You can write <a href="https://en.wikipedia.org/wiki/Markdown" class="external-link" target="_blank">Markdown</a> in the docstring, it will be interpreted and displayed correctly (taking into account docstring indentation).

=== "Python 3.10+"

    ```Python hl_lines="17-25"
    {!> ../../../docs_src/path_operation_configuration/tutorial004_py310.py!}
    ```

=== "Python 3.9+"

                // For postfix case, the last argument is the operation call invoked on a synthetic local variable `<unary>`. This local
                // variable is initialized by calling the `get` function.
                val operationCall = lastArg as? FirFunctionCall ?: return null
                val operationPartiallyAppliedSymbol = operationCall.toPartiallyAppliedSymbol(arrayAccessExpression) ?: return null

## Untrusted inputs during training and prediction

TensorFlow supports a wide range of input data formats. For example it can
process images, audio, videos, and text. There are several modules specialized
in taking those formats, modifying them, and/or converting them to intermediate
formats that can be processed by TensorFlow.

These modifications and conversions are handled by a variety of libraries that

  // If specified, the time in seconds before the operation should be retried. Some errors may indicate
  // the client must take an alternate action - for those errors this field may indicate how long to wait
  // before taking the alternate action.
  // +optional
  optional int32 retryAfterSeconds = 5;
}

// TableOptions are used when a Table is requested by the caller.

     *
     * * **Unreachable proxy servers.** A [ProxySelector] can be used to
     *   attempt multiple proxy servers in sequence, eventually falling back to a direct
     *   connection.
     *
     * Set this to false to avoid retrying requests when doing so is destructive. In this case the
     * calling application should do its own recovery of connectivity failures.
     */
    fun retryOnConnectionFailure(retryOnConnectionFailure: Boolean) =

### Application Interceptors

Interceptors are registered as either _application_ or _network_ interceptors. We'll use the `LoggingInterceptor` defined above to show the difference.

Register an _application_ interceptor by calling `addInterceptor()` on `OkHttpClient.Builder`:

```java
OkHttpClient client = new OkHttpClient.Builder()
    .addInterceptor(new LoggingInterceptor())
    .build();

Request request = new Request.Builder()

    connections that were allocated but never released, and will enforce HTTP/2
    stream limits. This update also fixes `Call.cancel()` to not do I/O on the
    calling thread.
 *  Fix: Don't log gzipped data in the logging interceptor.
 *  Fix: Don't resolve DNS addresses when connecting through a SOCKS proxy.
 *  Fix: Drop the synthetic `OkHttp-Selected-Protocol` response header.

 * Improvements and fixes in Keras loss masking:
    * Whether you represent a ragged tensor as a `tf.RaggedTensor` or using [keras masking](https://www.tensorflow.org/guide/keras/masking_and_padding), the returned loss values should be the identical to each other. In previous versions Keras may have silently ignored the mask.

* kubectl: fixed an issue with "too old resource version" errors continuously appearing when calling `kubectl delete` ([#72825](https://github.com/kubernetes/kubernetes/pull/72825), [@liggitt](https://github.com/liggitt))

    when (this) {
        is SyntheticJavaPropertyDescriptor -> return KtSymbolOrigin.JAVA_SYNTHETIC_PROPERTY
        is SyntheticFieldDescriptor -> return KtSymbolOrigin.PROPERTY_BACKING_FIELD
        is SamConstructorDescriptor -> return KtSymbolOrigin.SAM_CONSTRUCTOR
        is JavaClassDescriptor, is JavaCallableMemberDescriptor -> return KtSymbolOrigin.JAVA