if (op_id_it == tensor_tape.end()) {
      continue;
    }
    int64_t op_id = op_id_it->second;
    auto op_it = op_tape->find(op_id);
    auto result_op_it = result.op_tape.find(op_id);
    if (op_id == -1 || op_it == op_tape->end() ||
        result_op_it != result.op_tape.end()) {
      continue;
    }
    CHECK(result.op_tape.emplace(op_id, op_it->second).second);

//
// TODO(ashankar): Merge with TF_Session?
typedef struct TFE_Context TFE_Context;

TF_CAPI_EXPORT extern TFE_Context* TFE_NewContext(
    const TFE_ContextOptions* opts, TF_Status* status);
TF_CAPI_EXPORT extern void TFE_DeleteContext(TFE_Context* ctx);
TF_CAPI_EXPORT extern TF_DeviceList* TFE_ContextListDevices(TFE_Context* ctx,
                                                            TF_Status* status);

  virtual ~ImmediateExecutionDistributedManager() {}

  // Set up distributed execution environment on local and remote tasks.
  // When `reset_context` is true, initialize new cluster context state based
  // on cluster configurations provided in `server_def`; otherwise, update
  // existing context state with the provided `server_def`. Contexts created
  // on remote tasks will be considered stale and garbage collected after

typedef struct TFE_OpAttrs TFE_OpAttrs;

typedef struct TFE_Context TFE_Context;
typedef struct TFE_Op TFE_Op;

namespace tensorflow {
DEFINE_CONVERSION_FUNCTIONS(tensorflow::AbstractOpAttrs, TFE_OpAttrs);

// Set an AttrValue on the op. Doesn't handle the list types.
void SetOpAttrValueScalar(TFE_Context* ctx, TFE_Op* op,
                          const tensorflow::AttrValue& default_value,

#ifndef TENSORFLOW_C_EAGER_IMMEDIATE_EXECUTION_CONTEXT_H_
#define TENSORFLOW_C_EAGER_IMMEDIATE_EXECUTION_CONTEXT_H_

#include <functional>
#include <map>
#include <memory>
#include <vector>

#include "absl/types/optional.h"
#include "absl/types/span.h"
#include "tensorflow/c/eager/abstract_context.h"
#include "tensorflow/c/eager/immediate_execution_distributed_manager.h"

  }
};

// An abstract operation describes an operation by its type, name, and
// attributes. It can be "executed" by the context with some input tensors.
// It is allowed to reusing the same abstract operation for multiple execution
// on a given context, with the same or different input tensors.
class TracingOperation : public AbstractOperation {
 protected:
  explicit TracingOperation(AbstractOperationKind kind)

#ifndef TENSORFLOW_C_EAGER_GRADIENTS_H_
#define TENSORFLOW_C_EAGER_GRADIENTS_H_

#include "absl/container/flat_hash_map.h"
#include "tensorflow/c/eager/abstract_context.h"
#include "tensorflow/c/eager/abstract_tensor_handle.h"
#include "tensorflow/c/eager/tape.h"
#include "tensorflow/core/common_runtime/eager/attr_builder.h"

namespace tensorflow {
namespace gradients {

#include "absl/synchronization/notification.h"
#include "tensorflow/c/env.h"
#include "tensorflow/c/logging.h"
#include "tensorflow/c/tf_status.h"

namespace tf_gcs_filesystem {

/// \brief An LRU block cache of file contents, keyed by {filename, offset}.
///
/// This class should be shared by read-only random access files on a remote
/// filesystem (e.g. GCS).
class RamFileBlockCache {
 public:

  /// Flushes `*file` and syncs contents to filesystem.
  ///
  /// This call might not block, and when it returns the contents might not have
  /// been fully persisted.
  ///
  /// DEFAULT IMPLEMENTATION: No op.
  void (*flush)(const TF_WritableFile* file, TF_Status* status);

  /// Syncs contents of `*file` with the filesystem.
  ///

// the TFE_Context structure.
typedef struct TFE_Context TFE_Context;

namespace tensorflow {

DEFINE_CONVERSION_FUNCTIONS(tensorflow::ImmediateExecutionContext, TFE_Context);

}  // namespace tensorflow