pathspec: only match across submodule boundaries when requestedbw/pathspec-match-submodule-boundary

Commit 74ed43711fd (grep: enable recurse-submodules to work on <tree>
objects, 2016-12-16) taught 'tree_entry_interesting()' to be able to
match across submodule boundaries in the presence of wildcards.  This is
done by performing literal matching up to the first wildcard and then
punting to the submodule itself to perform more accurate pattern
matching.  Instead of introducing a new flag to request this behavior,
commit 74ed43711fd overloaded the already existing 'recursive' flag in
'struct pathspec' to request this behavior.

This leads to a bug where whenever any other caller has the 'recursive'
flag set as well as a pathspec with wildcards that all submodules will
be indicated as matches.  One simple example of this is:

	git init repo
	cd repo

	git init submodule
	git -C submodule commit -m initial --allow-empty

	touch "[bracket]"
	git add "[bracket]"
	git commit -m bracket
	git add submodule
	git commit -m submodule

	git rev-list HEAD -- "[bracket]"

Fix this by introducing the new flag 'recurse_submodules' in 'struct
pathspec' and using this flag to determine if matches should be allowed
to cross submodule boundaries.

This fixes https://github.com/git-for-windows/git/issues/1371.

Signed-off-by: Brandon Williams <bmwill@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

4 files changed, 24 insertions, 2 deletions

diff --git a/builtin/grep.c b/builtin/grep.c
index 6946cf4bbc..4bbbce2a64 100644
--- a/builtin/grep.c
+++ b/builtin/grep.c
@@ -1015,6 +1015,7 @@ int cmd_grep(int argc, const char **argv, const char *prefix)
 		       prefix, argv + i);
 	pathspec.max_depth = opt.max_depth;
 	pathspec.recursive = 1;
+	pathspec.recurse_submodules = !!recurse_submodules;
 
 #ifndef NO_PTHREADS
 	if (list.nr || cached || show_in_pager)
diff --git a/pathspec.h b/pathspec.h
index 60e6500401..d000b63d80 100644
--- a/pathspec.h
+++ b/pathspec.h
@@ -24,6 +24,7 @@ struct pathspec {
 	int nr;
 	unsigned int has_wildcard:1;
 	unsigned int recursive:1;
+	unsigned int recurse_submodules:1;
 	unsigned magic;
 	int max_depth;
 	struct pathspec_item {
diff --git a/t/t4208-log-magic-pathspec.sh b/t/t4208-log-magic-pathspec.sh
index 935df6a65c..a1705f70cf 100755
--- a/t/t4208-log-magic-pathspec.sh
+++ b/t/t4208-log-magic-pathspec.sh
@@ -93,4 +93,23 @@ test_expect_success 'command line pathspec parsing for "git log"' '
 	git log --merge -- a
 '
 
+test_expect_success 'tree_entry_interesting does not match past submodule boundaries' '
+	test_when_finished "rm -rf repo submodule" &&
+	git init submodule &&
+	test_commit -C submodule initial &&
+	git init repo &&
+	>"repo/[bracket]" &&
+	git -C repo add "[bracket]" &&
+	test_tick &&
+	git -C repo commit -m bracket &&
+	git -C repo rev-list HEAD -- "[bracket]" >expect &&
+
+	git -C repo submodule add ../submodule &&
+	test_tick &&
+	git -C repo commit -m submodule &&
+
+	git -C repo rev-list HEAD -- "[bracket]" >actual &&
+	test_cmp expect actual
+'
+
 test_done
diff --git a/tree-walk.c b/tree-walk.c
index 6a42e402b0..d6de5315be 100644
--- a/tree-walk.c
+++ b/tree-walk.c
@@ -1011,7 +1011,8 @@ static enum interesting do_match(const struct name_entry *entry,
 				 * character.  More accurate matching can then
 				 * be performed in the submodule itself.
 				 */
-				if (ps->recursive && S_ISGITLINK(entry->mode) &&
+				if (ps->recurse_submodules &&
+				    S_ISGITLINK(entry->mode) &&
 				    !ps_strncmp(item, match + baselen,
 						entry->path,
 						item->nowildcard_len - baselen))
@@ -1060,7 +1061,7 @@ match_wildcards:
 		 * character.  More accurate matching can then
 		 * be performed in the submodule itself.
 		 */
-		if (ps->recursive && S_ISGITLINK(entry->mode) &&
+		if (ps->recurse_submodules && S_ISGITLINK(entry->mode) &&
 		    !ps_strncmp(item, match, base->buf + base_offset,
 				item->nowildcard_len)) {
 			strbuf_setlen(base, base_offset + baselen);