checkVersionsOrder(b, c); // now b < c, but before MNG-5568, we had b > c
        checkVersionsOrder(a, c);
    }

    /**
     * Test <a href="https://jira.apache.org/jira/browse/MNG-6572">MNG-6572</a> optimization.
     */
    @Test
    void testMng6572() {
        String a = "20190126.230843"; // resembles a SNAPSHOT
        String b = "1234567890.12345"; // 10 digit number

// Test to ensure that a registered graph optimization pass is only executed
// once (i.e., on the main function side) in running distributed functions.
// This test creates a cluster with two workers, create a variable on the
// second worker, and run a distributed function (VariableAddFunction) whose ops
// span the local and remote workers. If the graph optimization pass is executed

       * doesn't need access to the futures again, so we can just pass `null`.
       *
       * TODO(b/112550045): Allocating a single, cheaper listener is (I think) only an optimization.
       * If we make some other optimizations, this one will no longer be necessary. The optimization
       * could actually hurt in some cases, as it forces us to keep all inputs in memory until the
       * final input completes.
       */
      @RetainedLocalRef

#### Optimizations

A naive implementation would simply regenerate all resources, of all subscribed types, for each client, on any configuration change. However, this scales poorly. As a result, we have many levels of optimizations to avoid doing this work.

        entry != null;
        entry = entry.getNextInKeyBucket()) {
      Object candidateKey = entry.getKey();

      /*
       * Assume that equals uses the == optimization when appropriate, and that
       * it would check hash codes as an optimization when appropriate. If we
       * did these things, it would just make things worse for the most
       * performance-conscious users.
       */

      if (compare(dividend, divisor) < 0) {
        return 0; // dividend < divisor
      } else {
        return 1; // dividend >= divisor
      }
    }

    // Optimization - use signed division if dividend < 2^63
    if (dividend >= 0) {
      return dividend / divisor;
    }

    /*
     * Otherwise, approximate the quotient, check, and correct if necessary. Our approximation is

       * crashes on Android when running in 32-bit mode. For maximum safety, we shouldn't use
       * Unsafe.getLong() at all, but the performance benefit on x86_64 is too great to ignore, so
       * as a compromise, we enable the optimization only on platforms that we specifically know to
       * work.
       *
       * In the future, the use of Unsafe.getLong() should be replaced by ByteBuffer.getLong(),

    return curr_version

  print('You have bazel %s installed.' % curr_version)
  return curr_version


def set_cc_opt_flags(environ_cp):
  """Set up architecture-dependent optimization flags.

  Also append CC optimization flags to bazel.rc..

  Args:
    environ_cp: copy of the os.environ.
  """
  if is_ppc64le():
    # gcc on ppc64le does not support -march, use mcpu instead

*   [TensorFlow Twitter](https://twitter.com/tensorflow)
*   [TensorFlow YouTube](https://www.youtube.com/channel/UC0rqucBdTuFTjJiefW5t-IQ)
*   [TensorFlow model optimization roadmap](https://www.tensorflow.org/model_optimization/guide/roadmap)
*   [TensorFlow White Papers](https://www.tensorflow.org/about/bib)
*   [TensorBoard Visualization Toolkit](https://github.com/tensorflow/tensorboard)

	// Limit the number of returned objects if > 0.
	Limit int

	// DiskID contains the disk ID of the disk.
	// Leave empty to not check disk ID.
	DiskID string
}

// supported FS for Nlink optimization in readdir.
const (
	xfs  = "XFS"
	ext4 = "EXT4"
)

// WalkDir will traverse a directory and return all entries found.
// On success a sorted meta cache stream will be returned.