As we were attempting to migrate to 2.18 some of our internal functional
tests failed. The tests that failed were testing merge and cherry-pick
when there was a merge conflict. Our tests run with sparse-checkout
enabled which is what exposed the bug.
What is happening is that in merge_recursive, the skip-worktree bit is
cleared on the cache entry but then the working directory isn't updated.
The end result is that git reports that the merged file has actually
been deleted.
We've identified the patch that introduced the regression as:
commit 1de70dbd1ada0069d1b6cd6345323906cc9a9ed3
Author: Elijah Newren <newren@xxxxxxxxx>
Date: Thu Apr 19 10:58:23 2018 -0700
merge-recursive: fix check for skipability of working tree updates
The can-working-tree-updates-be-skipped check has had a long and
blemished
history. The update can be skipped iff:
a) The merge is clean
b) The merge matches what was in HEAD (content, mode, pathname)
c) The target path is usable (i.e. not involved in D/F conflict)
I've written a test that can be used to reproduce the issue:
diff --git a/t/t3507-cherry-pick-conflict.sh
b/t/t3507-cherry-pick-conflict.sh
index 7c5ad08626..de0bdc8634 100755
--- a/t/t3507-cherry-pick-conflict.sh
+++ b/t/t3507-cherry-pick-conflict.sh
@@ -392,4 +392,17 @@ test_expect_success 'commit --amend -s places the
sign-off at the right place' '
test_cmp expect actual
'
+test_expect_success 'failed cherry-pick with sparse-checkout' '
+ pristine_detach initial &&
+ git config core.sparseCheckout true &&
+ echo /unrelated >.git/info/sparse-checkout &&
+ git read-tree --reset -u HEAD &&
+ test_must_fail git cherry-pick -Xours picked>actual &&
+ test_i18ngrep ! "Changes not staged for commit:" actual &&
+ echo "/*" >.git/info/sparse-checkout &&
+ git read-tree --reset -u HEAD &&
+ git config core.sparseCheckout false &&
+ rm .git/info/sparse-checkout
+'
+
test_done
Thanks,
Ben
