204

Is there any way to retrieve only one specific commit from a remote Git repo without cloning it on my PC? The structure of remote repo is absolutely same as that of mine and hence there won't be any conflicts but I have no idea how to do this and I don't want to clone that huge repository.

I am new to git, is there any way?

Daniel
  • 17,803
  • 7
  • 74
  • 142
Varun Chitre
  • 2,681
  • 4
  • 17
  • 31

10 Answers10

112

Starting with Git version 2.5+ (Q2 2015), fetching a single commit (without cloning the full repo) is actually possible.

See commit 68ee628 by Fredrik Medley (moroten), 21 May 2015.
(Merged by Junio C Hamano -- gitster -- in commit a9d3493, 01 Jun 2015)

You now have a new config (on the server side)

uploadpack.allowReachableSHA1InWant

Allow upload-pack to accept a fetch request that asks for an object that is reachable from any ref tip. However, note that calculating object reachability is computationally expensive.
Defaults to false.

If you combine that server-side configuration with a shallow clone (git fetch --depth=1), you can ask for a single commit (see t/t5516-fetch-push.sh:

git fetch --depth=1 ../testrepo/.git <full-length SHA1>

You can use the git cat-file command to see that the commit has been fetched:

git cat-file commit <full-length SHA1>

"git upload-pack" that serves "git fetch" can be told to serve commits that are not at the tip of any ref, as long as they are reachable from a ref, with uploadpack.allowReachableSHA1InWant configuration variable.

As noted by matt in the comments:

Note that SHA must be the full unabbreviated SHA, otherwise Git will claim it couldn't find the commit

The full documentation is:

upload-pack: optionally allow fetching reachable sha1

With uploadpack.allowReachableSHA1InWant configuration option set on the server side, "git fetch" can make a request with a "want" line that names an object that has not been advertised (likely to have been obtained out of band or from a submodule pointer).
Only objects reachable from the branch tips, i.e. the union of advertised branches and branches hidden by transfer.hideRefs, will be processed.
Note that there is an associated cost of having to walk back the history to check the reachability.

This feature can be used when obtaining the content of a certain commit, for which the sha1 is known, without the need of cloning the whole repository, especially if a shallow fetch is used.

Useful cases are e.g.

  • repositories containing large files in the history,
  • fetching only the needed data for a submodule checkout,
  • when sharing a sha1 without telling which exact branch it belongs to and in Gerrit, if you think in terms of commits instead of change numbers.
    (The Gerrit case has already been solved through allowTipSHA1InWant as every Gerrit change has a ref.)

Git 2.6 (Q3 2015) will improve that model.
See commit 2bc31d1, commit cc118a6 (28 Jul 2015) by Jeff King (peff).
(Merged by Junio C Hamano -- gitster -- in commit 824a0be, 19 Aug 2015)

refs: support negative transfer.hideRefs

If you hide a hierarchy of refs using the transfer.hideRefs config, there is no way to later override that config to "unhide" it.
This patch implements a "negative" hide which causes matches to immediately be marked as unhidden, even if another match would hide it.
We take care to apply the matches in reverse-order from how they are fed to us by the config machinery, as that lets our usual "last one wins" config precedence work (and entries in .git/config, for example, will override /etc/gitconfig).

So you can now do:

git config --system transfer.hideRefs refs/secret
git config transfer.hideRefs '!refs/secret/not-so-secret'

to hide refs/secret in all repos, except for one public bit in one specific repo.

Git 2.7 (Nov/Dec 2015) will improve again:

See commit 948bfa2, commit 00b293e (05 Nov 2015), commit 78a766a, commit 92cab49, commit 92cab49, commit 92cab49 (03 Nov 2015), commit 00b293e, commit 00b293e (05 Nov 2015), and commit 92cab49, commit 92cab49, commit 92cab49, commit 92cab49 (03 Nov 2015) by Lukas Fleischer (lfos).
Helped-by: Eric Sunshine (sunshineco).
(Merged by Jeff King -- peff -- in commit dbba85e, 20 Nov 2015)

config.txt: document the semantics of hideRefs with namespaces

Right now, there is no clear definition of how transfer.hideRefs should behave when a namespace is set.
Explain that hideRefs prefixes match stripped names in that case. This is how hideRefs patterns are currently handled in receive-pack.

hideRefs: add support for matching full refs

In addition to matching stripped refs, one can now add hideRefs patterns that the full (unstripped) ref is matched against.
To distinguish between stripped and full matches, those new patterns must be prefixed with a circumflex (^).

Hence the new documentation:

transfer.hideRefs:

If a namespace is in use, the namespace prefix is stripped from each reference before it is matched against transfer.hiderefs patterns.
For example, if refs/heads/master is specified in transfer.hideRefs and the current namespace is foo, then refs/namespaces/foo/refs/heads/master is omitted from the advertisements but refs/heads/master and refs/namespaces/bar/refs/heads/master are still advertised as so-called "have" lines.
In order to match refs before stripping, add a ^ in front of the ref name. If you combine ! and ^, ! must be specified first.

