submodule: submodule_move_head omits old argument in forced case
authorStefan Beller <sbeller@google.com>
Fri, 5 Jan 2018 20:03:04 +0000 (12:03 -0800)
committerJunio C Hamano <gitster@pobox.com>
Fri, 5 Jan 2018 20:35:35 +0000 (12:35 -0800)
When using hard reset or forced checkout with the option to recurse into
submodules, the submodules need to be reset, too.

It turns out that we need to omit the duplicate old argument to read-tree
in all forced cases to omit the 2 way merge and use the more assertive
behavior of reading the specific new tree into the index and updating
the working tree.

Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
submodule.c
t/lib-submodule-update.sh

index 2967704..47ddc9b 100644 (file)
@@ -1657,7 +1657,9 @@ int submodule_move_head(const char *path,
        else
                argv_array_push(&cp.args, "-m");
 
-       argv_array_push(&cp.args, old ? old : EMPTY_TREE_SHA1_HEX);
+       if (!(flags & SUBMODULE_MOVE_HEAD_FORCE))
+               argv_array_push(&cp.args, old ? old : EMPTY_TREE_SHA1_HEX);
+
        argv_array_push(&cp.args, new ? new : EMPTY_TREE_SHA1_HEX);
 
        if (run_command(&cp)) {
index fb0173e..1f38a85 100755 (executable)
@@ -1015,4 +1015,18 @@ test_submodule_forced_switch_recursing_with_args () {
                        test_submodule_content sub1 origin/modify_sub1
                )
        '
+
+       test_expect_success "$command: changed submodule worktree is reset" '
+               prolog &&
+               reset_work_tree_to_interested add_sub1 &&
+               (
+                       cd submodule_update &&
+                       rm sub1/file1 &&
+                       : >sub1/new_file &&
+                       git -C sub1 add new_file &&
+                       $command HEAD &&
+                       test_path_is_file sub1/file1 &&
+                       test_path_is_missing sub1/new_file
+               )
+       '
 }