threads / patch / 52017

patchstash: avoid recursive hard reset on submodules

Subject: [PATCH] stash: avoid recursive hard reset on submodules

## tl;dr

8 messages between Oct 10, 2019 and Oct 15, 2019. Diffs are folded; open one to read it.

replies: 7people: 2as markdown or json

Jakob Jarmar· Oct 10, 2019, 23:34 UTC · lore

git stash push does not recursively stash submodules, but if submodule.recurse is set, it may recursively reset --hard them. Having only the destructive action recurse is likely to be surprising behaviour, and unlikely to be desirable, so the easiest fix should be to ensure that the call to git reset --hard never recurses into submodules.

This matches the behavior of check_changes_tracked_files, which ignores submodules.

Signed-off-by: Jakob Jarmar <jakob@jarmar.se>
---
 builtin/stash.c     | 2 +-
 git-legacy-stash.sh | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
Show changes to 2 files +2 −2

builtin/stash.c, git-legacy-stash.sh

diff --git a/builtin/stash.c b/builtin/stash.c
index b5a301f24d..c986c258f2 100644
--- a/builtin/stash.c
+++ b/builtin/stash.c
@@ -1383,7 +1383,7 @@ static int do_push_stash(const struct pathspec
*ps, const char *stash_msg, int q
             struct child_process cp = CHILD_PROCESS_INIT;
             cp.git_cmd = 1;
             argv_array_pushl(&cp.args, "reset", "--hard", "-q",
-                     NULL);
+                     "--no-recurse-submodules", NULL);
             if (run_command(&cp)) {
                 ret = -1;
                 goto done;
diff --git a/git-legacy-stash.sh b/git-legacy-stash.sh
index f60e9b3e87..07ad4a5459 100755
--- a/git-legacy-stash.sh
+++ b/git-legacy-stash.sh
@@ -370,7 +370,7 @@ push_stash () {
             git diff-index -p --cached --binary HEAD -- "$@" |
             git apply --index -R
         else
-            git reset --hard -q
+            git reset --hard -q --no-recurse-submodules
         fi

         if test "$keep_index" = "t" && test -n "$i_tree"
-- 
2.23.0
Jakob Jarmar· Oct 11, 2019, 00:11 UTC · re: Jakob Jarmar · lore

[PATCH v2] stash: avoid recursive hard reset on submodules

git stash push does not recursively stash submodules, but if submodule.recurse is set, it may recursively reset --hard them. Having only the destructive action recurse is likely to be surprising behaviour, and unlikely to be desirable, so the easiest fix should be to ensure that the call to git reset --hard never recurses into submodules.

This matches the behavior of check_changes_tracked_files, which ignores submodules.

Signed-off-by: Jakob Jarmar <jakob@jarmar.se>
---

Sorry for sending a patch with messed up whitespace. This one should be correct.

 builtin/stash.c     | 2 +-
 git-legacy-stash.sh | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
Show changes to 2 files +2 −2

builtin/stash.c, git-legacy-stash.sh

diff --git a/builtin/stash.c b/builtin/stash.c
index b5a301f24d..c986c258f2 100644
--- a/builtin/stash.c
+++ b/builtin/stash.c
@@ -1383,7 +1383,7 @@ static int do_push_stash(const struct pathspec *ps, const char *stash_msg, int q
 			struct child_process cp = CHILD_PROCESS_INIT;
 			cp.git_cmd = 1;
 			argv_array_pushl(&cp.args, "reset", "--hard", "-q",
-					 NULL);
+					 "--no-recurse-submodules", NULL);
 			if (run_command(&cp)) {
 				ret = -1;
 				goto done;
diff --git a/git-legacy-stash.sh b/git-legacy-stash.sh
index f60e9b3e87..07ad4a5459 100755
--- a/git-legacy-stash.sh
+++ b/git-legacy-stash.sh
@@ -370,7 +370,7 @@ push_stash () {
 			git diff-index -p --cached --binary HEAD -- "$@" |
 			git apply --index -R
 		else
-			git reset --hard -q
+			git reset --hard -q --no-recurse-submodules
 		fi
 
 		if test "$keep_index" = "t" && test -n "$i_tree"
-- 
2.23.0
Junio C Hamano· Oct 11, 2019, 06:02 UTC · re: Jakob Jarmar · lore

