// TODO if the destinations would change based on which cluster then add cluster to srCkey
							fromKey := from.Config().ClusterLocalFQDN()
							toKey := to.Config().ClusterLocalFQDN()
							if testTopology[fromKey] == nil {
								testTopology[fromKey] = map[string]int{}
							}
							testTopology[fromKey][toKey]++
						})
				},
				expect: map[string]map[string]int{
					"a.echo1.svc.cluster.local": {

					continue
				}
				if _, f := res[fromKey]; !f {
					res[fromKey] = map[Reference]*Grants{}
				}
				if _, f := res[fromKey][toKey]; !f {
					res[fromKey][toKey] = &Grants{
						AllowedNames: sets.New[string](),
					}
				}
				if to.Name != nil {
					res[fromKey][toKey].AllowedNames.Insert(string(*to.Name))
				} else {
					res[fromKey][toKey].AllowAll = true
				}
			}
		}
	}

		files, err := zipindex.ReadDir(b[len(b)-size:], objSize, nil)
		if err == nil {
			return files, gr.ObjInfo, nil
		}
		var terr zipindex.ErrNeedMoreData
		if errors.As(err, &terr) {
			size = int(terr.FromEnd)
			if size <= 0 || size > 100<<20 {
				return nil, ObjectInfo{}, errors.New("zip directory too large")
			}
		} else {
			return nil, ObjectInfo{}, err
		}
	}
}

  // This decomposes resource ops like ResourceGather into read-variable op
  // followed by gather. This is used when the saved model import path is used
  // during which resources don't get frozen in the python layer.
  pass_manager->addNestedPass<mlir::func::FuncOp>(
      mlir::TFDevice::CreateDecomposeResourceOpsPass());

  pass_manager->addPass(mlir::TF::CreateTFRegionControlFlowToFunctional());
}

	// build -n. We tried using a file that does not exist, but that fails on
	// systems with GCC version 4.2.1; that is the last GPLv2 version of GCC,
	// so some systems have frozen on it. Now we pass an empty file on stdin,
	// which should work at least for GCC and clang.
	//
	// If the argument is "-Wl,", then it is testing the linker. In that case,

    to the layer came from a keras input.
*   Clean up `BatchNormalization` layer's `trainable` property to act like
    standard python state when it's used inside `tf.functions` (frozen at
    tracing time), instead of acting like a pseudo-variable whose updates *kind
    of sometimes* get reflected in already-traced `tf.function` traces.
*   Add the `Conv1DTranspose` layer.

may resize the object.  The gradient TensorArray is statically sized based
on the size of the forward TensorArray when this operation executes.
Furthermore, the size of the forward TensorArray is frozen by this call.
As a result, the flow is used to ensure that the call to generate the gradient
TensorArray only happens after all writes are executed.