threads / discuss / 29747

Stash during incomplete merge

Subject: Stash during incomplete merge

## tl;dr

4 messages between Feb 26, 2012 and Feb 29, 2012.

replies: 3people: 2as markdown or json

Phil Hord· Feb 26, 2012, 18:36 UTC · lore
Hi list,

I was cherry-picking changes from an old branch recently when I ran into unexpected behavior with git stash pop. When I git-stash-save after resolving a merge-conflict, the subsequent git-stash-pop does not restore my index.

I think it is the same problem being asked about here: http://stackoverflow.com/questions/9009354/git-stash-during-a-merge-conflict

Is this expected behavior or a bug?

<http://stackoverflow.com/questions/9009354/git-stash-during-a-merge-conflict>Here's a script the demonstrates the anomaly, but my actual encounter involved more files, some of which I added to the index and some I did not:

# Create a sample merge-conflict git init tmp-repo && cd tmp-repo echo foo > foo.txt && git add foo.txt && git commit -m "foo" git checkout -b A master && echo foo-A > foo.txt && git commit -am "foo-A" git checkout -b B master && echo foo-B > foo.txt && git commit -am "foo-B" git merge A git status # Resolve the conflict echo foo-AB > foo.txt && git add foo.txt git status git stash # test test test... Resume... git stash pop

Here's some of the final output:

$ git merge A Auto-merging foo.txt CONFLICT (content): Merge conflict in foo.txt Recorded preimage for 'foo.txt' Automatic merge failed; fix conflicts and then commit the result.

$ git status # On branch B # Unmerged paths: # (use "git add/rm <file>..." as appropriate to mark resolution) # # both modified: foo.txt # no changes added to commit (use "git add" and/or "git commit -a")

$ # Resolve the conflict $ echo foo-AB > foo.txt && git add foo.txt $ git status # On branch B # Changes to be committed: # # modified: foo.txt #

$ # Now foo.txt is in my index. But I have to test something before I commit. $ git stash Saved working directory and index state WIP on B: 80f2a13 foo-B HEAD is now at 80f2a13 foo-B

$ # test test test... Resume... $ git stash pop

# On branch B # Changes not staged for commit: # (use "git add <file>..." to update what will be committed) # (use "git checkout -- <file>..." to discard changes in working directory) # # modified: foo.txt # no changes added to commit (use "git add" and/or "git commit -a") Dropped refs/stash@{0} (460a6d5c67a3db613fd27f1854ecc7b89eeaa207)

Neal Kreitzinger· Feb 28, 2012, 20:22 UTC · re: Phil Hord · lore

Re: Stash during incomplete merge

On 2/26/2012 12:36 PM, Phil Hord wrote:
Show 77 quoted lines
> Hi list,
>
> I was cherry-picking changes from an old branch recently when I ran into
> unexpected behavior with git stash pop.  When I git-stash-save after
> resolving a  merge-conflict, the subsequent git-stash-pop does not
> restore my index.
>
> I think it is the same problem being asked about here:
> http://stackoverflow.com/questions/9009354/git-stash-during-a-merge-conflict
>
> Is this expected behavior or a bug?
>
> <http://stackoverflow.com/questions/9009354/git-stash-during-a-merge-conflict>Here's
> a script the demonstrates the anomaly, but my actual encounter involved
> more files, some of which I added to the index and some I did not:
>
> # Create a sample merge-conflict
> git init  tmp-repo&&  cd tmp-repo
> echo foo>  foo.txt&&  git add foo.txt&&  git commit -m "foo"
> git checkout -b A master&&  echo foo-A>  foo.txt&&  git commit -am "foo-A"
> git checkout -b B master&&  echo foo-B>  foo.txt&&  git commit -am "foo-B"
> git merge A
> git status
> # Resolve the conflict
> echo foo-AB>  foo.txt&&  git add foo.txt
> git status
> git stash
> # test test test...  Resume...
> git stash pop
>
>
> Here's some of the final output:
>
> $ git merge A
> Auto-merging foo.txt
> CONFLICT (content): Merge conflict in foo.txt
> Recorded preimage for 'foo.txt'
> Automatic merge failed; fix conflicts and then commit the result.
>
> $ git status
> # On branch B
> # Unmerged paths:
> #   (use "git add/rm<file>..." as appropriate to mark resolution)
> #
> #       both modified:      foo.txt
> #
> no changes added to commit (use "git add" and/or "git commit -a")
>
> $ # Resolve the conflict
> $ echo foo-AB>  foo.txt&&  git add foo.txt
> $ git status
> # On branch B
> # Changes to be committed:
> #
> #       modified:   foo.txt
> #
>
> $ # Now foo.txt is in my index.  But I have to test something before I
> commit.
> $ git stash
> Saved working directory and index state WIP on B: 80f2a13 foo-B
> HEAD is now at 80f2a13 foo-B
>
> $ # test test test...  Resume...
> $ git stash pop
>
> # On branch B
> # Changes not staged for commit:
> #   (use "git add<file>..." to update what will be committed)
> #   (use "git checkout --<file>..." to discard changes in working
> directory)
> #
> #       modified:   foo.txt
> #
> no changes added to commit (use "git add" and/or "git commit -a")
> Dropped refs/stash@{0} (460a6d5c67a3db613fd27f1854ecc7b89eeaa207)
>

Was the foo.txt in your worktree still the same as the one you resolved and added to the index?

If so, at that point you can just do "git add foo.txt" to put it back in your index.