Re: [PATCH v2] stash: avoid recursive hard reset on submodules

Jakob Jarmar <jakob.j@gmail.com> writes:
Show 8 quoted lines
> git stash push does not recursively stash submodules, but if
> submodule.recurse is set, it may recursively reset --hard them. Having
> only the destructive action recurse is likely to be surprising
> behaviour, and unlikely to be desirable, so the easiest fix should be to
> ensure that the call to git reset --hard never recurses into submodules.
>
> This matches the behavior of check_changes_tracked_files, which ignores
> submodules.

Makes sense. Can we demonstrate an existing breakage and the fix in a new test or two in perhaps t3906-stash-submodule.sh?

Thanks.
Show 37 quoted lines
>
> Signed-off-by: Jakob Jarmar <jakob@jarmar.se>
> ---
>
> Sorry for sending a patch with messed up whitespace. This one should be
> correct.
>
>  builtin/stash.c     | 2 +-
>  git-legacy-stash.sh | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/builtin/stash.c b/builtin/stash.c
> index b5a301f24d..c986c258f2 100644
> --- a/builtin/stash.c
> +++ b/builtin/stash.c
> @@ -1383,7 +1383,7 @@ static int do_push_stash(const struct pathspec *ps, const char *stash_msg, int q
>  			struct child_process cp = CHILD_PROCESS_INIT;
>  			cp.git_cmd = 1;
>  			argv_array_pushl(&cp.args, "reset", "--hard", "-q",
> -					 NULL);
> +					 "--no-recurse-submodules", NULL);
>  			if (run_command(&cp)) {
>  				ret = -1;
>  				goto done;
> diff --git a/git-legacy-stash.sh b/git-legacy-stash.sh
> index f60e9b3e87..07ad4a5459 100755
> --- a/git-legacy-stash.sh
> +++ b/git-legacy-stash.sh
> @@ -370,7 +370,7 @@ push_stash () {
>  			git diff-index -p --cached --binary HEAD -- "$@" |
>  			git apply --index -R
>  		else
> -			git reset --hard -q
> +			git reset --hard -q --no-recurse-submodules
>  		fi
>  
>  		if test "$keep_index" = "t" && test -n "$i_tree"
Jakob Jarmar· Oct 11, 2019, 22:24 UTC · re: Junio C Hamano · lore

[PATCH v3] stash: avoid recursive hard reset on submodules

git stash push does not recursively stash submodules, but if submodule.recurse is set, it may recursively reset --hard them. Having only the destructive action recurse is likely to be surprising behaviour, and unlikely to be desirable, so the easiest fix should be to ensure that the call to git reset --hard never recurses into submodules.

This matches the behavior of check_changes_tracked_files, which ignores submodules.

Signed-off-by: Jakob Jarmar <jakob@jarmar.se>
---
Notes:
    I went for two separate tests with the same basic setup, one that does
    push + pop (which should be valid regardless of future changes), and one
    that just does push before checking that the submodule still is dirty
    (which would become invalid if git stash were to learn --recurse-submodules
    in the future, but matches this approach).
 builtin/stash.c            |  2 +-
 git-legacy-stash.sh        |  2 +-
 t/t3906-stash-submodule.sh | 42 +++++++++++++++++++++++++++++++++++++-
 3 files changed, 43 insertions(+), 3 deletions(-)
Show changes to 3 files +43 −3

