* All changes should have an associated GitHub issue and/or a release note.
    * In the event that a bug cannot be automatically backported, the istio-testing bot creates an issue for a failed
    attempt and assigns it to the developer. This issue **is not** sufficient for requesting approval.
* For manual cherry-picks:
    * Large fixes, where the LOC>100 not withstanding unit tests changes, require subject-matter expert approval.

[XLA](https://www.tensorflow.org/xla) and
[JAX](https://jax.readthedocs.io/en/latest/jax-101/02-jitting.html)
documentation should be safe, while some of the testing and debugging tools that
come with the compiler are not designed to be used with untrusted data and
should be used with caution when working with untrusted models.

### Saved graphs and checkpoints

When loading untrusted serialized computation graphs (in form of a `GraphDef`,

  }

  /*
   * To avoid the need to call requireNonNull so much, we could consider more clever approaches,
   * such as:
   *
   * - Make checkOpen return the non-null `seq`. Then callers can assign that to a local variable or
   *   even back to `this.seq`. However, that may suggest that we're defending against concurrent
   *   mutation, which is not an actual risk because we use `synchronized`.

  }

  /*
   * To avoid the need to call requireNonNull so much, we could consider more clever approaches,
   * such as:
   *
   * - Make checkOpen return the non-null `seq`. Then callers can assign that to a local variable or
   *   even back to `this.seq`. However, that may suggest that we're defending against concurrent
   *   mutation, which is not an actual risk because we use `synchronized`.

import org.checkerframework.checker.nullness.qual.Nullable;

/**
 * A generalized interval on any ordering, for internal use. Supports {@code null}. Unlike {@link
 * Range}, this allows the use of an arbitrary comparator. This is designed for use in the
 * implementation of subcollections of sorted collection types.
 *
 * <p>Whenever possible, use {@code Range} instead, which is better supported.
 *
 * @author Louis Wasserman
 */

  ASSERT_TRUE(executed);
  auto handle_cleaner = tensorflow::gtl::MakeCleanup(
      [var_handle]() { TFE_DeleteTensorHandle(var_handle); });

  // Assign to the variable, copying to the custom device.
  std::unique_ptr<TFE_TensorHandle, decltype(&TFE_DeleteTensorHandle)> one(
      TestScalarTensorHandle(context.get(), 111.f), TFE_DeleteTensorHandle);

It is based on <a href="https://www.python-httpx.org" class="external-link" target="_blank">HTTPX</a>, which in turn is designed based on Requests, so it's very familiar and intuitive.

With it, you can use <a href="https://docs.pytest.org/" class="external-link" target="_blank">pytest</a> directly with **FastAPI**.

## Using `TestClient`

/// info

  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
  EXPECT_TF_META("v", -1, TF_ATTR_BOOL, -1);

  unsigned char value;
  TF_OperationGetAttrBool(oper, "v", &value, s_);
  EXPECT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
  EXPECT_EQ(1, value);
}

TEST_F(CApiAttributesTest, BoolList) {
  const unsigned char list[] = {0, 1, 1, 0, 0, 1, 1};
  const size_t list_size = TF_ARRAYSIZE(list);

You might remember that request bodies are also declared with Pydantic models.

Here **FastAPI** won't get confused because you are using `Depends`.

///

/// check

The way this dependency system is designed allows us to have different dependencies (different "dependables") that all return a `User` model.

We are not restricted to having only one dependency that can return that type of data.

///

## Other models

                  .ok());
  ASSERT_TRUE(worker_server2->Start().ok());

  TF_Status* status = TF_NewStatus();
  TFE_ContextOptions* opts = TFE_NewContextOptions();
  TFE_ContextOptionsSetAsync(opts, static_cast<unsigned char>(/*enable=*/true));
  TFE_ContextOptionsSetDevicePlacementPolicy(opts, TFE_DEVICE_PLACEMENT_SILENT);
  TFE_Context* ctx = TFE_NewContext(opts, status);
  EXPECT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);