1. Lori Pollock Professor, CIS Program Analysis, Software Development & Maintenance Tools, Optimizing Compilers
‘81 B.S. CS and Econ, Allegheny
’81-’86 PhD in CS, U of Pittsburgh
Married Mark
’86-’ Assistant Prof, Rice U
Lauren ‘88; Lindsay ‘90
’ Assistant, Associate, Full Prof UD CIS
Matt ’95
Today: Lauren & Lindsay here at UD, Matt 14, 3 PhD students and a few undergraduate researchers

2. What I do here at UD Research Graduate Teaching Undergraduates
Software Engineering and Compilation Lab (Hiperspace)
213 Smith Hall
Collaborations
Vijay Shanker (UD CIS), Terry Harvey (UD CIS), Lisa Marvel (Army Research Lab), Martin Swany (UD CIS), Guang Gao (UD ECE)
Funding
Primarily NSF grants; previously some Army funding
Graduate Teaching
CISC 672 Compilers
CISC 673 Program Analysis and Transformations
CISC 879 Software Testing and Maintenance
CISC 879 Software Tools and Environments
Undergraduates
Programming XO laptops for local middle school teachers
Study abroad programs

3. What I do outside UD
Associate Editor, Transactions on Software Engineering and Methodology (TOSEM)
Computing Research Association (CRA)’s Committee on the Status of Women in Computer Research (CRA-W)
Mentoring – speaker at mentoring
workshops for undergrads, grads,
assistant and associate profs, and
industry lab researchers
Program committees, conf org,
NSF panels, paper reviews,… (typical of university researchers)

4. PhD Students in Training
Antony Danalis
PhD
Emily Gibson Hill
PhD
Giri Sridhara
PhD
Masters: Divya Muppaneni
Current Undergraduates: Eric Enslen, Sana Malik, Katie Baldwin

5. Recently Completed PhD 2007-08
David Shepherd
Postdoc, Startup
Sara Sprenkle
Assistant Prof
Washington & Lee U
Mike Jochen
Assistant Prof
East Stroudsburg U
Ben Breech
Postdoc, Nasa

6. Overview: Research Projects
Natural Language
Analysis of Programs
Testing Web
Applications
Collaboration with Sara Sprenkle
Emily, Giri
Program Analysis
Compiler Technology
Ben
Optimization of
Cluster
Parallel Programs
Runtime Test
Generation via
Dynamic Compilers
Antony
Software Tools…………..Testing… Compilers….……Parallel Computing

7. Optimizing Cluster Parallel Programs
Research Problem - How can scientific codes be scaled
to a cluster of many CPUs?
Major Challenge – Communication Costs
Approach and Contributions:
An integrated system to hide communication latency
-Surveyor: Collect “knowledge” of cluster
-Compiler: analyze dependencies and transform to create maximal communication/computation overlap
-Communication Library: Use a companion library to MPI

8. ASPhALT: Automatic System for Parallel AppLication Transformations
Contribution: FIRST to cluster-optimize MPI codes

9. Testing Web Applications
Web Application Structure
Client
(HTML)
Server
(Java)
Database
(MySql)
Browser
Front End
Back End
Combination of
Stand-alone applications
GUIs and Database applications
Distributed applications
Numerous technologies and components

10. Traditional Software Testing Process
Hard to obtain when
testing web applications!!
Application
Representation
Implementation
Test Case Generator
User-session-based Testing
Replay Tool
Test
Cases
Application
Specification
Expected
Results
Test
Cases
Actual
Results
Oracle
Pass/ Fail

11. User-session-based Testing Process
register.jsp?name=ss&pass=tst
login.jsp?name=ss&pass=tst
logout.jsp
Users
User-session-based
User
Sessions
Beta Web Application (v.0.9) Deployment
Log User
Requests
Replay Tool
Test Cases
Create test
cases
Create/Reduce
Test Cases
Expected
Results
Test
Cases
Web Application
Implementation (v.1.0)
Actual
Results
Oracle
Pass/ Fail

12. Maintenance Testing for Web Applications
Research Problem: How can we exploit user session logging for testing of web applications after initial deployment, with minimal tester effort?
Contributions: Scalable, practical, automated structural testing framework for web applications
* Test case generation
* Test suite reduction
* Test oracles
* Test coverage criteria in terms of URLs, parameters, values

13. Analyzing the Names in Software
Research Problem
% software costs are in reading and navigating large software systems to fix bugs and add new features. Can we help with automation of search, navigation, location of relevant code?
- Key: Programmers leave clues of their intent as they choose names.
Focus on actions
-Correspond to verbs
-Verbs need Direct Object
- Phrases more useful
Proposed Approach
Develop, extend, and apply natural language-based analysis to the identifier names and comments
Contribution - Aid understanding, debugging, maintenance, development

