revision: mark blobs needed for resolve-undo as reachable

The resolve-undo extension was added to the index in cfc5789a
(resolve-undo: record resolved conflicts in a new index extension
section, 2009-12-25).  This extension records the blob object names
and their modes of conflicted paths when the path gets resolved
(e.g. with "git add"), to allow "undoing" the resolution with
"checkout -m path".  These blob objects should be guarded from
garbage-collection while we have the resolve-undo information in the
index (otherwise unresolve operation may try to use a blob object
that has already been pruned away).

But the code called from mark_reachable_objects() for the index
forgets to do so.  Teach add_index_objects_to_pending() helper to
also add objects referred to by the resolve-undo extension.

Also make matching changes to "fsck", which has code that is fairly
similar to the reachability stuff, but have parallel implementations
for all these stuff, which may (or may not) someday want to be unified.

Signed-off-by: Junio C Hamano <gitster@pobox.com>

1 files changed, 71 insertions, 0 deletions

diff --git a/t/t2030-unresolve-info.sh b/t/t2030-unresolve-info.sh
index f691e6d903..2d8c70b03a 100755
--- a/t/t2030-unresolve-info.sh
+++ b/t/t2030-unresolve-info.sh
@@ -194,4 +194,75 @@ test_expect_success 'rerere forget (add-add conflict)' '
 	test_i18ngrep "no remembered" actual
 '
 
+test_expect_success 'resolve-undo keeps blobs from gc' '
+	git checkout -f main &&
+
+	# First make sure we do not have any cruft left in the object store
+	git repack -a -d &&
+	git prune --expire=now &&
+	git prune-packed &&
+	git gc --prune=now &&
+	git fsck --unreachable >cruft &&
+	test_must_be_empty cruft &&
+
+	# Now add three otherwise unreferenced blob objects to the index
+	git reset --hard &&
+	B1=$(echo "resolve undo test data 1" | git hash-object -w --stdin) &&
+	B2=$(echo "resolve undo test data 2" | git hash-object -w --stdin) &&
+	B3=$(echo "resolve undo test data 3" | git hash-object -w --stdin) &&
+	git update-index --add --index-info <<-EOF &&
+	100644 $B1 1	frotz
+	100644 $B2 2	frotz
+	100644 $B3 3	frotz
+	EOF
+
+	# These three blob objects are reachable (only) from the index
+	git fsck --unreachable >cruft &&
+	test_must_be_empty cruft &&
+	# and they should be protected from GC
+	git gc --prune=now &&
+	git cat-file -e $B1 &&
+	git cat-file -e $B2 &&
+	git cat-file -e $B3 &&
+
+	# Now resolve the conflicted path
+	B0=$(echo "resolve undo test data 0" | git hash-object -w --stdin) &&
+	git update-index --add --cacheinfo 100644,$B0,frotz &&
+
+	# These three blob objects are now reachable only from the resolve-undo
+	git fsck --unreachable >cruft &&
+	test_must_be_empty cruft &&
+
+	# and they should survive GC
+	git gc --prune=now &&
+	git cat-file -e $B0 &&
+	git cat-file -e $B1 &&
+	git cat-file -e $B2 &&
+	git cat-file -e $B3 &&
+
+	# Now we switch away, which nukes resolve-undo, and
+	# blobs B0..B3 would become dangling.  fsck should
+	# notice that they are now unreachable.
+	git checkout -f side &&
+	git fsck --unreachable >cruft &&
+	sort cruft >actual &&
+	sort <<-EOF >expect &&
+	unreachable blob $B0
+	unreachable blob $B1
+	unreachable blob $B2
+	unreachable blob $B3
+	EOF
+	test_cmp expect actual &&
+
+	# And they should go away when gc runs.
+	git gc --prune=now &&
+	git fsck --unreachable >cruft &&
+	test_must_be_empty cruft &&
+
+	test_must_fail git cat-file -e $B0 &&
+	test_must_fail git cat-file -e $B1 &&
+	test_must_fail git cat-file -e $B2 &&
+	test_must_fail git cat-file -e $B3
+'
+
 test_done