[PATCH] submodule.c: duplicate real_path's return value

["Nguyễn Thái Ngọc Duy" <pclouds@gmail.com>]

real_path returns the pointer to an internal buffer, which may be
overwritten by the next real_path call. Duplicate the return value for
safety.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
---
 Apply on top of jl/submodule-mv.
 
 Junio, I think this is the reason some tests failed when you merged
 nd/magic-pathspecs in (I suspect you needed to do parse_pathspec
 call, which may do real_path internaly).

 I think "bug" fix like this should go in early, hence this patch. I'm
 going to rebase nd/magic-pathspecs after this series graduates to
 master anyway, feel free to reject.

 submodule.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/submodule.c b/submodule.c
index 6b01a02..9f442d8 100644
--- a/submodule.c
+++ b/submodule.c
@@ -1102,7 +1102,8 @@ void connect_work_tree_and_git_dir(const char *work_tree, const char *git_dir)
 	struct strbuf configfile_name = STRBUF_INIT;
 	struct strbuf gitfile_content = STRBUF_INIT;
 	struct strbuf gitfile_name = STRBUF_INIT;
-	const char *real_work_tree = real_path(work_tree);
+	char *real_work_tree = xstrdup(real_path(work_tree));
+	char *to_free = real_work_tree;
 	const char *pathspec[] = { real_work_tree, git_dir, NULL };
 	const char *max_prefix = common_prefix(pathspec);
 	FILE *fp;
@@ -1157,4 +1158,5 @@ void connect_work_tree_and_git_dir(const char *work_tree, const char *git_dir)
 
 	strbuf_release(&core_worktree_setting);
 	strbuf_release(&configfile_name);
+	free(to_free);
 }
-- 
1.8.2.82.gc24b958