* Fix: Support Shoutcast HTTP responses like `ICY 200 OK`.
 * Fix: Don't unzip if there isn't a response body.
 * Fix: Don't leak gzip streams on redirects.
 * Fix: Don't do DNS lookups on invalid hosts.
 * Fix: Exhaust the underlying stream when reading gzip streams.
 * Fix: Support the `PATCH` method.
 * Fix: Support request bodies on `DELETE` method.
 * Fix: Drop the `okhttp-protocols` module.

        sampling if it encounters an empty dataset. This preserves the sampling
        ratio throughout training. The prior behavior was to continue sampling,
        skipping over exhausted datasets, until all datasets are exhausted. By
        default, the original behavior (`stop_on_empty_dataset=False`) is
        preserved.
    *   Removed previously deprecated tf.data statistics related APIs:

  }

  /**
   * Returns a new source that writes bytes to [cacheRequest] as they are read by the source
   * consumer. This is careful to discard bytes left over when the stream is closed; otherwise we
   * may never exhaust the source stream and therefore not complete the cached response.
   */
  @Throws(IOException::class)
  private fun cacheWritingResponse(
    cacheRequest: CacheRequest?,
    response: Response,
  ): Response {

considered a vulnerability in TensorFlow.

### Resource allocation

A denial of service caused by one model could bring down the entire server, but
we don't consider this as a vulnerability, given that models can exhaust
resources in many different ways and solutions exist to prevent this from
happening (e.g., rate limits, ACLs, monitors to restart broken servers).

### Model sharing

   * Returns the total number of chars read. Does not close the stream.
   *
   * @since 20.0
   */
  @CanIgnoreReturnValue
  public static long exhaust(Readable readable) throws IOException {
    long total = 0;
    long read;
    CharBuffer buf = createBuffer();
    while ((read = readable.read(buf)) != -1) {
      total += read;

        // See https://github.com/google/guava/issues/3542
        Reader reader = new StringReader(cannotDecode);
        InputStream decodingStream = encoding.decodingStream(reader);
        try {
          ByteStreams.exhaust(decodingStream);
          fail("Expected DecodingException");
        } catch (DecodingException expected) {
          // Don't assert on the expectedMessage; the messages for exceptions thrown from the

    } finally {
      closer.close();
    }

    closer = Closer.create();
    try {
      InputStream in = closer.register(openStream());
      return ByteStreams.exhaust(in);
    } catch (Throwable e) {
      throw closer.rethrow(e);
    } finally {
      closer.close();
    }
  }

  /** Counts the bytes in the given input stream using skip if possible. */

   * reached. Returns the total number of bytes read. Does not close the stream.
   *
   * @since 20.0
   */
  @CanIgnoreReturnValue
  public static long exhaust(InputStream in) throws IOException {
    long total = 0;
    long read;
    byte[] buf = createBuffer();
    while ((read = in.read(buf)) != -1) {
      total += read;
    }
    return total;
  }

 *  Fix: Don't corrupt the response cache when a 304 (Not Modified) response
    overrides the stored "Content-Encoding" header.
 *  Fix: Gracefully shut down the HTTP/2 connection before it exhausts the
    namespace of stream IDs (~536 million streams).
 *  Fix: Never pass a null `Route` to `Authenticator`. There was a bug where
    routes were omitted for eagerly-closed connections.

## Version 3.9.1

!g.startReady { // The first invocation sets the starting counter, if not fixed. g.startReady = true g.start = counter } counter -= g.start // Ensure the counter is strictly increasing. if counter == math.MaxUint64 { panic("crypto/cipher: counter exhausted") } if counter < g.next { panic("crypto/cipher: counter decreased or remained the same") } g.next = counter + 1 } fips140.RecordApproved() return g.g.sealAfterIndicator(dst, nonce, plaintext, data) } func (g *GCMWithCounterNonce) Open(dst, nonce,...