// Wraps a pointer to an operation implementation.
//
// WARNING: Since the underlying object could be ref-counted a user of this
// interface cannot destruct the underlying operation object. Instead, call
// TFE_DeleteOp who calls Release() on the operation pointer and deletes
// the TFE_Op structure.
typedef struct TFE_Op TFE_Op;

namespace tensorflow {

DEFINE_CONVERSION_FUNCTIONS(tensorflow::ImmediateExecutionOperation, TFE_Op);

//
// WARNING: Since the underlying object could be ref-counted a user of this
// interface cannot destruct the underlying handle object. Instead, call
// TFE_DeleteTensorHandle who calls Release() on the handle pointer and deletes
// the TFE_TensorHandle structure.
typedef struct TFE_TensorHandle TFE_TensorHandle;

namespace tensorflow {

)

load("@pypi//:requirements.bzl", "install_deps")

install_deps()

# Initialize the TensorFlow repository and all dependencies.
#
# The cascade of load() statements and tf_workspace?() calls works around the
# restriction that load() statements need to be at the top of .bzl files.
# E.g. we can not retrieve a new repository with http_archive and then load()
# a macro from that repository in the same file.

  // a tensor with the result.
  AbstractTensorHandle* AggregateGradients(
      gtl::ArraySlice<AbstractTensorHandle*> gradient_tensors) const override;

  // Calls the passed-in backward function.
  // op_type is the op's name provided in RecordOperation.
  Status CallBackwardFunction(
      const string& op_type, GradientFunction* gradient_function,

// This file extends/implements core graph optimizer base classes in terms of
// the C API defined in grappler.h. A class "CSomething" represents a
// "Something" that can be manipulated via calls in the C interface and a C
// struct called "TP_Something".

#include "tensorflow/c/experimental/grappler/grappler.h"

#include <algorithm>
#include <cstddef>
#include <cstring>
#include <string>

// ops return an error. This call clears the error state and re-enables
// execution of newly issued ops.
//
// Note that outputs of discarded ops remain in a corrupt state and should not
// be used for future calls.
// TODO(agarwal): mark the affected handles and raise errors if they are used.
TF_CAPI_EXPORT extern void TFE_ExecutorClearError(TFE_Executor*);

// Sets a custom Executor for the current thread. All nodes created by this

# under the root of srcjars jars created by bazel, rather than under
# the maven-style src/main/java subdirectory.
#
# Bazel manages annotation generated source as follows: First, it
# calls javac with options that create generated files under a
# bazel-out directory. Next, it archives the generated source files
# into a srcjar directly under the root. There doesn't appear to be a

        `compression=None` to `tf.data.experimental.service.distribute(...)`.
    *   `tf.data.Dataset.batch()` now supports `num_parallel_calls` and
        `deterministic` arguments. `num_parallel_calls` is used to indicate that
        multiple input batches should be computed in parallel. With
        `num_parallel_calls` set, `deterministic` is used to indicate that
        outputs can be obtained in the non-deterministic order.

  std::list<size_t> calls;
  auto fetcher = [&calls, block_size](const string& filename, size_t offset,
                                      size_t n, char* buffer,
                                      TF_Status* status) -> int64_t {
    EXPECT_EQ(n, block_size);
    EXPECT_FALSE(calls.empty()) << "at offset = " << offset;
    if (!calls.empty()) {
      EXPECT_EQ(offset, calls.front());
      calls.pop_front();
    }

TEST(ExpiringLRUCacheTest, LookupOrCompute) {
  // max_age of 0 means we should always compute.
  uint64 num_compute_calls = 0;
  tf_gcs_filesystem::ExpiringLRUCache<int>::ComputeFunc compute_func =
      [&num_compute_calls](const string& key, int* value, TF_Status* status) {
        *value = num_compute_calls;
        num_compute_calls++;
        return TF_SetStatus(status, TF_OK, "");
      };