Optimize StringBuilder/StringBuffer serialization

[pandalee99]

What do these changes do?

To optimize the code, I avoiding the unnecessary string conversion and directly operating on the internal character array and length of the StringBuffer. This would involve modifying the write() and read() methods to work directly with the StringBuffer's internal data structure.

Related issue number

Closes #94

Check code requirements

• tests added / passed (if needed)
• Ensure all linting tests pass, see here for how to run them

[chaokunyang]

Hi @pandalee99 , thanks for taking time to contribute to fury, I left some comments, could you please take a look at it?

[chaokunyang]

StringBuilder didn't provide a zero-copy constructor, can we optimize write in this PR only to reduce complexity.

For example, we can keep protocol consistent with StringSerializer by :

  static Tuple2<ToByteFunction, Function> builderCache;

  private static synchronized Tuple2<ToByteFunction, Function> getBuilderFunc() {
    if (builderCache == null) {
      Function getValue = (Function) Functions.makeGetterFunction(
        StringBuilder.class.getSuperclass(), "getValue");
      if (Platform.JAVA_VERSION > 8) {
        ToByteFunction getCoder = (ToByteFunction) Functions.makeGetterFunction(
          StringBuilder.class.getSuperclass(), "getCoder");
        builderCache = Tuple2.of(getCoder, getValue);
      } else {
        builderCache = Tuple2.of(null, getValue);
      }
    }
    return builderCache;
  }

  public static abstract class AbstractStringBuilderSerializer<T> extends Serializer<T> {
    protected final ToByteFunction getCoder;
    protected final Function getValue;
    protected final StringSerializer stringSerializer;

    public AbstractStringBuilderSerializer(Fury fury, Class type) {
      super(fury, type);
      Tuple2<ToByteFunction, Function> builderFunc = getBuilderFunc();
      getCoder = builderFunc.f0;
      getValue = builderFunc.f1;
      stringSerializer = fury.getStringSerializer();
    }

    @Override
    public void write(MemoryBuffer buffer, T value) {
      if (Platform.JAVA_VERSION > 8) {
        byte coder = getCoder.applyAsByte(value);
        byte[] v = (byte[]) getValue.apply(value);
        buffer.writeByte(coder);
        buffer.writeBytesWithSizeEmbedded(v);
      } else {
        char[] v = (char[]) getValue.apply(value);
        if (StringSerializer.isAscii(v)) {
          stringSerializer.writeJDK8Ascii(buffer, v, v.length);
        } else {
          stringSerializer.writeJDK8UTF16(buffer, v, v.length);
        }
      }
    }
  }

  public static final class StringBuilderSerializer extends AbstractStringBuilderSerializer<StringBuilder> {
    
    public StringBuilderSerializer(Fury fury) {
      super(fury, StringBuilder.class);
    }

    @Override
    public StringBuilder read(MemoryBuffer buffer) {
      return new StringBuilder(stringSerializer.readJavaString(buffer));
    }
  }

Same thing can be done for StringBuffer

[pandalee99]

    char[] v = (char[]) getValue.apply(value);

Thank you, but I found a problem, the problem is, if using this way to get char[], then the final result is wrong, the length of char[] will be 19, the content will be "strnullnullnullnull..." This could be a jdk bug

[chaokunyang]

@pandalee99 io.fury.util.function.ToByteFunction is not defined in jdk library, we can't use it to generate accessor lambda for jdk classes, otherwise jdk will raise ClassNotFoundException. sorry for misleading comments.

Can we add a new makeFunction in Functions:

  public static Object makeGetterFunction(Method method) {
    return makeGetterFunction(method, method.getReturnType());
  }
  
  public static Object makeGetterFunction(Method method, Class<?> returnType) {
    MethodHandles.Lookup lookup = _JDKAccess._trustedLookup(method.getDeclaringClass());
    try {
      // Why `lookup.findGetter` doesn't work?
      // MethodHandle handle = lookup.findGetter(field.getDeclaringClass(), field.getName(),
      // field.getType());
      MethodHandle handle = lookup.unreflect(method);
      return _JDKAccess.makeGetterFunction(lookup, handle, returnType);
    } catch (IllegalAccessException ex) {
      throw new RuntimeException(ex);
    }
  }

then creating accessor by :

    Method getCoder = StringBuilder.class.getSuperclass().getDeclaredMethod("getCoder");
    ToIntFunction<CharSequence> o = (ToIntFunction<CharSequence>) makeGetterFunction(lookup, lookup.unreflect(getCoder), int.class);
    System.out.println(o);
    System.out.println(o.applyAsInt(new StringBuilder("abc")));
    System.out.println(o.applyAsInt(new StringBuilder("abc你好")));

[pandalee99]

thanks.
but,Some of the method of the Functions are repeated. and the lookup in the java.lang.CharacterName is private .maybe have other way, and the ToIntFunction maybe not able to Type conversion
and then,
How to solve the problem of local build failure in jdk9+? When I compile, it tells me that sun.misc does not exist

[pandalee99]

so good!

[chaokunyang]

LGTM, thank you so much for your contribution. It's not an easy work for such optimization