/// `const`.
///
/// For consistency, the arguments on all these functions follow the same
/// pattern: first we have the opaque pointer argument ("this" above), then the
/// input arguments, then the in-out arguments (if any) and we finish the
/// argument list with the out arguments. We only use the return type for an out
/// parameter if that is a plain C type, as this ensures ABI compatibility

To set a different version, use the `TF_PYTHON_VERSION` environment variable,
e.g.

```
export TF_PYTHON_VERSION=3.11
```

To specify the version via a Bazel command argument, use the following:

```
--repo_env=TF_PYTHON_VERSION=3.11
```

## Requirements updater

Requirements updater is a standalone tool, intended to simplify process of

The C++ files in this directory (***\*_ops.h*** and ***\*_ops.cc***) are
autogenerated from the registered Op and API definitions.

To regenerate them, run the script in this directory, `update_cpp_ops.sh`, with
no arguments. This script will overwrite the existing ops in-place at
***tensorflow/c/experimental/ops/\*_ops.{cc,h}***.

Run this `update_cpp_ops.sh` script when Op definitions change in the registry.

# input the branch name and paste the cherry-pick commit and click Run. A PR
# will be created.

name: Release Branch Cherrypick
on:
  workflow_dispatch:
    inputs:
      # We use this instead of the "run on branch" argument because GitHub looks
      # on that branch for a workflow.yml file, and we'd have to cherry-pick
      # this file into those branches.
      release_branch:
        description: 'Release branch name (e.g. r2.9)'

  virtual ImmediateExecutionContext* GetContext() const = 0;

  // Following two methods are used to support custom device.
  // Return true if the inputs contain custom device tensor handle. It means
  // that the argument need to be handled by a custom device.
  virtual bool HasCustomDeviceInput() const = 0;

  virtual const tensorflow::OpDef* OpDef() const = 0;

  virtual Status InputLength(const char* input_name, int* length) = 0;

# -u: error if undefined variable used
# -o pipefail: entire command fails if pipe fails. watch out for yes | ...
# -o history: record shell history
set -euo pipefail -o history

# Check for required arguments
if [ -z "$1" ]; then
  echo "Usage: $0 <path_to_tensorflow_wheel> <python_version>"
  exit 1
fi

TENSORFLOW_WHEEL_PATH="$1"
PYTHON_VERSION="$2"

# All commands run relative to this directory

tf_cc_test(
    name = "modular_filesystem_test",
    size = "small",
    srcs = ["modular_filesystem_test.cc"],
    linkopts = ["-ldl"],
    tags = [
        "manual",  # Requires DSOs as arguments, eventual setup
        "notap",  # b/139060984, requires implementing modular support for Google filesystem
    ],
    deps = [
        ":modular_filesystem",
        "//tensorflow/core:framework_internal",

*   Extends `tf.keras.layers.Lambda` layers to support multi-argument lambdas,
    and keyword arguments when calling the layer.
*   Functional models now get constructed if *any* tensor in a layer call's
    arguments/keyword arguments comes from a keras input. Previously the
    functional api would only work if all of the elements in the first argument
    to the layer came from a keras input.

  // 1) nullptr indicates tensor is compact and row-majored.
  // 2) fill in the strides array as the real case for compact row-major data.
  // Here we choose option 2, since some frameworks didn't handle the strides
  // argument properly.
  dlm_tensor->dl_tensor.strides = stride_arr->data();

  dlm_tensor->dl_tensor.byte_offset =
      0;  // TF doesn't handle the strides and byte_offsets here
  return static_cast<void*>(dlm_tensor);
}

  DefineFunction("MyGrad2", &grad_func2);

  // func cannot be null
  TF_GraphCopyFunction(host_graph_, nullptr, func_, s_);
  EXPECT_EQ(TF_INVALID_ARGUMENT, TF_GetCode(s_));
  EXPECT_EQ(string("'func' argument to TF_GraphCopyFunction cannot be null"),
            string(TF_Message(s_)));

  // Cannot change gradient
  TF_GraphCopyFunction(host_graph_, func_, grad_func1, s_);