WALA Everywhere

Julian Dolby

IBM Thomas J. Watson Research Center

SOAP, PLDI, Barcelona, June 2017

Dramatis personæ

Analyze Everything

Analyze Everywhere

WALA Everywhere

• Illustrate a small sample of being everywhere

  • hybrid apps

  • WALA on Web browsers

  • Swift

Hybrid Apps

Hybrid Apps

• Promote portability of apps
  • “write once, run everywhere”
  • minimize cost of app across multiple platforms
• Different semantics complicates programming
• Static vs. dynamic types
  • cannot check all type errors syntactically
  • no overloading in JavaScript
• Argument count flexible in JavaScript
  • try to pass wrong number of arguments to Java

Syntax Issues

HybridDroid

• Soundy analysis framework for Android hybrid apps

  • support for most implicit inter-language flows (Backed by APIs and Dalvik VM source code)

  • support most type compatibility in browsers (backed by experiments with trials & errors)

• Implementation on top of WALA

• Uses generic WALA cross-language support

Syntax Checking Results

Why run WALA on the Client?

• Integrate with Rich Client interfaces
  • simplify integration
  • avoid network round trips
  • use client compute resources
  • already in prototype application inside IBM
• Client-side applications
  • analyze dynamically-loaded scripts
  • support IDE's like Eclipse Orion
• Browsers are everywhere

How to Run WALA on the Client?

• TeaVM compiles Java to JavaScript
  • compiles JVM bytecode to JavaScript
  • reachability analysis to choose code to compile
  • Java 8, including lambdas
• Significant but limited library support
  • collections and key libraries implemented
  • I/O libraries very limited
  • little reflection support

WALA on TeaVM

• Core of WALA compiles with TeaVM
  • avoid esoteric features like soft references
  • avoid the file system
  • avoid any use of reflection
• Basic program analysis works
  • read JavaScript
  • call graphs
  • system dependence graphs
• Enough for flow-sensitive taint analysis

WALA Client Demo

function Document_prototype_write(x) {

}

function id(x) {
    return x // line 6;
}

var document = { URL: "whatever" };
var url = id(document.URL); // line 10
Document_prototype_write(url); // line 11

var notUrl = id("not a url");
Document_prototype_write(notUrl);

WALA Client Demo

WALA for Swift

• Apple a dominant mobile platform

  • likely most popular phone in the audience

  • Swift primary programming language

• WALA meant to be flexible

  • analyzing new language Swift a test

  • support Apple-IBM alliance

• Use WALA to analyze Swift

  • this work in early stage

Leverage Apple Infrastructure

• Use open source Apple code

  • up-to-date with evolving language

  • written largely in C++, unlike WALA

• Bridge C++ code to WALA

  • large code base, so explicit JNI tedious

  • generate wrappers with SWIG

• Expose Apple Swift code to Java

SWIG

%module example     
int fact();      

public class example {

  public static int fact() {
    return exampleJNI.fact();   
  }

}

public class exampleJNI {
  static native int fact();
}

...Java_exampleJNI_fact(...) {
  jint jresult = 0 ;
  int result;
  
  (void)jenv;
  (void)jcls;
  result = (int)fact();
  jresult = (jint)result; 
  return jresult;
}

• SWIG generates “normal” Java code around JNI

WALA using SWIG

 instance.performParseOnly();
 if (instance.getASTContext().hadError())
     System.out.println("Parse error");
 ModuleDecl module = instance.getMainModule();
 ASTWalker walker = new ASTWalker() {
     @Override
     public boolean walkToDeclPre(Decl s) {
         // visited a declaration
         return true;
     }
     // ...
 };
 module.walk(walker);

• Java code calling Apple C++ code
  • resembles WALA JavaScript Rhino code
  • ugliness hidden by SWIG

Status

• Beginnings of SWIG interface working

  • SWIG integrated into Apple build process

  • both AST and SIL partially wrapped

• Interface not just for WALA

  • other Java tools could benefit

  • would love collaborators

WALA Everywhere

• World getting less monolingual, not more

  • domains have favorites, e.g. Python for ML

  • new languages like Swift arise

• Analysis frameworks must adapt

  • need to analyze code people use

  • WALA has proven flexible