auto attrs = oper->node.attrs();
  int count = 0;
  AttrValueMap::const_iterator it;
  for (it = attrs.begin(); it != attrs.end(); it++) {
    if (count == i) {
      strncpy(output, it->first.c_str(), it->first.length());
      status->status = absl::OkStatus();
      return;
    }
    count++;
  }
  status->status = OutOfRange("Operation only has ", count,

  if (dst_fd < 0) {
    close(src_fd);
    return -1;
  }

  // Both files have been opened, do the transfer.
  // Since errno would be overridden by `close` below, save it here.
  int error_code = 0;
  if (CopyFileContents(dst_fd, src_fd, size) < 0) error_code = errno;

  close(src_fd);
  close(dst_fd);
  if (error_code != 0) {
    errno = error_code;
    return -1;
  } else {

void TF_SetTracingImplementation(const char* name, TF_Status* s) {
  tsl::Set_TF_Status_from_Status(s, SetDefaultTracingEngine(name));
}

// Creates a new TensorFlow function, it is an execution context attached to a
// given tracing context.
TF_ExecutionContext* TF_CreateFunction(const char* fn_name, TF_Status* s) {
  return wrap(CreateTracingExecutionContext(fn_name, s));
}

}

void RamFileBlockCache::RemoveFile_Locked(const std::string& filename) {
  Key begin = std::make_pair(filename, 0);
  auto it = block_map_.lower_bound(begin);
  while (it != block_map_.end() && it->first.first == filename) {
    auto next = std::next(it);
    RemoveBlock(it);
    it = next;
  }
}

void RamFileBlockCache::RemoveBlock(BlockMap::iterator entry) {

  EXPECT_EQ(cache.CacheSize(), 0);
}

TEST(RamFileBlockCacheTest, ParallelReads) {
  // This fetcher won't respond until either `callers` threads are calling it
  // concurrently (at which point it will respond with success to all callers),
  // or 10 seconds have elapsed (at which point it will respond with an error).
  const int callers = 4;
  BlockingCounter counter(callers);

// Transfers up to `size` bytes from `dst_fd` to `src_fd`.
//
// This method uses a temporary buffer to hold contents.
int CopyFileContents(int dst_fd, int src_fd, off_t size) {
  // Use a copy buffer of 128KB but don't store it on the stack.
  constexpr static size_t kPosixCopyFileBufferSize = 128 * 1024;
  std::unique_ptr<char[]> buffer(new char[kPosixCopyFileBufferSize]);

  off_t offset = 0;
  int bytes_transferred = 0;
  int rc = 1;

  // Although, `ops::SparseSoftmaxCrossEntropyWithLogits` returns 2 tensors, the
  // second tensor isn't needed for computing gradient so we could safely drop
  // it.
  outputs[0] = loss;
  backprop->Unref();
  return absl::OkStatus();
}

Status BiasAddModel(AbstractContext* ctx,
                    absl::Span<AbstractTensorHandle* const> inputs,

  // Import it in a fresh graph.
  TF_DeleteGraph(graph);
  graph = TF_NewGraph();
  TF_ImportGraphDefOptions* opts = TF_NewImportGraphDefOptions();
  TF_GraphImportGraphDef(graph, graph_def, opts, s);
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* scalar = TF_GraphOperationByName(graph, "scalar");

  // Import it in a fresh graph with an unused input mapping.

  EXPECT_EQ(num_compute_calls, 5);
  cache2.LookupOrCompute("d", &value, compute_func, &status);
  TF_EXPECT_OK(status.status);
  EXPECT_EQ(value, 3);
  EXPECT_EQ(num_compute_calls, 5);

  // Re-read "a", ensure it is re-computed.
  cache2.LookupOrCompute("a", &value, compute_func, &status);
  TF_EXPECT_OK(status.status);
  EXPECT_EQ(value, 5);
  EXPECT_EQ(num_compute_calls, 6);
}

TEST(ExpiringLRUCacheTest, Clear) {

    } else {
      // In sync mode, forwarding can't really happen since the client code will
      // have a reference count on the input tensor while the kernel is being
      // executed and thus it cannot be re-used for the return tensor.
      EXPECT_NE(orig_ptr, TF_TensorData(t));
    }
  }

  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_DeleteTensorHandle(m);