Fix error on op selection (#17233)

With
https://github.com/dagster-io/dagster/pull/16640/files#diff-3a0d9c2e7d488c33f5be6fdf25298f9b03277b1bc73417d53e0a51688d32e8b6,
we started passing assetCheckSelection: [] for op jobs. This was
untested, and hit this invariant because unlike for assets where [] gets
turned in to None, we maintain a difference between None and [] with
asset checks

This is ripe for rethinking

python_modules/dagster-graphql/dagster_graphql/test/utils.py

@@ -195,12 +195,16 @@ def infer_pipeline_selector(
     graphql_context: WorkspaceRequestContext,
     pipeline_name: str,
     op_selection: Optional[Sequence[str]] = None,
+    asset_selection: Optional[Sequence[GqlAssetKey]] = None,
+    asset_check_selection: Optional[Sequence[GqlAssetCheckHandle]] = None,
 ) -> Selector:
     selector = infer_repository_selector(graphql_context)
     selector.update(
         {
             "pipelineName": pipeline_name,
             "solidSelection": op_selection,
+            "assetSelection": asset_selection,
+            "assetCheckSelection": asset_check_selection,
         }
     )
     return selector

python_modules/dagster-graphql/dagster_graphql_tests/graphql/test_execute_pipeline.py

@@ -687,6 +687,33 @@ def test_nested_graph_op_selection_and_config(self, graphql_context: WorkspaceRe
         assert step_did_not_run(logs, "plus_one")
         assert step_did_not_run(logs, "subgraph.op_2")
 
+    def test_nested_graph_op_selection_and_config_with_non_null_asset_and_check_selection(
+        self, graphql_context: WorkspaceRequestContext
+    ):
+        selector = infer_pipeline_selector(
+            graphql_context,
+            "nested_job",
+            ["subgraph.adder", "subgraph.op_1"],
+            asset_selection=[],
+            asset_check_selection=[],
+        )
+        run_config = {"ops": {"subgraph": {"ops": {"adder": {"inputs": {"num2": 20}}}}}}
+        result = sync_execute_get_run_log_data(
+            context=graphql_context,
+            variables={
+                "executionParams": {
+                    "selector": selector,
+                    "runConfigData": run_config,
+                }
+            },
+        )
+        logs = result["messages"]
+        assert isinstance(logs, list)
+        assert step_did_succeed(logs, "subgraph.adder")
+        assert step_did_succeed(logs, "subgraph.op_1")
+        assert step_did_not_run(logs, "plus_one")
+        assert step_did_not_run(logs, "subgraph.op_2")
+
     def test_memoization_job(self, graphql_context: WorkspaceRequestContext):
         selector = infer_pipeline_selector(graphql_context, "memoization_job")
         run_config = {}

python_modules/dagster/dagster/_core/instance/__init__.py

@@ -1461,7 +1461,11 @@ def create_run(
         check.opt_set_param(asset_selection, "asset_selection", of_type=AssetKey)
         check.opt_set_param(asset_check_selection, "asset_check_selection", of_type=AssetCheckKey)
 
-        if asset_selection is not None or asset_check_selection is not None:
+        # asset_selection will always be None on an op job, but asset_check_selection may be
+        # None or []. This is because [] and None are different for asset checks: None means
+        # include all asset checks on selected assets, while [] means include no asset checks.
+        # In an op job (which has no asset checks), these two are equivalent.
+        if asset_selection is not None or asset_check_selection:
             check.invariant(
                 op_selection is None,
                 "Cannot pass op_selection with either of asset_selection or asset_check_selection",