if err := syscall.Exec(os.Args[0], argv, os.Environ()); err != nil {
			t.Fatal(err)
		}
	}

	var cmds []*exec.Cmd
	defer func() {
		for _, cmd := range cmds {
			cmd.Process.Kill()
		}
	}()

	args := append(append([]string(nil), os.Args[1:]...), "-test.run=^Test18146$")
	for n := attempts; n > 0; n-- {
		cmd := exec.Command(os.Args[0], args...)
		cmd.Env = append(os.Environ(), "test18146=exec")

   * Section 7.2.1.2. The iteration order follows the lexicographical order. This means that the
   * first permutation will be in ascending order, and the last will be in descending order.
   *
   * <p>Duplicate elements are considered equal. For example, the list [1, 1] will have only one
   * permutation, instead of two. This is why the elements have to implement {@link Comparable}.
   *

 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 * 
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 * 

      values[dstIndex] = 0;
      entries[dstIndex] = UNSET;
    }
  }

  public void clear() {
    modCount++;
    Arrays.fill(keys, 0, size, null);
    Arrays.fill(values, 0, size, 0);
    Arrays.fill(table, UNSET);
    Arrays.fill(entries, UNSET);
    this.size = 0;
  }

enum ContextDevicePlacementPolicy {
  // Running operations with input tensors on the wrong device will fail.
  DEVICE_PLACEMENT_EXPLICIT = 0,
  // Copy the tensor to the right device but log a warning.
  DEVICE_PLACEMENT_WARN = 1,
  // Silently copy the tensor, which has a performance cost since the operation
  // will be blocked till the copy completes. This is the default policy.
  DEVICE_PLACEMENT_SILENT = 2,

                           TF_Status*);

// TF_ExecuteOperation will, if in eager mode, execute, if in graph mode, maybe
// capture some inputs and then add a node in the graph. The output tensors are
// returned through the provided TF_OutputList.
// Any active tape will observe the effects of this execution.
void TF_ExecuteOperation(TF_AbstractOp* op, int num_inputs,

                                                TF_Status* status);

// This function will block till the operation that produces `h` has
// completed. This is only valid on local TFE_TensorHandles. The pointer
// returned will be on the device in which the TFE_TensorHandle resides (so e.g.
// for a GPU tensor this will return a pointer to GPU memory). The pointer is

  LoadingCache<Integer, Integer> cache;

  int max;

  static AtomicLong requests = new AtomicLong(0);
  static AtomicLong misses = new AtomicLong(0);

  @BeforeExperiment
  void setUp() {
    // random integers will be generated in this range, then raised to the
    // power of (1/concentration) and floor()ed
    max = Ints.checkedCast((long) Math.pow(distinctKeys, concentration));

    cache =
        CacheBuilder.newBuilder()

  TF_EXPECT_OK(ReadCache(&cache, "", 0, block_size, &out));
  EXPECT_TRUE(first_block);
  EXPECT_EQ(out.size(), block_size);
  // Reading at offset file_size + 4 will read the second block (since the read
  // at file_size + 4 = 28 will be aligned to an offset of 16) but will return
  // OutOfRange because the offset is past the end of the 24-byte file.
  Status status = ReadCache(&cache, "", file_size + 4, 4, &out);

  + Make sure you fill in the earliest version that you know has the issue.
+ Fork the repository on GitHub.

Making and Submitting Changes
--------------

We accept Pull Requests via GitHub. The [developer mailing list][dev-ml-list] is the
main channel of communication for contributors.  
There are some guidelines which will make applying PRs easier for us: