> and you can rely on it after doing initial sanity check and targeted local testing.

### Submitting Your Change

After you submit your pull request, a Gradle developer will review it. It is normal for this to take several iterations, so don't get discouraged by change requests. They ensure the high quality that we all enjoy.

If you need to check on [CI](http://builds.gradle.org/) status as an external contributor, you can click "Log in as guest".

  public void testExpectedFpp() {
    BloomFilter<Object> bf = BloomFilter.create(HashTestUtils.BAD_FUNNEL, 10, 0.03);
    double fpp = bf.expectedFpp();
    assertEquals(0.0, fpp);
    // usually completed in less than 200 iterations
    while (fpp != 1.0) {
      boolean changed = bf.put(new Object());
      double newFpp = bf.expectedFpp();
      // if changed, the new fpp is strictly higher, otherwise it is the same

			first = append(first, i)
		}

		// 800 chances to get a different iteration order.
		// See bug 8736 for why we need so many tries.
		for n := 0; n < 800; n++ {
			idx := 0
			for i := range m {
				if i != first[idx] {
					// iteration order changed.
					continue NextRound
				}
				idx++
			}
		}
		t.Fatalf("constant iteration order on round %d: %v", round, first)
	}
}

		if s.vars[i].name == name {
			s.vars[i].value = value
			return
		}
	}
	s.errorf("undefined variable: %s", name)
}

// setTopVar overwrites the top-nth variable on the stack. Used by range iterations.
func (s *state) setTopVar(n int, value reflect.Value) {
	s.vars[len(s.vars)-n].value = value
}

// varValue returns the value of the named variable.
func (s *state) varValue(name string) reflect.Value {

	// also no stale generation values left. Therefore, it's safe to flush
	// any buffers that remain in that generation's slot.
	const debugDeadlock = false
	systemstack(func() {
		// Track iterations for some rudimentary deadlock detection.
		i := 0
		detectedDeadlock := false

		for mToFlush != nil {
			prev := &mToFlush
			for mp := *prev; mp != nil; {
				if mp.trace.seqlock.Load()%2 != 0 {

		go runCommand(cb, bres)
		if got, want := <-ares, fmt.Sprintf("fd3: listener %s\n", la.Addr()); got != want {
			t.Errorf("iteration %d, process A got:\n%s\nwant:\n%s\n", i, got, want)
		}
		if got, want := <-bres, fmt.Sprintf("fd3: listener %s\n", lb.Addr()); got != want {
			t.Errorf("iteration %d, process B got:\n%s\nwant:\n%s\n", i, got, want)
		}
		la.Close()
		lb.Close()
		for _, f := range ca.ExtraFiles {

	// Walk blocks in postorder ordering. This improves convergence.
	po := lv.f.Postorder()

	// Iterate through the blocks in reverse round-robin fashion. A work
	// queue might be slightly faster. As is, the number of iterations is
	// so low that it hardly seems to be worth the complexity.

	for change := true; change; {
		change = false
		for _, b := range po {
			be := lv.blockEffects(b)

			newliveout.Clear()

		p = p[overage:]
		w.skipped += int64(overage)
	}
	p = w.fill(&w.suffix, p)

	// w.suffix is full now if p is non-empty. Overwrite it in a circle.
	for len(p) > 0 { // 0, 1, or 2 iterations.
		n := copy(w.suffix[w.suffixOff:], p)
		p = p[n:]
		w.skipped += int64(n)
		w.suffixOff += n
		if w.suffixOff == w.N {
			w.suffixOff = 0
		}
	}
	return lenp, nil
}

  }

  /**
   * Returns a view of a table where each value is transformed by a function. All other properties
   * of the table, such as iteration order, are left intact.
   *
   * <p>Changes in the underlying table are reflected in this view. Conversely, this view supports
   * removal operations, and these are reflected in the underlying table.
   *

    return (Multiset<T>) iterable;
  }

  /**
   * Returns a copy of {@code multiset} as an {@link ImmutableMultiset} whose iteration order puts
   * the highest count first, with ties broken by the iteration order of the original multiset.
   *
   * @since 11.0
   */
  public static <E> ImmutableMultiset<E> copyHighestCountFirst(Multiset<E> multiset) {