Fix --take-ownership

If a resource exists in the cluster and is to be adopted by helm install
--take-ownership, it is left unchanged while helm reports the
installation to have succeeded.

This is due to CRs and CRDs being merged without three-way-merge, which
results in an empty patch.

By using a three-way-merge transparently when --take-ownership is used,
the helm behaves as expected without breaking previous behavior.

Fixes #30622

Signed-off-by: Patrick Seidensal <pseidensal@suse.com>

pkg/kube/client_test.go

@@ -488,6 +488,14 @@ func TestUpdate(t *testing.T) {
 	}
 }
 
+func TestUpdate(t *testing.T) {
+	testUpdate(t, false)
+}
+
+func TestUpdateThreeWayMerge(t *testing.T) {
+	testUpdate(t, true)
+}
+
 func TestBuild(t *testing.T) {
 	tests := []struct {
 		name      string

pkg/kube/fake/failing_kube_client.go

@@ -125,6 +125,14 @@ func (f *FailingKubeClient) Update(r, modified kube.ResourceList, options ...kub
 	return f.PrintingKubeClient.Update(r, modified, options...)
 }
 
+// Update returns the configured error if set or prints
+func (f *FailingKubeClient) UpdateThreeWayMerge(r, modified kube.ResourceList, ignoreMe bool) (*kube.Result, error) {
+	if f.UpdateError != nil {
+		return &kube.Result{}, f.UpdateError
+	}
+	return f.PrintingKubeClient.Update(r, modified, ignoreMe)
+}
+
 // Build returns the configured error if set or prints
 func (f *FailingKubeClient) Build(r io.Reader, _ bool) (kube.ResourceList, error) {
 	if f.BuildError != nil {

pkg/kube/interface.go

@@ -52,6 +52,13 @@ type Interface interface {
 	GetWaiter(ws WaitStrategy) (Waiter, error)
 }
 
+// InterfaceThreeWayMerge was introduced to avoid breaking backwards compatibility for Interface implementers.
+//
+// TODO Helm 4: Remove InterfaceThreeWayMerge and integrate its method(s) into the Interface.
+type InterfaceThreeWayMerge interface {
+	UpdateThreeWayMerge(original, target ResourceList, force bool) (*Result, error)
+}
+
 // Waiter defines methods related to waiting for resource states.
 type Waiter interface {
 	// Wait waits up to the given timeout for the specified resources to be ready.