checkout -p: handle tree arguments correctly again

This fixes a segmentation fault.

The bug is caused by dereferencing `new_branch_info->commit` when it is
`NULL`, which is the case when the tree-ish argument is actually a tree,
not a commit-ish. This was introduced in 5602b500c3 (builtin/checkout:
fix `git checkout -p HEAD...` bug, 2020-10-07), where we tried to ensure
that the special tree-ish `HEAD...` is handled correctly.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

builtin/checkout.c

@@ -479,9 +479,11 @@ static int checkout_paths(const struct checkout_opts *opts,
 		 * with the hex of the commit (whether it's in `...` form or
 		 * not) for the run_add_interactive() machinery to work
 		 * properly. However, there is special logic for the HEAD case
-		 * so we mustn't replace that.
+		 * so we mustn't replace that.  Also, when we were given a
+		 * tree-object, new_branch_info->commit would be NULL, but we
+		 * do not have to do any replacement, either.
 		 */
-		if (rev && strcmp(rev, "HEAD"))
+		if (rev && new_branch_info->commit && strcmp(rev, "HEAD"))
 			rev = oid_to_hex_r(rev_oid, &new_branch_info->commit->object.oid);
 
 		if (opts->checkout_index && opts->checkout_worktree)

t/t2016-checkout-patch.sh

@@ -123,4 +123,9 @@ test_expect_success PERL 'none of this moved HEAD' '
 	verify_saved_head
 '
 
+test_expect_success PERL 'empty tree can be handled' '
+	test_when_finished "git reset --hard" &&
+	git checkout -p $(test_oid empty_tree) --
+'
+
 test_done