builtin/stash.c, git-legacy-stash.sh, t/t3906-stash-submodule.sh

diff --git a/builtin/stash.c b/builtin/stash.c
index b5a301f24d..c986c258f2 100644
--- a/builtin/stash.c
+++ b/builtin/stash.c
@@ -1383,7 +1383,7 @@ static int do_push_stash(const struct pathspec *ps, const char *stash_msg, int q
 			struct child_process cp = CHILD_PROCESS_INIT;
 			cp.git_cmd = 1;
 			argv_array_pushl(&cp.args, "reset", "--hard", "-q",
-					 NULL);
+					 "--no-recurse-submodules", NULL);
 			if (run_command(&cp)) {
 				ret = -1;
 				goto done;
diff --git a/git-legacy-stash.sh b/git-legacy-stash.sh
index f60e9b3e87..07ad4a5459 100755
--- a/git-legacy-stash.sh
+++ b/git-legacy-stash.sh
@@ -370,7 +370,7 @@ push_stash () {
 			git diff-index -p --cached --binary HEAD -- "$@" |
 			git apply --index -R
 		else
-			git reset --hard -q
+			git reset --hard -q --no-recurse-submodules
 		fi
 
 		if test "$keep_index" = "t" && test -n "$i_tree"
diff --git a/t/t3906-stash-submodule.sh b/t/t3906-stash-submodule.sh
index d7219d6f8f..83106fa958 100755
--- a/t/t3906-stash-submodule.sh
+++ b/t/t3906-stash-submodule.sh
@@ -1,6 +1,6 @@
 #!/bin/sh
 
-test_description='stash apply can handle submodules'
+test_description='stash can handle submodules'
 
 . ./test-lib.sh
 . "$TEST_DIRECTORY"/lib-submodule-update.sh
@@ -21,4 +21,44 @@ KNOWN_FAILURE_CHERRY_PICK_SEES_EMPTY_COMMIT=1
 KNOWN_FAILURE_NOFF_MERGE_DOESNT_CREATE_EMPTY_SUBMODULE_DIR=1
 test_submodule_switch "git_stash"
 
+setup_basic() {
+	git init sub &&
+	(
+		cd sub &&
+		test_commit sub_file
+	) &&
+	git init main &&
+	(
+		cd main &&
+		git submodule add ../sub &&
+		test_commit main_file
+	) &&
+	test_when_finished "rm -rf main sub"
+}
+
+test_expect_success 'stash push with submodule.recurse=true preserves dirty submodule worktree' '
+	setup_basic &&
+	(
+		cd main &&
+		git config submodule.recurse true &&
+		echo "x" >main_file.t &&
+		echo "y" >sub/sub_file.t &&
+		git stash push &&
+		test_must_fail git -C sub diff --quiet
+	)
+'
+
+test_expect_success 'stash push and pop with submodule.recurse=true preserves dirty submodule worktree' '
+	setup_basic &&
+	(
+		cd main &&
+		git config submodule.recurse true &&
+		echo "x" >main_file.t &&
+		echo "y" >sub/sub_file.t &&
+		git stash push &&
+		git stash pop &&
+		test_must_fail git -C sub diff --quiet
+	)
+'
+
 test_done
-- 
2.23.0
Junio C Hamano· Oct 12, 2019, 01:46 UTC · re: Jakob Jarmar · lore

Re: [PATCH v3] stash: avoid recursive hard reset on submodules

Jakob Jarmar <jakob.j@gmail.com> writes:
Show 9 quoted lines
> diff --git a/t/t3906-stash-submodule.sh b/t/t3906-stash-submodule.sh
> index d7219d6f8f..83106fa958 100755
> --- a/t/t3906-stash-submodule.sh
> +++ b/t/t3906-stash-submodule.sh
> @@ -1,6 +1,6 @@
>  #!/bin/sh
>  
> -test_description='stash apply can handle submodules'
> +test_description='stash can handle submodules'
Good attention to the detail ;-)
> +setup_basic() {

Style. SP on both sides of () in our shell scripts (as seen in the existing shell function in the same file).

