FULL PRODUCT VERSION :
java version "1.7.0_51"
Java(TM) SE Runtime Environment (build 1.7.0_51-b13)
Java HotSpot(TM) Client VM (build 24.51-b03, mixed mode, sharing)

A DESCRIPTION OF THE PROBLEM :
This bug requires 2 conditions to be triggered:
- 1 or more consecutive nested character classes (surrounded by [ ]) following right after an intersection &&
- Right after the series of nested character classes mentioned above is at least one "flat" character class

Then the series of nested character classes are lost.

For example:
// Working cases
String ra = "[A-Z&&[0-9A-Z]]"; // Nothing after the nested character class
String rb = "[A-Z&&0-9A-Z]"; // Everything is "flat"
String rc = "[A-Z&&0-9[A-Z]]"; // [A-Z] is after "flat" character class 0-9
 
// Failing cases
String rx = "[A-Z&&[A-Z]0-9]"; // [A-Z] is lost
String ry = "[A-Z&&[A-F][G-Z]0-9]"; // [A-F][G-Z] is lost


REPRODUCIBILITY :
This bug can be reproduced always.

---------- BEGIN SOURCE ----------
/* Author: Hong Dai Thanh/nhahtdh */
import java.lang.reflect.*;
import java.util.regex.*;

class BugLostCharacterClass {
    private static Class<?> CharProperty;
    private static Class<?> Node;
    
    static {
        try {
            CharProperty = Class.forName("java.util.regex.Pattern$CharProperty");
            Node = Class.forName("java.util.regex.Pattern$Node");
        } catch (Exception e) {
            throw new RuntimeException(e);
        }
    }
    
    private static void indent(int level) {
        System.out.print(new String(new char[level * 3]).replace('\0',' '));
    }
    
    private static void dissectCharClass(int indent, Object o) throws Exception {
        Field[] fields = o.getClass().getDeclaredFields();
        
        Method enclosing = o.getClass().getEnclosingMethod();
        
        indent(indent);
        System.out.println((enclosing != null ? enclosing.getName() + " " : "") + o.getClass().getName());
        
        for (Field f: fields) {
            f.setAccessible(true);
            Object c = f.get(o);
            
            if (CharProperty.isAssignableFrom(f.getType())) {
                dissectCharClass(indent + 1, c);
            } else {
                indent(indent + 1);
                System.out.println(f.getName() + ":" + c);
            }
        }
    }
    
    private static void dissectNode(Object o) throws Exception {
        Field nextField = Node.getDeclaredField("next");
        nextField.setAccessible(true);

        while (o != null) {
            if (CharProperty.isInstance(o)) {
                dissectCharClass(0, o);
            } else {
                System.out.println(o.getClass().getName());
            }
            
            o = nextField.get(o);
        }
    }
    
    private static void dissectPattern(Pattern p) throws Exception {
        Field rootField = Pattern.class.getDeclaredField("root");
        rootField.setAccessible(true);
        
        Object root = rootField.get(p);
        
        dissectNode(root);
        System.out.println();
    }
    
    public static void main(String args[]) throws Exception {
        String ra = "[A-Z&&[0-9A-Z]]";
        String rb = "[A-Z&&0-9A-Z]";
        String rc = "[A-Z&&0-9[A-Z]]";
        
        String rx = "[A-Z&&[A-Z]0-9]";
        String ry = "[A-Z&&[A-F][G-Z]0-9]";
        //--------------------------------------------
        
        // Structure verification via reflection
        
        // COMPARISON CASES
        // These 3 character classes work as expected.
        dissectPattern(Pattern.compile(ra));
        dissectPattern(Pattern.compile(rb));
        dissectPattern(Pattern.compile(rc));
        
        // EXPECTED: The regex below should be equivalent to the 3 regex in COMPARISON CASES
        // ACTUAL: The nested character class [A-Z] is lost during compilation. This regex can't match anything.
        dissectPattern(Pattern.compile(rx));
        
        // EXPECTED: The regex below should be equivalent to the 3 regex in COMPARISON CASES
        // ACTUAL: The nested character class [A-F] and [G-Z] is lost during compilation. This regex can't match anything.
        dissectPattern(Pattern.compile(ry));
        
        //---------------------------------------------
        // Normal blackbox testing
        
        // ACTUAL=EXPECTED: true
        System.out.println("A".matches(ra));
        System.out.println("A".matches(rb));
        System.out.println("A".matches(rc));
        
        // EXPECTED: true
        // ACTUAL: false
        System.out.println("A".matches(rx));
        
        // EXPECTED: true
        // ACTUAL: false
        System.out.println("A".matches(ry));
    }
}
---------- END SOURCE ----------

CUSTOMER SUBMITTED WORKAROUND :
* To fix this bug:

In clazz() method of Pattern class

                           } else { // abc&&def
                                unread();
                                if (rightNode == null) { //L
                                    rightNode = clazz(false);
                                } else { //L
                                    rightNode = union(rightNode, clazz(false)); //L
                                }//L
                            }

Lines ending with //L are lines to be added

* To workaround:

Enclose the whole thing inside a nested character class, instead of leaving some "flat" character class outside.