} else {
      // Remove the stale block and continue.
      RemoveFile_Locked(key.first);
    }
  }

  // Insert a new empty block, setting the bookkeeping to sentinel values
  // in order to update them as appropriate.
  auto new_entry = std::make_shared<Block>();
  lru_list_.push_front(key);
  lra_list_.push_front(key);
  new_entry->lru_iterator = lru_list_.begin();

    [following API review practices](https://github.com/tensorflow/community/blob/master/governance/api-reviews.md).
*   When you contribute a new feature to TensorFlow, the maintenance burden is
    (by default) transferred to the TensorFlow team. This means that the benefit
    of the contribution must be compared against the cost of maintaining the
    feature.

// EagerService to maintain consistency between client and worker. The
// context_id is initialized with a dummy value and is later set when the worker
// is initialized (either locally or remotely). The context_id can change during
// the process lifetime although this should cause the worker to be
// reinitialized (e.g. cleared caches) as well.

// can correspond to the data contained in another dimension in on-host
// representation. The dimensions are indexed using the standard TensorFlow
// major-to-minor order (slowest varying dimension first),
// not the XLA's minor-to-major order.
// On-device dimensions can be padded. TFE_TensorDebugInfoOnDeviceDim returns
// the number of elements in a dimension after padding.

  /// be accessed while holding the Block's mu lock. The data vector should only
  /// be accessed after state == FINISHED, and it should never be modified.
  ///
  /// In order to prevent deadlocks, never grab the block-cache-wide mu_ lock
  /// AFTER grabbing any block's mu lock. It is safe to grab mu without locking
  /// mu_.
  struct Block {
    /// The block data.

  // from a backward gradient function. Fills `out_grads` corresponding to
  // `output_tensors`. `out_grads` must not be null.
  //
  // Executes the backward function in order to trace its gradient, which will
  // waste computation if executing eagerly (when graph building the unneeded
  // computation is pruned). Temporarily sets `backward_tape` so that

@@ -47,29 +47,7 @@
 #endif
 
 #if _STRING_ARCH_unaligned
-/* If we can do unaligned memory accesses we must know the endianess.  */
-# include <endian.h>
 # include <bits/types.h>
-
-# if __BYTE_ORDER == __LITTLE_ENDIAN
-#  define __STRING2_SMALL_GET16(src, idx) \
-     (((const unsigned char *) (const char *) (src))[idx + 1] << 8	      \
-      | ((const unsigned char *) (const char *) (src))[idx])

  };

  tensorflow::mutex execution_mutex_;
  ExecutionState execution_state_ TF_GUARDED_BY(execution_mutex_) =
      ExecutionState::kIdle;
  // Tells the worker thread that there is new work.
  tensorflow::condition_variable start_execute_;
  // The worker thread notifies that work has finished.
  tensorflow::condition_variable finished_execute_;
  // Notifies a StartExecute that the previous Join has finished.

  // parallel.
  //
  // Conceptually this is a thread pool with one thread per device. It requires
  // less synchronization than a thread pool would for this task, since Execute
  // acquires each thread in order (and so only one Execute will schedule
  // blocking collective operations at a time), and avoids some dynamic
  // allocation/scheduling.
  //
  // TODO(allenl): Keep a map from outer thread to list of inner threads rather