodule sha1 as the corresponding commit in the submodule is not yet part of the project history, but also just a proposed change.

If $sha1 was not part of the default fetch, we try to fetch the $sha1 directly. Some servers however do not support direct fetch by sha1, which leads git-fetch to fail quickly.
We can fail ourselves here as the still missing sha1 would lead to a failure later in the checkout stage anyway, so failing here is as good as we can get.

---

MVG points out in the comments to commit fb43e31 (git 2.9, Feb 2016)

It would seem to me that commit fb43e31 requests the missing commit by SHA1 id, so the uploadpack.allowReachableSHA1InWant and uploadpack.allowTipSHA1InWant settings on the server will probably affect whether this works.
I wrote a post to the git list today, pointing out how the use of shallow submodules could be made to work better for some scenarios, namely if the commit is also a tag.
Let's wait and see.

I guess this is a reason why fb43e31 made the fetch for a specific SHA1 a fallback after the fetch for the default branch.
Nevertheless, in case of “--depth 1” I think it would make sense to abort early: if none of the listed refs matches the requested one, and asking by SHA1 isn't supported by the server, then there is no point in fetching anything, since we won't be able to satisfy the submodule requirement either way.

---

Update August 2016 (3 years later)

With Git 2.10 (Q3 2016), you will be able to do

 git config -f .gitmodules submodule.<name>.shallow true

See "Git submodule without extra weight" for more.

---

Git 2.13 (Q2 2017) do add in commit 8d3047c (19 Apr 2017) by Sebastian Schuberth (sschuberth).
^{(Merged by Sebastian Schuberth -- sschuberth -- in commit 8d3047c, 20 Apr 2017)}

a clone of this submodule will be performed as a shallow clone (with a history depth of 1)

However, Ciro Santilli adds in the comments (and details in his answer)

shallow = true on .gitmodules only affects the reference tracked by the HEAD of the remote when using --recurse-submodules, even if the target commit is pointed to by a branch, and even if you put branch = mybranch on the .gitmodules as well.

---

Git 2.20 (Q4 2018) improves on the submodule support, which has been updated to read from the blob at HEAD:.gitmodules when the .gitmodules file is missing from the working tree.

See commit 2b1257e, commit 76e9bdc (25 Oct 2018), and commit b5c259f, commit 23dd8f5, commit b2faad4, commit 2502ffc, commit 996df4d, commit d1b13df, commit 45f5ef3, commit bcbc780 (05 Oct 2018) by Antonio Ospite (ao2).
^{(Merged by Junio C Hamano -- gitster -- in commit abb4824, 13 Nov 2018)}

submodule: support reading .gitmodules when it's not in the working tree

When the .gitmodules file is not available in the working tree, try using the content from the index and from the current branch.
This covers the case when the file is part of the repository but for some reason it is not checked out, for example because of a sparse checkout.

This makes it possible to use at least the 'git submodule' commands which read the gitmodules configuration file without fully populating the working tree.

Writing to .gitmodules will still require that the file is checked out, so check for that before calling config_set_in_gitmodules_file_gently.

Add a similar check also in git-submodule.sh::cmd_add() to anticipate the eventual failure of the "git submodule add" command when .gitmodules is not safely writeable; this prevents the command from leaving the repository in a spurious state (e.g. the submodule repository was cloned but .gitmodules was not updated because config_set_in_gitmodules_file_gently failed).

Moreover, since config_from_gitmodules() now accesses the global object store, it is necessary to protect all code paths which call the function against concurrent access to the global object store.
Currently this only happens in builtin/grep.c::grep_submodules(), so call grep_read_lock() before invoking code involving config_from_gitmodules().

NOTE: there is one rare case where this new feature does not work properly yet: nested submodules without .gitmodules in their working tree.

Answer 3

Reading through the git-submodule "source", it looks like git submodule add can handle submodules that already have their repositories present. In that case...

$ git clone $remote1 $repo
$ cd $repo
$ git clone --depth 5 $remotesub1 $sub1
$ git submodule add $remotesub1 $sub1
#repeat as necessary...

You'll want to make sure the required commit is in the submodule repo, so make sure you set an appropriate --depth.

Edit: You may be able to get away with multiple manual submodule clones followed by a single update:

$ git clone $remote1 $repo
$ cd $repo
$ git clone --depth 5 $remotesub1 $sub1
#repeat as necessary...
$ git submodule update

Answer 4

Summary of buggy / unexpected / annoying behaviour as of Git 2.14.1

1. shallow = true in .gitmodules only affects git clone --recurse-submodules if the HEAD of the remote submodule points to the required commit, even if the target commit is pointed to by a branch, and even if you put branch = mybranch on the .gitmodules as well.

   Local test script. Same behaviour on GitHub 2017-11, where HEAD is controlled by the default branch repo setting:

   git clone --recurse-submodules https://github.com/cirosantilli/test-shallow-submodule-top-branch-shallow
   cd test-shallow-submodule-top-branch-shallow/mod
   git log
   # Multiple commits, not shallow.
   

2. git clone --recurse-submodules --shallow-submodules fails if the commit is neither referenced by a branch or tag with a message: error: Server does not allow request for unadvertised object.

   Local test script. Same behaviour on GitHub:

   git clone --recurse-submodules --shallow-submodules https://github.com/cirosantilli/test-shallow-submodule-top-sha
   # error
   

   I also asked on the mailing list: https://marc.info/?l=git&m=151863590026582&w=2 and the reply was:

   In theory this should be easy. :)

   In practice not so much, unfortunately. This is because cloning will just obtain the latest tip of a branch (usually master). There is no mechanism in clone to specify the exact sha1 that is wanted.

   The wire protocol supports for asking exact sha1s, so that should be covered. (Caveat: it only works if the server operator enables uploadpack.allowReachableSHA1InWant which github has not AFAICT)

   git-fetch allows to fetch arbitrary sha1, so as a workaround you can run a fetch after the recursive clone by using "git submodule update" as that will use fetches after the initial clone.

TODO test: allowReachableSHA1InWant.

Answer 5

I created a slightly different version, for when it's not running at the bleeding edge, which not all projects do. The standard submodule additions did't work nor did the script above. So I added a hash lookup for the tag ref, and if it doesn't have one, it falls back to full clone.

#!/bin/bash
git submodule init
git submodule | while read hash name junk; do
    spath=$(git config -f .gitmodules --get submodule.$name.path)
    surl=$(git config -f .gitmodules --get submodule.$name.url)
    sbr=$(git ls-remote --tags $surl | sed -r "/${hash:1}/ s|^.*tags/([^^]+).*\$|\1|p;d")
    if [ -z $sbr ]; then
        git clone $surl $spath
    else
        git clone -b $sbr --depth 1 --single-branch $surl $spath
    fi
done
git submodule update 

Answer 6

Git 2.9.0 support submodules shallow clone directly, so now you can just call:

git clone url://to/source/repository --recursive --shallow-submodules