14. Our Research Focus and Impact
Exploration
Understanding …
Software Maintenance Tools
NLPA
Help build these software maintenance tools
So green (motivation) comes up first, then (orange- what we are going to use), then NLPA.
Natural Language Analysis
Word relations
(synonyms,
antonyms, …
Part of speech
tagging
Abbreviations…
Word splitting

15. How do SE tool users search code now?
(Re)formulate Query
Determine Relevance of Results
User formulates query
Query executed by search method
User views search results
Repeat as necessary
Search
Results
Query
Search Method
Source
Code
Our focus: Natural language (NL) queries
Reason for NL vs RE
NL queries are more flexible than queries such as REs, and in previous work we’ve shown that keyword-style queries are superior for users who are new to a software system
note difference between developers and users (likely not original developer)
Iterative Query Refinement & Search Process
Our focus: NL queries (google example)
“compile report” vs. “compil*report”
User faces 2 challenges:
Decide what query words to search for
Determine whether the results are relevant

16. Our Contextual Search Process
Determine Relevance of Results
User
(Re)formulate Query
Search
Results
Hierarchical
Query
Search Method
Search Method
Source
Code
Information Extraction Process
Partial Phrase Matching
NL Phrase Mapping
Information Extraction Process
Source
Code

17. Another of our Tools: Dora the Program Explorer*
Query
Natural Language Query
Maintenance request
Expert knowledge
Query expansion
Program Structure
Representation
Current: call graph
Seed starting point
Dora
Before I give you the details of how we identify the relevant neighborhood, let me give you some background put the problem into context
Relevant Neighborhood
Subgraph relevant to query
Relevant Neighborhood
* Dora comes from exploradora, the Spanish word for a female explorer.

18. Illustrating some issues: Extracting Clues from Signatures
Split Name into Words
Part-of-speech tag method name
Chunk method name
Identify Verb and Direct-Object (DO)
public UserList getUserListFromFile( String path ) throws IOException {
try {
File tmpFile = new File( path );
return parseFile(tmpFile);
} catch( java.io.IOException e ) {
throw new IOrException( ”UserList format issue" + path + " file " + e ); }
Next up, Zak
POS Tag
get<verb> User<adj> List<noun> From <prep> File <noun>
Chunk
get<verb phrase> User List<noun phrase> From File <prep phrase>

19. Developing Basic NL Analyses
Search
Exploration
Understanding …
Software Maintenance Tools
NLPA
Help build these software maintenance tools
So green (motivation) comes up first, then (orange- what we are going to use), then NLPA.
Natural Language Analysis
Word relations
(synonyms,
antonyms, …
Part of speech
tagging
Abbreviations…
Word splitting

20. Automatic Abbreviation Expansion
Don’t want to miss relevant code with abbreviations
Given a code segment, identify character sequences that are short forms and determine long form
non-dictionary word
Split Identifiers:
Identify non-dictionary words
Determine long form
To identify character sequences, or tokens, in code boils down to splitting the identifiers. The hardest case is no boundary cases. If not properly split, abbreviations will be missed, such as in string length.
no boundary
Approach: Mine expansions from code [MSR 08]

21. Issues with a Simple Dictionary Approach
Manually create a lookup table of common abbreviations in code
Vocabulary evolves over time, must maintain table
Same abbreviation can have different expansions depending on domain AND context: ?
Control Flow Graph
Context-Free Grammar
configuration
configure
cfg
The simplest approach is a manually created dictionary of common short forms in code. However, this has a couple drawbacks.

22. What have we learned overall?
Evaluation studies indicate
Natural language analysis has far more potential to improve software maintenance tools than we initially believed
Existing technology falls short
Synonyms, collocations, morphology, word frequencies, part-of-speech tagging, AOIG
Keys to further success
Improve recall
Extract additional NL clues

23. What are we doing now? Emily Giri
Developing a general software word usage model
Implementing SWUM
Evaluating for search
Giri
Developing techniques for automatic comment generation
Which statements to include in summary content?
How to generate phrases for that content?
Providing feedback on SWUM refinement
Divya
Analyzing specific Java statement structures
Developing templates for comment phrase generation

24. Overview: Research Projects
Natural Language
Analysis of Programs
Testing Web
Applications
Collaboration with Sara Sprenkle
Emily, Giri
Program Analysis
Compiler Technology
Optimization of
Cluster
Parallel Programs
Antony
NEW…
Natural Language
Analysis of Parallel
Programs
Software Tools…………..Testing… Compilers….……Parallel Computing