Show 12 quoted lines
> +	git init sub &&
> +	(
> +		cd sub &&
> +		test_commit sub_file
> +	) &&
> +	git init main &&
> +	(
> +		cd main &&
> +		git submodule add ../sub &&
> +		test_commit main_file
> +	) &&
> +	test_when_finished "rm -rf main sub"

Have test_when_finished that removes main and sub _before_ you start creating sub and main.

When the &&-cascade breaks anywhere, the control may not even reach your test_when_finished that registers the clean-up procedure. Imagine "git init sub" succeeds but "git init main" somehow fails---you still want to clean up "sub".

Other than that, looks reasonably well done.  
Thanks for working on this.
Show 28 quoted lines
> +}
> +
> +test_expect_success 'stash push with submodule.recurse=true preserves dirty submodule worktree' '
> +	setup_basic &&
> +	(
> +		cd main &&
> +		git config submodule.recurse true &&
> +		echo "x" >main_file.t &&
> +		echo "y" >sub/sub_file.t &&
> +		git stash push &&
> +		test_must_fail git -C sub diff --quiet
> +	)
> +'
> +
> +test_expect_success 'stash push and pop with submodule.recurse=true preserves dirty submodule worktree' '
> +	setup_basic &&
> +	(
> +		cd main &&
> +		git config submodule.recurse true &&
> +		echo "x" >main_file.t &&
> +		echo "y" >sub/sub_file.t &&
> +		git stash push &&
> +		git stash pop &&
> +		test_must_fail git -C sub diff --quiet
> +	)
> +'
> +
>  test_done
Jakob Jarmar· Oct 12, 2019, 15:24 UTC · re: Junio C Hamano · lore

Re: [PATCH v3] stash: avoid recursive hard reset on submodules

On Sat, Oct 12, 2019 at 10:46:36AM +0900, Junio C Hamano wrote:
Show 20 quoted lines
> > +	git init sub &&
> > +	(
> > +		cd sub &&
> > +		test_commit sub_file
> > +	) &&
> > +	git init main &&
> > +	(
> > +		cd main &&
> > +		git submodule add ../sub &&
> > +		test_commit main_file
> > +	) &&
> > +	test_when_finished "rm -rf main sub"
> 
> Have test_when_finished that removes main and sub _before_ you start
> creating sub and main.  
> 
> When the &&-cascade breaks anywhere, the control may not even reach
> your test_when_finished that registers the clean-up procedure.
> Imagine "git init sub" succeeds but "git init main" somehow
> fails---you still want to clean up "sub".
Ah, of course, will fix this (and the other point as well).
> Other than that, looks reasonably well done.  
Thanks. This was fun to figure out and fix.
Jakob Jarmar· Oct 12, 2019, 15:38 UTC · re: Junio C Hamano · lore

[PATCH v4] stash: avoid recursive hard reset on submodules

git stash push does not recursively stash submodules, but if submodule.recurse is set, it may recursively reset --hard them. Having only the destructive action recurse is likely to be surprising behaviour, and unlikely to be desirable, so the easiest fix should be to ensure that the call to git reset --hard never recurses into submodules.

This matches the behavior of check_changes_tracked_files, which ignores submodules.

Signed-off-by: Jakob Jarmar <jakob@jarmar.se>
---
Notes:
    1. Added space between function name and parentheses
    2. Moved test_when_finished cleanup to top of setup_basic
 builtin/stash.c            |  2 +-
 git-legacy-stash.sh        |  2 +-
 t/t3906-stash-submodule.sh | 42 +++++++++++++++++++++++++++++++++++++-
 3 files changed, 43 insertions(+), 3 deletions(-)
Show changes to 3 files +43 −3

