Precisely Selecting Regression Test for AspectOriented Programs

1
Precisely Selecting Regression Test for
Aspect-Oriented Programs

• Guoqing Xu
• The Ohio State University
• xug_at_cse.ohio-state.edu

2
Outline

• Big picture and background
• Problem statement and motivation
• Our analysis
• Implementation status
• Related work

3
Big Picture

• AOSD improves separation of concerns
• Challenges for analysis of AO programs
• Control flow in bytecode is not equivalent to the
  program logic in source code
• Compiler-specific weaving
• Dynamic/static parts of pointcut designators
• No analysis framework implementation available
• Challenges for testing of AO programs
• Coverage criteria test generation for exercising
  of aspects regression test selection

4
Big Picture

• Long-term goals
• Framework that supports static analyses similar
  to popular analyses for OO program
• Develop AO-specific techniques for analysis and
  testing
• Topic of this talk
• Regression test selection for AO programs
• Work in progress

5
Regression Test Selection

• Testing after software modifications
• Re-running the entire regression test suite is
  expensive
• Select a subset of tests to run
• Safe test selection chooses every test case that
  may reveal a fault

6
Related Previous Work

• Rothermel and Harrold, TOSEM 97
• Graph traversal algorithms
• Harrold et al., OOPSLA 01
• Java interclass graph (JIG)

Program P
Execute P/ Record Coverage
Program P
Ps Edge Coverage Matrix
Select Tests
Dangerous Entities containing edges in P
Identify Dangerous Entities
Program P
Program P
Copied from Harrold et al., OOPSLA 01
7
Regression Testing for AOSD

• When AO features are added/modified, the program
  needs to be regression tested
• Case 1 P is an OO program and P is an AO
  version of P
• Case 2 Both P and P are AO programs
• How should regression test selection be performed
  for AO software?

8
Outline

• Big picture and background
• Problem statement and motivation
• Our analysis
• Implementation status
• Related work

9
Existing Work Applied to AO Programs

• How to compare two JIGs?
• The JIG of the woven code includes redundant
  nodes and edges and does not correspond to the
  logical control flow as presented in the source
• Need new representations
• How to recover CFG edges from the execution trace
  when computing edge-coverage matrix? (when P is
  AO program)
• The execution trace is compiler-specific
• Need instrumentation before/during weaving

10
Existing Work Applied to AO Programs

• class C
• public void m(int i)
• 3.
• void bar(C p, int k)
• p.m(k)
• Java version P
• CFG edge
• Call edge

11
Example

• class C
• public void m(int i)
• 3.
• void bar(C p, int k)
• p.m(k)
• aspect Sample
• void around(C c, int i)
• call(C.m(int))
• target(c)
• args(i)
• 14. proceed(p , x)
• 15.

AspectJ Version P
12
Some Results

• When we applied the Harrold et.al. 01 algorithm
  to several subjects

13
Possible Approach

• Create clean CFGs in which the wrapper code
  inserted during weaving is removed
• Graph traversal and comparison corresponds to the
  logical structure of the code, not the
  compiler-specific woven code
• New representation AJIG
• AspectJ Inter-module Graph more later
• For regression test selection, need to consider
  additional issues

14
A More Complex Problem

• class C
• public void m(int i)
• 3.
• void bar(C p, int k)
• C.m(k)
• aspect Sample
• void around(C c, int i)
• call(C.m(int))
• target(c)
• args(i)
• System.out.println(i)
• proceed(p , x)
• 15.

15
Why it needs to be addressed

• This is an issue not only for AO software, but
  also for procedural and OO software
• Advices are often free of side effects
• Study in Rinard et al. FSE 04 reported 6
  observer advices out of ten inspected advices
• Recommended for safe AO programming
• Adding side-effect-free advices should not result
  in overly conservative regression test selection
• Approach use side-effect analysis

16
Outline

• Big picture and background
• Problem statement and motivation
• Our analysis
• Implementation status
• Related work

17
Our Work

• Consider both situations
• Case 1 P is an OO version, P is an AO version
• Case 2 both P and P are AO versions
• Analysis to select regression tests
• Build a new control flow representation AJIG
• Apply existing graph-traversal algorithm on AJIG
• Side-effect analysis when comparing AJIGs

18
AJIG

• AspectJ Inter-module Graph (recent work)
• For the Java parts, same as JIG.
• Shadow node
• A shadow node is associated with
• a set of JIGs of advices
• the precedence of these advices
• an integrated shadow advice JIG
• AJIG supports all static AspectJ join point types
• Conservatively approximate the dynamic part of
  pointcut designators.

19
Side Effect

• Side effect node
• Mutate the objects that have existed before the
  method/advice is invoked
• For an advice,
• All the nodes in the CFG path that does not
  contain a proceed node
• The proceed node has side effect if its actual
  parameters and parameters of the advice dont
  point to same objects.
• Mutates the object returned by the proceed node
• The return node has side effect if the return
  value of the advice and that of the proceed node
  dont point to the same object

20
Side Effect (Contd)

• Side effect related node in AJIG
• Has side effects
• Has some dependency on the nodes that have side
  effects
• Safe edge in AJIG
• An edge is safe edge, if the sink node of this
  edge is not side effect related.

21
New Test Selection criterion

• New Test Selection criterion
• Dangerous set S computation for each edge e in
  P, and its counter part e in P
• e is not equivalent to e and
• Neither e nor e is a safe edge.
• Select a test that execute one or more edges in S

22
Selection

• Computing dangerous set S by comparing AJIGs.
• We plan to use some form of side-effect analysis
• Large body of existing work
• Selecting tests

23
Implementation progress

• The implementation of algorithms described in
  Harrold et.al. 01 v
• Building AspectJ Inter-module Graph v
• Make an extension for the abc compiler that
  generates the Jimple based CFGs for aspects
  between the weaving of inter-type declarations
  and advices.
• Instrument advices at different phases.
• Side effect analysis ongoing work
• Evaluation

24
Evaluation Plan

• Benchmarks ten to twenty moderate size (50-100
  classes) AO projects taken from tmbenches, used
  by the abc compiler.
• Experiments
• collect data.
• The number of tests selected over versions.
• The comparison between enabling side-effect
  analysis and disabling it.

25
Related Work

• Static/Dynamic Analysis for AO programs.
• abc compiler AOSD 05 PLDI 05 TR 04.
• Static analysis of aspects Sereni and Moor, AOSD
  02.
• Zhaos work on the analysis and testing of AO
  programs COMPSAC 03 WPC 02AOSD 06.
• Classification system for AO programs Rinard et.
  al. FSE 04.

26
Related Work (Contd)

• Regression Test Selection
• LW ICSM 91, CRV ISCE 94, RH TOSEM 97, Ball
  ISSTA 98, Harrold OOPSLA 01, OSH FSE 04
• Change Impact Analysis
• KGH ICSM 94, RT PASTE 01, OAH FSE 03,
  OAL ICSE 04, RST OOPSLA 04,

27

• Thank you!!
• Questions??