Re: bug: git pull may delete untracked files in submodule without notice
To
Duy Nguyen
Christian Spanier
Cc
Git Mailing List
Phillip Wood
From
Phillip Wood
See Also
Prev Ref 1
Date
2019-05-03 15:10:01 UTC
On 03/05/2019 11:28, Duy Nguyen wrote:
> On Fri, May 3, 2019 at 5:25 PM Christian Spanier <cspanier@boxie.eu> wrote:
>>
>> Hi,
>>
>> 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

Phillip

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.
>>
>> Greetings,
>> 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
>> directory)
>> #
>> #         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 ..
>>
>> # NOW THE BUG:
>>
>> # 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 :(
> 
> 
>