## 0.28.0

* Implement dependency cache per request.
    * This avoids calling each dependency multiple times for the same request.
    * This is useful while calling external services, performing costly computation, etc.

    }
  }

  /**
   * This verifies that putUnencodedChars(String) and hashUnencodedChars(String) are equivalent,
   * even for funny strings composed by (possibly unmatched, and mostly illegal) surrogate
   * characters. (But doesn't test that they do the right thing - just their consistency).
   */
  private static void assertHashStringWithSurrogatesEquivalence(

      rawSource.use {
        val source = rawSource.buffer()
        val urlLine = source.readUtf8LineStrict()
        // Choice here is between failing with a correct RuntimeException
        // or mostly silently with an IOException
        url = urlLine.toHttpUrlOrNull() ?: throw IOException("Cache corruption for $urlLine").also {
          Platform.get().log("cache corruption", WARN, it)
        }

 * <p>Path types are often exclusive. For example, a dependency should not be both on the Java class path
 * and on the Java module path.</p>
 *
 * <h2>Relationship with Java compiler standard location</h2>
 * This enumeration is closely related to the {@link JavaFileManager.Location} enumerations.
 * A difference is that the latter enumerates input and output files, while {@code JavaPathType}

   * collection-based APIs. Further, this method allows precise control over the runtime type of the
   * output array, and may, under certain circumstances, be used to save allocation costs.
   *
   * <p>Suppose {@code x} is a queue known to contain only strings. The following code can be used
   * to dump the queue into a newly allocated array of {@code String}:
   *
   * <pre>

channelsdvr.net
u.channelsdvr.net

// Fastly Inc. : http://www.fastly.com/
// Submitted by Fastly Security <security@fastly.com>
edgecompute.app
fastly-edge.com
fastly-terrarium.com
freetls.fastly.net
map.fastly.net
a.prod.fastly.net
global.prod.fastly.net
a.ssl.fastly.net
b.ssl.fastly.net
global.ssl.fastly.net
fastlylb.net
map.fastlylb.net

 *       <i>deeply</i> immutable.
 * </ul>
 *
 * <h4>Performance notes</h4>
 *
 * <ul>
 *   <li>Implementations can be generally assumed to prioritize memory efficiency, then speed of
 *       access, and lastly speed of creation.
 *   <li>The {@code copyOf} methods will sometimes recognize that the actual copy operation is
 *       unnecessary; for example, {@code copyOf(copyOf(anArrayList))} should copy the data only

 *       <i>deeply</i> immutable.
 * </ul>
 *
 * <h4>Performance notes</h4>
 *
 * <ul>
 *   <li>Implementations can be generally assumed to prioritize memory efficiency, then speed of
 *       access, and lastly speed of creation.
 *   <li>The {@code copyOf} methods will sometimes recognize that the actual copy operation is
 *       unnecessary; for example, {@code copyOf(copyOf(anArrayList))} should copy the data only

   * collection-based APIs. Further, this method allows precise control over the runtime type of the
   * output array, and may, under certain circumstances, be used to save allocation costs.
   *
   * <p>Suppose {@code x} is a queue known to contain only strings. The following code can be used
   * to dump the queue into a newly allocated array of {@code String}:
   *
   * <pre>

 *   long xBits = Double.doubleToRawLongBits(x);
 *   long yBits = Double.doubleToRawLongBits(y);
 *   return xBits == yBits;
 * }
 * }
 *
 * <p>It is possible to write a more scalable updater, at the cost of giving up strict atomicity.
 * See for example <a
 * href="http://gee.cs.oswego.edu/dl/jsr166/dist/docs/java.base/java/util/concurrent/atomic/DoubleAdder.html">
 * DoubleAdder</a>.
 *
 * @author Doug Lea