//    only worker 1.
// 2. Create a context B using A.
// 3. Create the variable in B.
// 4. Create another single host server def C with only worker 0.
// 5. Start the GRPC server for worker 0 using C.
// 6. Create a context D using the full cluster server def E.
// 7. Read the variable in D.
TEST(CAPI, SingleHostServerDefV1Works) {
  // Create a server def that represents a 2-process cluster and a client.

  const int64_t n = data.size();
  Status status;
  for (const std::vector<int64_t>& dims :
       std::vector<std::vector<int64_t>>{{n}, {1, n}, {n, 1}, {n / 2, 2}}) {
    // Create C++ Tensor
    Tensor src(tensorflow::DT_STRING, TensorShape(dims));
    for (int64_t i = 0; i < src.NumElements(); ++i) {
      src.flat<tstring>()(i) = data[i];
    }

#include "tensorflow/core/platform/stacktrace_handler.h"
#include "tensorflow/core/platform/test.h"
#include "tensorflow/core/util/command_line_flags.h"

#if defined(PLATFORM_WINDOWS)
// Make mkdir resolve to _mkdir to create the test temporary directory.
#include <direct.h>
#define mkdir(name, mode) _mkdir(name)

// Windows defines the following macros to convert foo to fooA or fooW,

  void RunT(const std::vector<std::pair<TF_Operation*, TF_Tensor*>>& inputs,
            std::initializer_list<TF_Output> outputs,
            const std::vector<std::vector<int32_t>>& expected_results) {
    // Create a session for this graph
    CSession csession(host_graph_, s_);
    ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);

    // Run
    csession.SetInputs(inputs);
    csession.SetOutputs(outputs);

  /// missing directory entry and all its descendants will be created by the
  /// plugin.
  ///
  /// DEFAULT IMPLEMENTATION: Creates directories one by one. Needs
  /// `path_exists`, `is_directory`, and `create_dir`.
  void (*recursively_create_dir)(const TF_Filesystem* filesystem,
                                 const char* path, TF_Status* status);

  /// Deletes the file specified by `path`.
  ///
  /// Plugins:

# -------------------------------------------

# TensorFlow uses several libraries that may also be used by applications
# linking against the C and C++ APIs (such as libjpeg).  When we create
# the shared library, only export the core TF API functions to avoid
# causing library conflicts (e.g., those reported in github issue 1924).
# On Linux, tell the linker (-Wl,<option>) to use a version script that

              'Android builds?'), 'Searching for NDK and SDK installations.',
             'Not configuring the WORKSPACE for Android builds.'):
    create_android_ndk_rule(environ_cp)
    create_android_sdk_rule(environ_cp)

  system_specific_test_config(environ_cp)

  configure_ios(environ_cp)

  print('Preconfigured Bazel build configs. You can use any of the below by '

# Configure short or long logs
build:short_logs --output_filter=DONT_MATCH_ANYTHING
build:verbose_logs --output_filter=

# Instruction set optimizations
# TODO(gunan): Create a feature in toolchains for avx/avx2 to
#   avoid having to define linux/win separately.
build:avx_linux --copt=-mavx
build:avx_linux --host_copt=-mavx
build:avx_win --copt=/arch:AVX

TF_CAPI_EXPORT extern TF_Output TF_OperationInput(TF_Input oper_in);

// Get list of all inputs of a specific operation.  `inputs` must point to
// an array of length at least `max_inputs` (ideally set to
// TF_OperationNumInputs(oper)).  Beware that a concurrent
// modification of the graph can increase the number of inputs of
// an operation.
TF_CAPI_EXPORT extern void TF_OperationAllInputs(TF_Operation* oper,