cos / walafacade   0.1.2

Eclipse Public License 1.0 GitHub

facade for using WALA from Scala

Scala versions: 2.11 2.10

WALAFacade

WALAFacade is a facade for making WALA easier to use from Scala. It is mostly syntactic sugar over the WALA API to make its use more concise. It allows bypassing boilerplate code required because of Java's less flexible type system.

The implementation is working well but it is not, by far, comprehensive. Features have been added when needed. The discrete WALAFacade project appeared when we needed to share code between projects. We intent to evolve it in the same demand-driven manner from now on. If you cannot find something you need, feel free to add it, or post an feature request issue.

Features

  • short aliases for commonly used types (e.g., PutI instead of SSAPutInstruction)

  • richer API for many WALA types (e.g., get all instructions puting to a LocalP(LocalPointerKey): P(cgNode, ssaValue).puts). Rich types are implemented as Scala value classes so the performance impact should be negligible

  • alternative CallGraphBuilder and related classes for fast bootstrapping (integrated with typesafe/config)

  • implicit two-way conversion between WALA types and rich types or Scala alternatives (e.g., can use a Function1[T, Boolean] instead of a Predicate[T])

  • uses implicit parameters to piggyback outer information (e.g., can find out the IField(i.e., F) for a PutI by putI.f while having an implicit IClassHierarchy in scope)

  • wraps primitives in value classes for better type safety (e.g., i.uses returns an Iterable[V] where V is a values class that wraps an Int that has SSA value number semantics)

Design principles:

  • keep the facade API close to the original API. A fair amount of thought has been put into the WALA API so there is no point in moving away from it when the gain is not significant.

  • favor usability, but keep performance in mind

Getting started

If you use sbt, the easiest way to get the library is from maven central:

libraryDependencies += "edu.illinois.wala" %% "walafacade" % "0.2"

Still, since the library is young you will likely want to get the code so you can modify it easily instead of using it as an external dependency.

There are also ready-made example projects you can simply clone:

Steps

Clone WALA (optional)

This step is not required anymore as WALA is available from Maven Central now. Still, if you would like to also hack on the WALA source code, clone WALA and install it to your local Maven repo:

  • export JAVA_HOME="<java's home on your system>" (on OS X: /usr/libexec/java_home) Also, you might want to add this line to your system loading profile
  • git clone https://github.com/wala/WALA.git (or use my fork https://github.com/cos/wala/tree/for-iterace instead. It makes dispatch more precise. See the discussion about cilib results in the ISSTA paper)
  • cd WALA
  • mvn clean install -DskipTests=true
Clone WALAFacade
  • git clone https://github.com/cos/WALAFacade.git
  • cd WALAFacade
  • sbt publishLocal
Generate an Eclipse project or import into IntelliJ

To generate an Eclipse project:

  1. Generate an .project file by running sbt eclipse in the WALAFacade directory.
  2. Import the project into an Eclipse. WALA dependencies will be linked from the Ivy repository.

Alternatively, import into IntelliJ as an sbt project.

Basic usage

Configuring the analysis

This guide provides a good introduction to setting up WALA: http://wala.sourceforge.net/wiki/index.php/UserGuide:Getting_Started

Alternatively, WALAFacade allows you to use a typesafe/config file instead of wala.properties. Steps:

  1. Create a file called application.conf (or alternatives) and put it on the project's classpath.
  2. Load an implicit config in your scope Config conf = ConfigFactory.load()
  3. val pa = FlexibleCallGraphBuilder() - the pointer analysis will run on instantiation
  4. Use the results. E.g., pa.cg is the call graph, pa.heap is the heap graph

The analysis can be customized by overriding the following default methods of the FlexibleCallGraph class:

def policy = { import ZeroXInstanceKeys._;  ALLOCATIONS }
def cs = new ContextInsensitiveSelector()
def contextInterpreter: RTAContextInterpreter
def contextInterpreter = new DefaultSSAInterpreter(...)
def instanceKeys = new ZeroXInstanceKeys(...)

Example

A basic config file - needs to be in your classpath. No other configuration necessary.

wala {
  jre-lib-path = "/Library/Java/JavaVirtualMachines/jdk1.8.0_45.jdk/Contents/Home/jre/lib/rt.jar"
  dependencies.binary += "target/scala-2.11/classes"
  dependencies.jar += "andthis.jar"
  exclusions += ".*BesidesDefaultsIgnoreAlsoThisPattern.*"
  entry {
    signature-pattern = ".*Foo.*main.*"
  }
}

And a program that finds in all the code reachable from methods named bar all written fields and the names of the variables written to each.

// remember that 
// type N = CGNode // call graph nodes
// type PutI = SSAPutInstruction
// type LocalP = LocalPointerKey

import com.ibm.wala.ipa.callgraph.impl.ContextInsensitiveSelector
import com.ibm.wala.ipa.callgraph.propagation.cfa.nCFAContextSelector
import com.ibm.wala.util.graph.traverse.DFS

import edu.illinois.wala.ipa.callgraph.FlexibleCallGraphBuilder
import edu.illinois.wala.ipa.callgraph.propagation.P

import com.typesafe.config.ConfigFactory

// convenience object that activates all implicit converters
import edu.illinois.wala.Facade._

import scala.collection.JavaConversions._

object Test extends App {
  implicit val config = ConfigFactory.load()

  // creates a new pointer analysis with a special context selector
  // implicitly uses the above config file
  val pa = new FlexibleCallGraphBuilder() {
    override def cs = new nCFAContextSelector(2, new ContextInsensitiveSelector());
  }

  // make cg, heap, etc. available in scope

  import pa._

  // more verbose to each understanding
  val startNodes = cg filter { n: N => n.m.name == "bar" }
  val reachableNodes = DFS.getReachableNodes(cg, startNodes)
  val foo = reachableNodes flatMap { n =>
    n.instructions collect {
      case i: PutI =>
        val p: LocalP = P(n, i.v)
        val variableNames: Iterable[String] = p.variableNames()
        val fieldName: F = i.f.get
        (fieldName, variableNames)
    }
  }

  // and a 3-liner doing exactly the name thing
  DFS
    .getReachableNodes(cg, cg filter { _.m.name == "bar" })
    .flatMap { n => n.instructions collect { case i: PutI => (i.f.get, P(n, i.v).variableNames()) } }
}

Finding your way around

The code has little comments and documentation but much of it should be self-explenatory.

Conventions

  • the package structure mirrors WALA's. E.g., classes wrapping/pertaining to com.ibm.wala.ssa are found in edu.illinois.wala.ssa

  • many packages contain a trait called Wrapper which does most of the implicit magic. The Wrappers are arranged in a tree structure, with each package's Wrapper extending all the Wrappers of its subpackages.

  • the wrapper for the outermost package, i.e. edu.illinois.wala, is actually called Facade and imports all other Wrappers

  • all type aliases are in edu.illinois.wala/TypeAliases

Related projects

cuplv/walautil is a collection of WALA utility functions for Scala projects you may want to check out.