Update java-extension.md

Author: taowen

java-extension.md

@@ -12,7 +12,7 @@ The general direction to provide extensiblity, is to give the user a callback, i
 All kinds of annotation support can be implemented using the callback. Also the callback should control the codegen if possible, not 
 slowing down the runtime execution.
 
-# Decoder interface
+# Customize type decoding
 
 ```java
 public interface Decoder {
@@ -64,3 +64,43 @@ JsonIterator iter = JsonIterator.parse("\"1481365190000\"");
 List<Integer> listOfInt = iter.read(listOfIntType);
 System.out.println(listOfInt); // [1, 4, 8, 1, 3, 6, 5, 1, 9, 0, 0, 0, 0]
 ```
+
+# Customize field decoding
+
+Sometimes, we can not just customize the type, all different occurence of the type will need different encoding. So we can precisely control individual field decoding. The callback interface is still `Decoder`, we just need to register it to field instead of type.
+
+```java
+public class TestObject1 {
+    public String field1;
+}
+
+JsonIterator.registerFieldDecoder(TestObject1.class, "field1", new Decoder() {
+
+    @Override
+    public Object decode(JsonIterator iter) throws IOException {
+        return Integer.toString(iter.readInt());
+    }
+});
+JsonIterator iter = JsonIterator.parse("{'field1': 100}".replace('\'', '"'));
+TestObject1 myObject = iter.read(TestObject1.class);
+assertEquals("100", myObject.field1);
+```
+
+For field decoding, there is one thing special: primitive types. If the field is of type like `int` or `long`, we do not want to return `Object` from `Decoder` then unbox it to primitive types. The decoder registered by user should not be forced to go through this process. And `IntDecoder` is what we need:
+
+```java
+public class TestObject2 {
+    public int field1;
+}
+
+JsonIterator.registerFieldDecoder(TestObject2.class, "field1", new Decoder.IntDecoder() {
+
+    @Override
+    public int decodeInt(JsonIterator iter) throws IOException {
+        return Integer.valueOf(iter.readString());
+    }
+});
+JsonIterator iter = JsonIterator.parse("{'field1': '100'}".replace('\'', '"'));
+TestObject2 myObject = iter.read(TestObject2.class);
+assertEquals(100, myObject.field1);
+```