tensorflow::mutex_lock c(graph->mu);
  const auto& debug_str = graph->graph.ToGraphDefDebug().DebugString();
  *len = debug_str.size();
  char* ret = static_cast<char*>(malloc(*len + 1));
  memcpy(ret, debug_str.c_str(), *len + 1);
  return ret;
}

char* TF_FunctionDebugString(TF_Function* func, size_t* len) {
  const auto& debug_str = DebugString(func->record->fdef());
  *len = debug_str.size();

  CONFIG_TOGGLE(implementation_selector);
  CONFIG_TOGGLE(function_optimization);
  CONFIG_TOGGLE(common_subgraph_elimination);
  CONFIG_TOGGLE(arithmetic_optimization);
  CONFIG_TOGGLE(debug_stripper);
  CONFIG_TOGGLE(constant_folding);
  CONFIG_TOGGLE(shape_optimization);
  CONFIG_TOGGLE(auto_mixed_precision);
  CONFIG_TOGGLE(auto_mixed_precision_onednn_bfloat16);
  CONFIG_TOGGLE(auto_mixed_precision_mkl);

#include "tensorflow/core/framework/types.h"
#include "tensorflow/core/graph/graph.h"
#include "tensorflow/core/platform/base64.h"
#include "tensorflow/core/platform/strcat.h"
#include "tensorflow/core/util/debug_data_dumper.h"

using tensorflow::errors::InvalidArgument;

namespace tensorflow {
namespace {

Status ValidateNonRefOutput(const Node* node, int idx) {
  const DataType& dt = node->output_type(idx);

// explicitly by running a TPUReplicatedOutput operation on the parallel device.
//
// TODO(allenl): There are some use-cases that are only supported by copying to
// host at the moment (e.g. debug print on a tensor, .numpy(), etc.). We either
// need to return something here or address these use-cases one by one.
TFE_TensorHandle* CopyTensorFromParallelDevice(TFE_Context* context,

        "Unable to find function with name: ", function_name);
    return;
  }

  tensorflow::GraphDebugInfo debug_info =
      tensorflow::StackTracesMapToGraphDebugInfo(
          function_record->stack_traces());

  string str = debug_info.SerializeAsString();
  void* data = tensorflow::port::Malloc(str.length());
  str.copy(static_cast<char*>(data), str.length(), 0);
  buf->data = data;

                            char* buffer, TF_Status* status) -> int64_t {
    counter.DecrementCount();
    if (!counter.WaitFor(std::chrono::seconds(10))) {
      // This avoids having the test time out, which is harder to debug.
      TF_SetStatus(status, TF_FAILED_PRECONDITION,
                   "desired concurrency not reached");
      return -1;
    }
    memset(buffer, 'x', n);
    TF_SetStatus(status, TF_OK, "");
    return n;