diff mbox series

[v2] merge: avoid write merge state when unable to write index

Message ID pull.1731.v2.git.1715836731784.gitgitgadget@gmail.com (mailing list archive)
State Superseded
Headers show
Series [v2] merge: avoid write merge state when unable to write index | expand

Commit Message

Kyle Zhao May 16, 2024, 5:18 a.m. UTC
From: Kyle Zhao <kylezhao@tencent.com>

Currently, when index.lock exists, if a merge occurs, the merge state
will be written and the index will be unchanged.

If the user exec "git commit" instead of "git merge --abort" after this,
a merge commit will be generated and all modifications from the source
branch will be lost.

Signed-off-by: Kyle Zhao <kylezhao@tencent.com>
---
    merge: avoid write merge state when unable to write index
    
    In some of our monorepos, code is sometimes lost after merging.
    
    After investigation, we discovered the problem.
    
    This happens if we perform "git pull" or "git merge" when another git
    process is writing to the index, especially in a monorepo (because its
    index will be larger).
    
    How to reproduce:
    
    git init demo
    cd demo
    touch 1.txt && git add . && git commit -m "1"
    git checkout -b source-branch
    touch 2.txt && git add . && git commit -m "2"
    git checkout master
    echo "1" >> 1.txt && git add . && git commit -m "3"
    touch .git/index.lock
    git merge source-branch
    rm .git/index.lock
    git commit -m "4"
    
    
    Then the modifications from the source branch are lost.
    
    Regards, Kyle

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1731%2Fkeyu98%2Fkz%2Ffix-merge-when-index-lock-exists-v2
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1731/keyu98/kz/fix-merge-when-index-lock-exists-v2
Pull-Request: https://github.com/gitgitgadget/git/pull/1731

Range-diff vs v1:

 1:  aab0be55989 ! 1:  86101e5b778 merge: avoid write merge state when unable to write index
     @@ Commit message
          will be written and the index will be unchanged.
      
          If the user exec "git commit" instead of "git merge --abort" after this,
     -    the commit will be successful and all modifications from the source
     +    a merge commit will be generated and all modifications from the source
          branch will be lost.
      
          Signed-off-by: Kyle Zhao <kylezhao@tencent.com>


 builtin/merge.c  |  6 ++++--
 t/t7600-merge.sh | 10 ++++++++++
 2 files changed, 14 insertions(+), 2 deletions(-)


base-commit: 19fe900cfce8096b7645ec9611a0b981f6bbd154

Comments

Junio C Hamano May 16, 2024, 4:18 p.m. UTC | #1
"Kyle Zhao via GitGitGadget" <gitgitgadget@gmail.com> writes:

> From: Kyle Zhao <kylezhao@tencent.com>
>
> Currently, when index.lock exists, if a merge occurs, the merge state
> will be written and the index will be unchanged.

"Currently, " is superfluous in this project, as by convention our
proposed log message begins with an observation of the current
status (and "what's wrong in it" follows) in the present tense.

My guess, from reading the tests, of the situation this patch
attempts to handle is something like this?

    When running a merge while the index is locked (presumably by
    another process), the merge state is written in SUCH and SUCH
    files, the index is not updated, and then the merge fails.  This
    leaves the resulting state inconsistent.

> If the user exec "git commit" instead of "git merge --abort" after this,
> a merge commit will be generated and all modifications from the source
> branch will be lost.

I do not think this is accurate description of the "an user action
can make it worse".  In reality, if the user runs "git commit", a
safety kicks in and they get:

    fatal: Unable to create '.../.git/index.lock': file exists.

In fact, your "How to reproduce" below the three-dash line removes
the stale index.lock file before running "git commit".

    From this state, if the index.lock file gets removed and the
    user runs "git commit", a merge commit is created, recording the
    tree from the inconsistent state the merge left.

may be a better description of the breakage.

But stepping back a bit, I do not think this extra paragraph is
needed at all.  If there were a competing process holding the
index.lock, it were in the process of updating the index, possibly
even to create a new commit.  If that process were indeed running
the "git commit" command, MERGE_HEAD and other state files we write
on our side will be taken into account by them and cause them to
record a merge, even though they may have been trying to record
something entirely different.  So regardless of what _this_ user,
whose merge failed due to a competing process that held the
index.lock file, does after the merge failure, the recorded history
can be hosed by the other process.

In any case, to prevent the other "git commit" process from using
"our" MERGE_HEAD and other state files to record a wrong contents,
the right fix is to make sure everybody who takes the lock on the
index file does *not* create any other state files to be read by
others before it successfully takes the lock.  That will also
prevent "git commit" we run after a failed merge (and removing the
lockfile) from recording an incorrect merge.

I do not offhand know if returning 2 (aka "I am not equipped to
handle this merg e at all") is a good way to do so, but if it is,
then the patch given here is absolutely the right thing to do.

>...
>     How to reproduce:
>     ...
>     git merge source-branch
>     rm .git/index.lock
>     git commit -m "4"

> diff --git a/t/t7600-merge.sh b/t/t7600-merge.sh
> index e5ff073099a..f03709ea4be 100755
> --- a/t/t7600-merge.sh
> +++ b/t/t7600-merge.sh
> @@ -236,6 +236,16 @@ test_expect_success 'merge c1 with c2' '
>  	verify_parents $c1 $c2
>  '
>  
> +test_expect_success 'merge c1 with c2 when index.lock exists' '
> +	test_when_finished rm .git/index.lock &&
> +	git reset --hard c1 &&
> +	touch .git/index.lock &&

Do not use "touch" when the timestamp is not the thing we care
about.  In this case, you want that file to _exist_, and the right
way to do so is

	>.git/index.lock &&

which already appears in t7400 (it uses a no-op ":" command, but
that is not needed).

Other than that, the patch looks good to me.

Thanks.
diff mbox series

Patch

diff --git a/builtin/merge.c b/builtin/merge.c
index 6a6d3798858..80e3438be25 100644
--- a/builtin/merge.c
+++ b/builtin/merge.c
@@ -698,8 +698,10 @@  static int try_merge_strategy(const char *strategy, struct commit_list *common,
 
 	if (repo_refresh_and_write_index(the_repository, REFRESH_QUIET,
 					 SKIP_IF_UNCHANGED, 0, NULL, NULL,
-					 NULL) < 0)
-		return error(_("Unable to write index."));
+					 NULL) < 0) {
+		error(_("Unable to write index."));
+		return 2;
+	}
 
 	if (!strcmp(strategy, "recursive") || !strcmp(strategy, "subtree") ||
 	    !strcmp(strategy, "ort")) {
diff --git a/t/t7600-merge.sh b/t/t7600-merge.sh
index e5ff073099a..f03709ea4be 100755
--- a/t/t7600-merge.sh
+++ b/t/t7600-merge.sh
@@ -236,6 +236,16 @@  test_expect_success 'merge c1 with c2' '
 	verify_parents $c1 $c2
 '
 
+test_expect_success 'merge c1 with c2 when index.lock exists' '
+	test_when_finished rm .git/index.lock &&
+	git reset --hard c1 &&
+	touch .git/index.lock &&
+	test_must_fail git merge c2 &&
+	test_path_is_missing .git/MERGE_HEAD &&
+	test_path_is_missing .git/MERGE_MODE &&
+	test_path_is_missing .git/MERGE_MSG
+'
+
 test_expect_success 'merge --squash c3 with c7' '
 	git reset --hard c3 &&
 	test_must_fail git merge --squash c7 &&