std::atomic<bool> extend_before_run;
};

struct TF_ImportGraphDefOptions {
  tensorflow::ImportGraphDefOptions opts;

  // Backing memory for TensorId fields in opts.
  // TODO(skyewm): it'd be better if ImportGraphDefOptions owned this.
  std::vector<tensorflow::string> tensor_id_data;
};

struct TF_ImportGraphDefResults {
  std::vector<TF_Output> return_tensors;
  std::vector<TF_Operation*> return_nodes;

the same address may also share the same underlying TCP socket connection. Sharing a connection has substantial performance benefits: lower latency, higher throughput (due to [TCP slow start](https://www.igvita.com/2011/10/20/faster-web-vs-tcp-slow-start/)) and conserved battery. OkHttp uses a [ConnectionPool](https://square.github.io/okhttp/4.x/okhttp/okhttp3/-connection-pool/) that automatically reuses HTTP/1.x connections and multiplexes HTTP/2 connections.

In OkHttp some fields of the address...

    * A "token" is just a string with some content that we can use later to verify this user.
    * Normally, a token is set to expire after some time.
        * So, the user will have to log in again at some point later.
        * And if the token is stolen, the risk is less. It is not like a permanent key that will work forever (in most of the cases).

    // Behold the weirdness!
    assertFalse(domain.publicSuffix().isPublicSuffix());
  }

  public void testPublicSuffixMultipleUnders() {
    // PSL has both *.uk and *.sch.uk; the latter should win.
    // See https://github.com/google/guava/issues/1176

    InternetDomainName domain = InternetDomainName.from("www.essex.sch.uk");
    assertTrue(domain.hasPublicSuffix());

   * under the assumption that for long-running instances, observed
   * contention levels will recur, so the cells will eventually be
   * needed again; and for short-lived ones, it does not matter.
   */

  /**
   * Padded variant of AtomicLong supporting only raw accesses plus CAS. The value field is placed
   * between pads, hoping that the JVM doesn't reorder them.
   *

   * query than the original; your mileage may vary. Precomputation takes time and requires more
   * memory, so it is only likely to be worthwhile if the precomputed matcher is queried very often.
   *
   * <p>This method has no effect (returns {@code this}) when called in GWT: it's unclear whether a

    return stringWriter.toString();
  }

  /**
   * Returns the stack trace of {@code throwable}, possibly providing slower iteration over the full
   * trace but faster iteration over parts of the trace. Here, "slower" and "faster" are defined in
   * comparison to the normal way to access the stack trace, {@link Throwable#getStackTrace()

### Is concurrency better than parallelism? { #is-concurrency-better-than-parallelism }

Nope! That's not the moral of the story.

Concurrency is different than parallelism. And it is better on **specific** scenarios that involve a lot of waiting. Because of that, it generally is a lot better than parallelism for web application development. But not for everything.

            <sourceFileExclude>**/ForceGuavaCompilation*</sourceFileExclude>
          </sourceFileExcludes>
          <!-- The above exclusion doesn't actually matter unless I prevent Javadoc from autodiscovering the source in the sourcepath, which defaults to the source directory :\ Boo for -sourcepath. -->
          <sourcepath>doesnotexist</sourcepath>

1. This is the generator function. It's a "generator function" because it contains `yield` statements inside.
2. By using a `with` block, we make sure that the file-like object is closed after the generator function is done. So, after it finishes sending the response.