R.. mentions in the comments the config uploadpack.allowAnySHA1InWant, which allows upload-pack to accept a fetch request that asks for any object at all. (Defaults to false).

See commit f8edeaa (Nov. 2016, Git v2.11.1) by David "novalis" Turner (novalis):

upload-pack: optionally allow fetching any sha1

It seems a little silly to do a reachabilty check in the case where we trust the user to access absolutely everything in the repository.

Also, it's racy in a distributed system -- perhaps one server advertises a ref, but another has since had a force-push to that ref, and perhaps the two HTTP requests end up directed to these different servers.

VonC
  • 1,042,979
  • 435
  • 3,649
  • 4,283
  • 5
    Can you give a more complete example on how to create a repo clone with just this single commit? I tried but failed.. Thanks! – Lars Bilke Aug 06 '15 at 12:55
  • @LarsBilke do you have the right version for the Git repo *server*? If the server has not the `uploadpack.allowReachableSHA1InWant` config activated, any single-commit clone would fail. – VonC Aug 06 '15 at 12:56
  • 1
    I want to push to GitHub. Maybe they don't allow this. – Lars Bilke Aug 06 '15 at 12:57
  • 2
    @LarsBilke we are talking about clone or pull here, not push. And I am pretty sure GitHub doesn't have Git 2.5 on the server side yet. – VonC Aug 06 '15 at 12:58
  • 2
    Now even better, there's `uploadpack.allowAnySHA1InWant` without the reachability-computation penalty (and DoS vector). – R.. GitHub STOP HELPING ICE Jun 26 '18 at 16:46
  • @R.. Thank you. I have included your comment in the answer for more visibility. – VonC Jun 26 '18 at 21:38
  • 1
    Thanks! I find it funny that they describe it as "trust the user to access" rather than "trust the repo authors not to push random crap they don't intend to make public". – R.. GitHub STOP HELPING ICE Jun 27 '18 at 00:27
  • Thank you. This is exactly the answer that I'm seeking. – ramwin Aug 16 '18 at 01:52
  • Note that SHA must be the full unabbreviated SHA, otherwise Git will claim it couldn't find the commit. – matt May 21 '21 at 20:15
  • 1
    @matt Good point, thank you. I have edited the answer accordingly, and included your comment in it, for more visibility. – VonC May 21 '21 at 21:25
102

You only clone once, so if you already have a clone of the remote repository, pulling from it won't download everything again. Just indicate what branch you want to pull, or fetch the changes and checkout the commit you want.

Fetching from a new repository is very cheap in bandwidth, as it will only download the changes you don't have. Think in terms of Git making the right thing, with minimum load.

Git stores everything in .git folder. A commit can't be fetched and stored in isolation, it needs all its ancestors. They are interrelated.

To reduce download size you can however ask git to fetch only objects related to a specific branch or commit:

git fetch origin refs/heads/branch:refs/remotes/origin/branch

This will download only commits contained in remote branch branch (and only the ones that you miss), and store it in origin/branch. You can then merge or checkout.

You can also specify only a SHA1 commit:

git fetch origin 96de5297df870:refs/remotes/origin/foo-commit

This will download only the commit of the specified SHA-1 96de5297df870 (and its ancestors that you miss), and store it as (non-existing) remote branch origin/foo-commit.

Flow
  • 22,048
  • 13
  • 91
  • 147
CharlesB
  • 75,315
  • 26
  • 174
  • 199
  • I wish to avoid this because I am working with Linux kernel sources, and I like to visit different repos and merge the commits I like from them which improve certain components I like and so I cannot clone every remote repo :( they are humongous in size – Varun Chitre Feb 14 '13 at 10:29
  • 3
    Looks like you're making a confusion on what clone means. When you fetch changes from a remote repo you don't clone it, you just get the commits in your history. Then you choose which commit you want to check out, or merge it in your history – CharlesB Feb 14 '13 at 10:44
  • 2
    It still downloads lots of data(430mb) with git fetch. The required commit is just of few kbs. There is no special command to do this really? And what if I want to remove the 'git fetched' repo? where is it stored? – Varun Chitre Feb 14 '13 at 11:23
  • Actually you can do partial fetch, but still it will download the intermediate commits between the ones you have. Commits are interrelated together, to build the history you need their succession. – CharlesB Feb 14 '13 at 13:02
  • 9
    This is rather out-of-date now. We have both the ability to perform a [shallow clone](http://stackoverflow.com/questions/1778088/how-to-clone-a-single-branch-in-git), as well as to [fetch a single commit](http://stackoverflow.com/a/30701724/1055489). Shallow clones are now allowed to push and fetch normally, without having to know the full history of the project, so it's no longer correct to say that a commit cannot exist alone without its ancestors. What you say about fetching after the initial clone is very true, but we also have even cheaper options. – Theodore Murdock Oct 12 '15 at 22:03
  • If you store it in /foo-commit(a non-existing branch) will it create this branch? And can I easly merge with this branch and if something goes wrong undo the merge? – Taacoo Jul 27 '16 at 13:53
  • 6
    The last command (using SHA1 commit) doesn't work for me. The command silently does "something" for a while, and then exits with no message or apparent side-effect. – HRJ Aug 26 '16 at 05:32
  • 1
    @HRJ Yes, I encountered this too, on Ubuntu 16.04 with Git `2.7.4-0ubuntu1.3`. However, when using `2.16.2-0ppa1~ubuntu16.04.1` from the git-core PPA, this works as it should. Sounds like a bug that got fixed. Couldn't find a reference to that with a quick search. If someone can get me a pointer on that, I would love to get this fix backported. – gertvdijk Feb 27 '18 at 12:46
  • 1
    The `git fetch origin` only works with **unabbreviated** hash. Tested on Ubuntu 18.04 with gerrit remote. – kubanczyk May 24 '19 at 13:30
65

I did a pull on my git repo:

git pull --rebase <repo> <branch>

Allowing git to pull in all the code for the branch and then I went to do a reset over to the commit that interested me.

git reset --hard <commit-hash>

Hope this helps.

Piu Sharma
  • 711
  • 5
  • 2
  • 1
    None of the answers worked, this one though, saved my life! Thanks a bunch! – michaeltintiuc Dec 08 '14 at 23:19
  • 1
    The reset --hard worked for me after cloning! Thanks. – Nick-ACNB Mar 20 '15 at 20:33
  • 3
    -1: "destructive" commands like `git reset --hard`, when shared in generalised solutions, can lead people into traps where they lose data (or, in this case: in a state where getting their data back is nontrivial). – Ry Biesemeyer Feb 07 '19 at 18:47
59

You can simply fetch a single commit of a remote repo with

git fetch <repo> <commit>

where,

  • <repo> can be a remote repo name (e.g. origin) or even a remote repo URL (e.g. https://git.foo.com/myrepo.git)
  • <commit> can be the SHA1 commit

for example

git fetch https://git.foo.com/myrepo.git 0a071603d87e0b89738599c160583a19a6d95545

after you fetched the commit (and the missing ancestors) you can simply checkout it with

git checkout FETCH_HEAD

Note that this will bring you in the "detached head" state.

Flow
  • 22,048
  • 13
  • 91
  • 147
19

You can simply fetch the remote repo with:

git fetch <repo>

where,

  • <repo> can be a remote repo name (e.g. origin) or even a remote repo URL (e.g. https://git.foo.com/myrepo.git)

for example:

git fetch https://git.foo.com/myrepo.git

after you fetched the repos you may merge the commits that you want (since the question is about retrieve one commit, instead merge you may use cherry-pick to pick just one commit):

git merge <commit>
  • <commit> can be the SHA1 commit

for example:

git cherry-pick 0a071603d87e0b89738599c160583a19a6d95545

or

git merge 0a071603d87e0b89738599c160583a19a6d95545

if is the latest commit that you want to merge, you also may use FETCH_HEAD variable : 

git cherry-pick (or merge) FETCH_HEAD
Sérgio
  • 6,132
  • 41
  • 49
4

This works best:

git fetch origin specific_commit
git checkout -b temp FETCH_HEAD

name "temp" whatever you want...this branch might be orphaned though

Alexander Mills
  • 1
  • 80
  • 344
  • 642
1

Finally i found a way to clone specific commit using git cherry-pick. Assuming you don't have any repository in local and you are pulling specific commit from remote,

1) create empty repository in local and git init

2) git remote add origin "url-of-repository"

3) git fetch origin [this will not move your files to your local workspace unless you merge]

4) git cherry-pick "Enter-long-commit-hash-that-you-need"

Done.This way, you will only have the files from that specific commit in your local.

Enter-long-commit-hash:

You can get this using -> git log --pretty=oneline

surya deepak
  • 29
  • 3
0

I think 'git ls-remote' ( http://git-scm.com/docs/git-ls-remote ) should do what you want. Without force fetch or pull.

Hubbitus
  • 4,481
  • 2
  • 37
  • 42
0

If the requested commit is in the pull requests of the remote repo, you can get it by its ID:

# Add the remote repo path, let's call it 'upstream':
git remote add upstream https://github.com/repo/project.git

# checkout the pull ID, for example ID '60':
git fetch upstream pull/60/head && git checkout FETCH_HEAD
Noam Manos
  • 10,586
  • 2
  • 68
  • 65
0

In a project we had a problem so that we had to revert back to a certain commit. We made it with the following command successfully:

git reset --hard <commitID>
Eren
  • 17
  • 3