What does git fetch really do?

Question

Now I want to merge a remote branch, naming it origin/branch1, with my local branch branch1, as my partner pushed a new commit to branch1 on remote after our last merge, and I haven't committed since his last commit and want to get update from this commit. I used the following commands:

$ git fetch origin branch1
Compressing...
*branch branch1 ->FETCH_HEAD
$ git merge origin/branch1
Already up-to-date

That was not what I intended. What was in my mind before doing this was using fetch to get what my partner added and made the remote branch origin/branch1 being updated. However, the "Already up-to-date" means I failed to get the updates in my local branch1. Then I checked the sha1 value of origin/branch1 by

$ git ls-remote origin

and found that it kept the stale value of old commit after we merged last time. It tells that the git fetch origin branch1 cannot update origin/branch1. I did another experiment, where my partner created another branch named "branch2" in his side and pushed a commit in branch2 to the remote origin. Then I still used

$ git fetch origin branch2
Compressing...
$ git merge origin/branch2
No branch named "origin/branch2"

The "Compressing" told me that the first command downloaded something in branch2 in origin successfully, however, the second command told me that there was no branch named origin/branch2! Thus I concluded that neither could git fetch origin branchname update origin/branchname locally, nor could it create a remote branch if it doesn't exist.

After I replace git fetch origin branch# with git fetch origin, all the git merges worked as I expected.

However, I often see the combination of

$ git fetch remote branch-name
$ git merge remote/branch-name

So my question is what is the difference between git fetch remote and git fetch remote branch-name? And in what conditions could I succeed in letting it work as my wish by this combination?

Answer 1

If you need to know what exactly git fetch is doing or any other git command, just prefix it with GIT_TRACE=1, so it'll give you trace output with any other commands which are invoked, e.g.

$ GIT_TRACE=1 git fetch origin master
03:08:15.704945 git.c:348               trace: built-in: git 'fetch' 'origin' 'master'
03:08:15.706183 run-command.c:347       trace: run_command: 'ssh' '[email protected]' 'git-upload-pack '\''FOO/BAR.git'\'''
03:08:16.006394 run-command.c:347       trace: run_command: 'rev-list' '--objects' '--stdin' '--not' '--all' '--quiet'
03:08:16.013096 run-command.c:347       trace: run_command: 'rev-list' '--objects' '--stdin' '--not' '--all'
03:08:16.013625 exec_cmd.c:129          trace: exec: 'git' 'rev-list' '--objects' '--stdin' '--not' '--all'
03:08:16.016617 git.c:348               trace: built-in: git 'rev-list' '--objects' '--stdin' '--not' '--all'
From github.com:FOO/BAR
 * branch            master     -> FETCH_HEAD
03:08:16.153070 run-command.c:347       trace: run_command: 'gc' '--auto'
03:08:16.153748 exec_cmd.c:129          trace: exec: 'git' 'gc' '--auto'
03:08:16.157704 git.c:348               trace: built-in: git 'gc' '--auto'

which basically is doing ssh to the remote host, running git-upload-pack which send objects packed back to git-fetch-pack which receives missing objects from another repository.

In man git-upload-pack we can read:

Invoked by git fetch-pack, learns what objects the other side is missing, and sends them after packing.

And in man git-fetch-pack we can read:

Invokes git-upload-pack on a possibly remote repository and asks it to send objects missing from this repository, to update the named heads. The list of commits available locally is found out by scanning the local refs/ hierarchy and sent to git-upload-pack running on the other end.

---

To answer the question, the difference between git fetch remote and git fetch remote branch-name, is that when you don't specify <refspec> parameter (such as branch), it fetches all branches and/or tags (refs, see: git ls-refs) from one or more other repositories along with the objects necessary to complete their histories. By default only tags which are reachable by the objects that are fetched are downloaded (e.g. you end up only with tags that point at branches that you are interested in).

And when you run with explicit branches and/or tags, git first determines what needs to be fetched, then getting only the relevant refs (branches or tags). E.g. git fetch origin master will fetch only the master branch.