(merge conflict) git add conflict-resolution (file is in worktree and index) git stash (file is in stash of worktree and stash of index) git stash pop (file is back in worktree but not back in index) git add conflict-resolution (file is back in the index)

If not, then you must have a more complicated case like a modified worktree version that differs from the index version (original conflict resolution) in which case you need to also consider whether or not it is an [a]"evil merge". Ideally, you intend to commit the index (conflict resolution) as the merge commit and add any worktree mods in addition to that as a commit after the merge commit. If so, then I see your problem. The next time, you can get both the worktree version and the index version back with:

git stash apply --index

git stash does not apply the index by default so you have to specify the --index option if you want the index back also. I recommend git stash apply instead of git stash pop in case you realize you forgot the index. Then you can rerun git stash apply --index. git stash pop can only be run once because it throws away the stash after popping it. You can run git stash drop after you confirm that you did the git stash apply correctly.

I assume you know that the stash is a stack and how to specify which stash you want by using the reflog syntax (see git-stash manpage: http://schacon.github.com/git/git-stash.html). Stashes are really [b]"commit objects" of the worktree and index trees (you can see them in gitk by viewing all refs), but they are treated according to "stash rules" instead of "commit rules".

If you still want to try and get that conflict-resolution from the "lost" index from that stash you popped you can try this variation of the "Recovering stashes that were cleared/dropped erroneously" procedure at the end of manpage:

(1) Find the lost stash: $ git fsck --unreachable | grep commit | cut -d" " -f3 | xargs git log --no-walk --grep=WIP --grep="index on"

(2) Review the list of commits and decide which pair is for the stash you want. Note on stash commit-object messages: WIP = stash commit-object of the worktree "index on" = stash commit-object of the index (this may not exist of the index didn't differ from the worktree)

(3) Reset the conflict-resolution-file(s) from the stash of the index:

$ git reset <sha1-of-stash-commit-object-of-index> -- <conflict-resolution-file>

git reset will just put it back in your index and leave the worktree alone.

You may also want to consider the --keep-index option on your "git stash save" if your "testing" workflow doesn't involve adds or commits before the git stash apply/pop.

Footnotes: a. If you resolved the conflict and then made further mods that is called an "evil merge" because you really didn't just merge the changes of the two parents, but also made additional changes that were not part of the changes made by either of the parents your are merging. The additional changes really should be their own commit after the merge. "Evil merges" are misleading (and also probably not documented in the merge commit message with a git commit --amend). b. There are only 4 types of objects in git: tag, commit, tree, blob.

Hope this helps. Maybe someone else has a better way to do this. Maybe my assumptions are incorrect and I'm missing something about what you are trying to do.

v/r, neal

Neal Kreitzinger· Feb 28, 2012, 22:24 UTC · re: Neal Kreitzinger · lore

Re: Stash during incomplete merge

On 2/28/2012 2:22 PM, Neal Kreitzinger wrote:
Show 5 quoted lines
>
> You may also want to consider the --keep-index option on your "git
> stash save" if your "testing" workflow doesn't involve adds or
> commits before the git stash apply/pop.
>

the very limited case I had in mind (and probably incorrect assumption) about your "testing" workflow was:

hack merge-conflicts $ git add conflict-resolution (conflict-resolution is in worktree and index) hack conflict-resolution with extra stuff (original conflict-resolution is still in index) uh-oh, i got carried away and started doing extra stuff (evil merge) and forgot to finish testing just the conflict-resolutions (pure merge-commit) $ git stash --keep-index (conflict-resolution is still in index and now back in wokrtree) finish testing just the conflict-resolutions (merge-commit-to-be) (conflict resolutions worked (I knew they would)) $ git stash pop (original conflict resolution is still in index, but extra-stuff is back in worktree) $ git commit (commit the conflict-resolutions/merge-commit) $ git add foo $ git commit (new foo stuff committed after merge commit)

v/r, neal

Neal Kreitzinger· Feb 29, 2012, 19:34 UTC · re: Neal Kreitzinger · lore

Re: Stash during incomplete merge

On 2/28/2012 4:24 PM, Neal Kreitzinger wrote:
Show 23 quoted lines
> On 2/28/2012 2:22 PM, Neal Kreitzinger wrote:
>>
>> You may also want to consider the --keep-index option on your "git
>>  stash save" if your "testing" workflow doesn't involve adds or
>> commits before the git stash apply/pop.
>>
> the very limited case I had in mind (and probably incorrect
> assumption) about your "testing" workflow was:
>
> hack merge-conflicts $ git add conflict-resolution
> (conflict-resolution is in worktree and index) hack
> conflict-resolution with extra stuff (original conflict-resolution
> is still in index) uh-oh, i got carried away and started doing extra
> stuff (evil merge) and forgot to finish testing just the
> conflict-resolutions (pure merge-commit) $ git stash --keep-index
> (conflict-resolution is still in index and now back in wokrtree)
> finish testing just the conflict-resolutions (merge-commit-to-be)
> (conflict resolutions worked (I knew they would)) $ git stash pop
> (original conflict resolution is still in index, but extra-stuff is
> back in worktree) $ git commit (commit the
> conflict-resolutions/merge-commit) $ git add foo $ git commit (new
> foo stuff committed after merge commit)
>

This imaginary scenario I proposed for --keep-index can get a merge-conflict on the pop so its invalid. Please disregard it. Sorry for the bum scoop.

v/r, neal

← back to recent threads