for it.

If the feature has merit, it will go through a thorough process of API design
and review. Any code should come after this.

[APIs]: https://en.wikipedia.org/wiki/Application_programming_interface
[issue]: https://github.com/google/guava/issues

Pull requests
-------------

Unless the change is a trivial fix such as for a typo, it's generally best to

   *
   * @since 19.0
   */
  public static final class UnsupportedLoadingOperationException
      extends UnsupportedOperationException {
    // Package-private because this should only be thrown by loadAll() when it is not overridden.
    // Cache implementors may want to catch it but should not need to be able to throw it.
    UnsupportedLoadingOperationException() {}

 *       Guava while continuing to run it internally, as we do with many other tests. This would
 *       suffice because we our Android users and tests are using the open-source version, which
 *       would no longer have the problematic test. But why bother when we can instead strip it with
 *       a more precisely named annotation?
 *   <li>While a dependency on Android ought to be easy if it's for annotations only, it will

      }
    }
  }

  /**
   * Closes the given {@link InputStream}, logging any {@code IOException} that's thrown rather than
   * propagating it.
   *
   * <p>While it's not safe in the general case to ignore exceptions that are thrown when closing an
   * I/O resource, it should generally be safe in the case of a resource that's being used only for

    The summary is that it ignores one file in favor of the other.
    util.concurrent, like nearly all our packages, has two .gwt.xml files: one
    for prod and one for tests. However, unlike our other packages, as of this
    writing it has test supersource but no prod supersource.

    GWT happens to use the prod .gwt.xml, so it looks for no supersource for
    tests, either. This causes it to fail to find AtomicLongMapTest.

   * safe range. If {@code safeMax < safeMin} then no code points are considered safe.
   *
   * <p>If a code point has no mapped replacement then it is checked against the safe range. If it
   * lies outside that, then {@link #escapeUnsafe} is called, otherwise no escaping is performed.
   *
   * @param replacementMap a map of characters to their escaped representations

   *
   * <ul>
   *   <li>If {@code exceptions} has a single exception and that exception is a {@link
   *       RuntimeException}, return it
   *   <li>If {@code exceptions} has a single exceptions and that exceptions is not a
   *       {@link RuntimeException}, return a simple {@code RuntimeException} that wraps it
   *   <li>Otherwise, return an instance of {@link ClusterException} that wraps the first exception

       * to test it.
       */
      return;
    }

    /*
     * Only under Windows (or hypothetically when running with some other non-POSIX, ACL-based
     * filesystem) does our prod code look up the username. Thus, this test doesn't necessarily test
     * anything interesting under most environments. Still, we can run it (except for Android, at

    assertEquals(1, runLatch.getCount());
  }

  public void testListenerCalledOnCancelFromRunning() throws Exception {
    exec.execute(task);
    runLatch.await();

    // Task has started up, cancel it while it's running.
    task.cancel(true);
    assertTrue(task.isDone());
    assertTrue(task.isCancelled());
    assertEquals(1, taskLatch.getCount());

    // Wait for the listeners to be called.

    tearDownStack.runTearDown();
  }

  /**
   * This first test doesn't test anything in Uninterruptibles, just demonstrates some normal
   * behavior of futures so that you can contrast the next test with it.
   */
  public void testRegularFutureInterrupted() throws ExecutionException {

    /*
     * Here's the order of events that we want.
     *
     * 1. The client thread begins to block on a get() call to a future.