Web lists-archives.com

Re: bug: git pull may delete untracked files in submodule without notice

On 03/05/2019 11:28, Duy Nguyen wrote:
On Fri, May 3, 2019 at 5:25 PM Christian Spanier <cspanier@xxxxxxxx> wrote:


I found a bug where Git may delete untracked files without notice in
certain situations. This bug effects Git 2.21.0 both on Linux and Windows.
In summary this happens when git pull merges a commit that replaces a
submodule folder with a symlink. Any files within the folder are deleted
without notice.

Maybe it's not the same, maybe it is. But Phillip recently did some
work protecting untracked files and I think he touched a test case
about submodule. Adding him so he can check, if he has time.

That was a different case. I've just tried adding an untracked file to the submodule directory in the test that replaces a submodule with a file (patch below hopefully thunderbird does not break it) and it still passes when running t1013-read-tree-submodule.sh (it's test 8) - so I think there is something wrong with the detection of untracked files in a submodule directory when we replace it with a file. Normally I'd ping Stefan about submodules but he's not around at the moment.

Best Wishes


diff --git a/t/lib-submodule-update.sh b/t/lib-submodule-update.sh
index 5b56b23166..618176ae54 100755
--- a/t/lib-submodule-update.sh
+++ b/t/lib-submodule-update.sh
@@ -735,6 +735,7 @@ test_submodule_recursing_with_args_common() {
                reset_work_tree_to_interested add_sub1 &&
                        cd submodule_update &&
+                       echo untracked>sub1/untracked &&
git branch -t replace_sub1_with_file origin/replace_sub1_with_file &&
                        $command replace_sub1_with_file &&
test_superproject_content origin/replace_sub1_with_file &&

Check out the script below for details.

This happend on some developer's machine and deleted a repository
containing about 200GiB of files and tons of uncommited local scripts,
log files and whatever, just because some other dev accidentally
commited a temporary change.

Christian Spanier

##### PREPARATION #####

# New empty repository #1
mkdir rep1
cd rep1
git init --bare .
cd ..

# New empty repository #2
mkdir rep2
cd rep2
git init --bare .
cd ..

# Clone repository #1 and create initial commit
git clone rep1 clone_rep1_user1
cd clone_rep1_user1
touch README
git add README
git commit -m "initial commit"
git push
cd ..

# Clone repository #2 and create initial commit
git clone rep2 clone_rep2
cd clone_rep2
touch README
git add README
git commit -m "initial commit"
git push
cd ..

# Add repository #2 as a submodule to repository #1
cd clone_rep1_user1
git submodule add ../rep2
git commit -m "add submodule"
git push
cd ..

# User 2 also clones repository #1 and #2 recursively
git clone --recursive rep1 clone_rep1_user2

# User 2 starts working in his folder and adds an important local file
which is
# not yet committed inside the submodule folder.
cd clone_rep1_user2/rep2
echo "important work" > uncommitted_file
cd ../../

# Meanwhile, user 1 temporarily switch out folder /clone_rep1_user1/rep2
with a
# symbolic link to a different folder (for whatever reason, maybe a copy
of an
# older version or anything).
mkdir rep2_alternative
cd clone_rep1_user1
mv rep2 ../rep2_backup
ln -s ../rep2_alternative rep2
# On Windows this can be done with 'mklink /D rep2 ../rep2_alternative',
# which requires admin privileges. The bug is not reproducible when
# using a directory junction with 'mklink /D /J ...'.

# He does some work on rep1 but then accidently adds the symbolic link
to his
# next commit and pushes the changes. Notice the typechange of rep2.
echo "some" > work
git status
# On branch master
# Your branch is up to date with 'origin/master'.
# Changes not staged for commit:
#   (use "git add <file>..." to update what will be committed)
#   (use "git checkout -- <file>..." to discard changes in working
#         typechange: rep2
# Untracked files:
#   (use "git add <file>..." to include in what will be committed)
#         work
# no changes added to commit (use "git add" and/or "git commit -a")
git add .
git commit -m "do some work"
git push
cd ..


# User 2 pulls the changes and loses his important work in
# rep2/uncommitted_file because Git replaces the folder with a symlink
# without checking for modified or uncommited files!
# He should get an error in this case!
cd clone_rep1_user2
git pull
cat rep2/uncommitted_file
# cat: rep2/uncommitted_file: Not a directory
# "important work" in rep2/uncommitted_file is gone :(