Do re-download corrupt objects with GVFS (#782)

As of 9e59b38, Git will loudly complain about corrupt objects.

That is fine, as long as the idea isn't to re-download locally-corrupted
objects. But that's exactly what we want to do in VFS for Git. This is
even tested for in the functional tests of VFS for Git, which has been
identified as a regression in
microsoft/VFSForGit#1853.

Author: dscho

commit.c

@@ -566,9 +566,14 @@ int repo_parse_commit_internal(struct repository *r,
 	int flags = OBJECT_INFO_LOOKUP_REPLACE | OBJECT_INFO_SKIP_FETCH_OBJECT |
 		    OBJECT_INFO_DIE_IF_CORRUPT;
 
-	/* But the GVFS Protocol _does_ support missing commits! */
+	/*
+	 * But the GVFS Protocol _does_ support missing commits!
+	 * And the idea with VFS for Git is to re-download corrupted objects,
+	 * not to fail!
+	 */
 	if (gvfs_config_is_set(GVFS_MISSING_OK))
-		flags ^= OBJECT_INFO_SKIP_FETCH_OBJECT;
+		flags &= ~(OBJECT_INFO_SKIP_FETCH_OBJECT |
+			   OBJECT_INFO_DIE_IF_CORRUPT);
 
 	if (!item)
 		return -1;

object-store.c

@@ -7,6 +7,7 @@
 #include "dir.h"
 #include "environment.h"
 #include "gettext.h"
+#include "gvfs.h"
 #include "gvfs-helper-client.h"
 #include "hex.h"
 #include "hook.h"
@@ -707,6 +708,9 @@ int read_object_process(const struct object_id *oid)
 	const char *cmd = find_hook(the_repository, "read-object");
 	uint64_t start;
 
+	if (!cmd)
+		die(_("could not find the `read-object` hook"));
+
 	start = getnanotime();
 
 	trace2_region_enter("subprocess", "read_object", the_repository);
@@ -1104,6 +1108,9 @@ void *repo_read_object_file(struct repository *r,
 	unsigned flags = OBJECT_INFO_DIE_IF_CORRUPT | OBJECT_INFO_LOOKUP_REPLACE;
 	void *data;
 
+	if (gvfs_config_is_set(GVFS_MISSING_OK))
+		flags &= ~OBJECT_INFO_DIE_IF_CORRUPT;
+
 	oi.typep = type;
 	oi.sizep = size;
 	oi.contentp = &data;

t/t1060-object-corruption.sh

@@ -3,6 +3,7 @@
 test_description='see how we handle various forms of corruption'
 
 . ./test-lib.sh
+. "$TEST_DIRECTORY"/lib-diff-data.sh
 
 # convert "1234abcd" to ".git/objects/12/34abcd"
 obj_to_file() {
@@ -62,6 +63,35 @@ test_expect_success 'streaming a corrupt blob fails' '
 	)
 '
 
+test_expect_success PERL 'truncated objects can be re-retrieved via GVFS' '
+	git init truncated &&
+	COPYING_test_data >truncated/COPYING &&
+	git -C truncated add COPYING &&
+	test_tick &&
+	git -C truncated commit -m initial COPYING &&
+
+	# set up the `read-object` hook so that it overwrites the corrupt object
+	mkdir -p truncated/.git/hooks &&
+	sed -e "1s|/usr/bin/perl|$PERL_PATH|" \
+	  -e "s/system/unlink \".git\/objects\/\" . substr(\$sha1, 0, 2) . \"\/\" . substr(\$sha1, 2); &/" \
+	  <$TEST_DIRECTORY/t0410/read-object \
+	  >truncated/.git/hooks/read-object &&
+	chmod +x truncated/.git/hooks/read-object &&
+
+	# ensure that the parent repository has a copy of the object, from
+	# where the `read-object` can read it
+	sha="$(git hash-object -w truncated/COPYING)" &&
+	file=$(obj_to_file $sha) &&
+	size=$(test_file_size $file) &&
+	chmod u+w truncated/$file &&
+	test-tool truncate truncated/$file $(($size-8)) &&
+
+	rm truncated/COPYING &&
+	test_must_fail git -C truncated reset --hard &&
+	git -C truncated -c core.gvfs=4 -c core.virtualizeObjects \
+		reset --hard
+'
+
 test_expect_success 'getting type of a corrupt blob fails' '
 	(
 		cd bit-error &&