* this graph.
   *
   * <p>This is equal to the union of {@link #predecessors(Object)} and {@link #successors(Object)}.
   *
   * <p>If {@code node} is removed from the graph after this method is called, the {@code Set}
   * {@code view} returned by this method will be invalidated, and will throw {@code
   * IllegalStateException} if it is accessed in any way, with the following exceptions:
   *
   * <ul>

  }

  /** Throw {@link IllegalStateException} if {@link #createTestSuite()} can't be called yet. */
  protected void checkCanCreate() {
    if (subjectGenerator == null) {
      throw new IllegalStateException("Call using() before createTestSuite().");
    }
    if (name == null) {
      throw new IllegalStateException("Call named() before createTestSuite().");
    }
    if (features == null) {

      loader.reset();
      assertThat(cache.getUnchecked(KEY_PREFIX + i)).isEqualTo(VALUE_PREFIX + i);
      assertWithMessage("Loader should NOT have been called @#%s", i)
          .that(loader.wasCalled())
          .isFalse();
    }

    // wait for entries to expire, but don't call expireEntries
    ticker.advance(EXPIRING_TIME * 10, MILLISECONDS);

    // add a single unexpired entry
    cache.getUnchecked(KEY_PREFIX + 11);

        all scripts can use them.
    -   `utilities/setup_docker.sh` creates a container called `tf` with all
        `TFCI_` variables shared to it.
3.  Top-level scripts (`wheel.sh`, etc.) reference `env` variables and call
    `utilities/` scripts.
    -   The `tfrun` function makes a command run correctly in Docker if Docker

  /**
   * Returns a new builder, expecting the specified number of entries to be added.
   *
   * <p>If {@code expectedSize} is exactly the number of entries added to the builder before {@link
   * Builder#build} is called, the builder is likely to perform better than an unsized {@link
   * #builder()} would have.
   *
   * <p>It is not specified if any performance benefits apply if {@code expectedSize} is close to,

    @Override public EventListener create(Call call) {
      long callId = nextCallId.getAndIncrement();
      System.out.printf("%04d %s%n", callId, call.request().url());
      return new PrintingEventListener(callId, System.nanoTime());
    }
  };

  final long callId;
  final long callStartNanos;

  public PrintingEventListener(long callId, long callStartNanos) {
    this.callId = callId;

      /*
       * requireNonNull is safe because we don't null out `sequencer` except:
       *
       * - after the requireNonNull call below. (And this object has its Runnable.run override
       *   called only once, just as it has its Executor.execute override called only once.)
       *
       * - if we return immediately from `execute` (in which case we never get here)
       *

	// If set to nil, it will not be called.

	// agreed is called if all disks agreed.
	agreed func(entry metaCacheEntry)

	// partial will be called when there is disagreement between disks.
	// if disk did not return any result, but also haven't errored
	// the entry will be empty and errs will
	partial func(entries metaCacheEntries, errs []error)

	// finished will be called when all streams have finished and

  /**
   * Returns a new builder, expecting the specified number of entries to be added.
   *
   * <p>If {@code expectedSize} is exactly the number of entries added to the builder before {@link
   * Builder#build} is called, the builder is likely to perform better than an unsized {@link
   * #builder()} would have.
   *
   * <p>It is not specified if any performance benefits apply if {@code expectedSize} is close to,

### Async Context Manager { #async-context-manager }

If you check, the function is decorated with an `@asynccontextmanager`.

That converts the function into something called an "**async context manager**".

{* ../../docs_src/events/tutorial003_py39.py hl[1,13] *}