builtin/stash.c, git-legacy-stash.sh, t/t3906-stash-submodule.sh

diff --git a/builtin/stash.c b/builtin/stash.c
index b5a301f24d..c986c258f2 100644
--- a/builtin/stash.c
+++ b/builtin/stash.c
@@ -1383,7 +1383,7 @@ static int do_push_stash(const struct pathspec *ps, const char *stash_msg, int q
 			struct child_process cp = CHILD_PROCESS_INIT;
 			cp.git_cmd = 1;
 			argv_array_pushl(&cp.args, "reset", "--hard", "-q",
-					 NULL);
+					 "--no-recurse-submodules", NULL);
 			if (run_command(&cp)) {
 				ret = -1;
 				goto done;
diff --git a/git-legacy-stash.sh b/git-legacy-stash.sh
index f60e9b3e87..07ad4a5459 100755
--- a/git-legacy-stash.sh
+++ b/git-legacy-stash.sh
@@ -370,7 +370,7 @@ push_stash () {
 			git diff-index -p --cached --binary HEAD -- "$@" |
 			git apply --index -R
 		else
-			git reset --hard -q
+			git reset --hard -q --no-recurse-submodules
 		fi
 
 		if test "$keep_index" = "t" && test -n "$i_tree"
diff --git a/t/t3906-stash-submodule.sh b/t/t3906-stash-submodule.sh
index d7219d6f8f..b93d1d74da 100755
--- a/t/t3906-stash-submodule.sh
+++ b/t/t3906-stash-submodule.sh
@@ -1,6 +1,6 @@
 #!/bin/sh
 
-test_description='stash apply can handle submodules'
+test_description='stash can handle submodules'
 
 . ./test-lib.sh
 . "$TEST_DIRECTORY"/lib-submodule-update.sh
@@ -21,4 +21,44 @@ KNOWN_FAILURE_CHERRY_PICK_SEES_EMPTY_COMMIT=1
 KNOWN_FAILURE_NOFF_MERGE_DOESNT_CREATE_EMPTY_SUBMODULE_DIR=1
 test_submodule_switch "git_stash"
 
+setup_basic () {
+	test_when_finished "rm -rf main sub" &&
+	git init sub &&
+	(
+		cd sub &&
+		test_commit sub_file
+	) &&
+	git init main &&
+	(
+		cd main &&
+		git submodule add ../sub &&
+		test_commit main_file
+	)
+}
+
+test_expect_success 'stash push with submodule.recurse=true preserves dirty submodule worktree' '
+	setup_basic &&
+	(
+		cd main &&
+		git config submodule.recurse true &&
+		echo "x" >main_file.t &&
+		echo "y" >sub/sub_file.t &&
+		git stash push &&
+		test_must_fail git -C sub diff --quiet
+	)
+'
+
+test_expect_success 'stash push and pop with submodule.recurse=true preserves dirty submodule worktree' '
+	setup_basic &&
+	(
+		cd main &&
+		git config submodule.recurse true &&
+		echo "x" >main_file.t &&
+		echo "y" >sub/sub_file.t &&
+		git stash push &&
+		git stash pop &&
+		test_must_fail git -C sub diff --quiet
+	)
+'
+
 test_done
-- 
2.23.0
Junio C Hamano· Oct 15, 2019, 01:35 UTC · re: Jakob Jarmar · lore

Re: [PATCH v4] stash: avoid recursive hard reset on submodules

Jakob Jarmar <jakob.j@gmail.com> writes:
Show 15 quoted lines
> git stash push does not recursively stash submodules, but if
> submodule.recurse is set, it may recursively reset --hard them. Having
> only the destructive action recurse is likely to be surprising
> behaviour, and unlikely to be desirable, so the easiest fix should be to
> ensure that the call to git reset --hard never recurses into submodules.
>
> This matches the behavior of check_changes_tracked_files, which ignores
> submodules.
>
> Signed-off-by: Jakob Jarmar <jakob@jarmar.se>
> ---
>
> Notes:
>     1. Added space between function name and parentheses
>     2. Moved test_when_finished cleanup to top of setup_basic
Looks good; will queue.
Thanks.

← back to recent threads