A High-Powered Version of Behavior-Driven Development

Transcription

A High-Powered Version of Behavior-Driven Development
A High-Powered Version of
Behavior-Driven Development
Outline
1.
2.
3.
4.
5.
6.
Introduction
Problem Statement
Proposed Solution
Target Audience
Related Work
User Scenarios
7.
7. Scenario Specification
8. Architecture
9. Our Contributions
10.Planned Evaluation
11.Future Work
12.Conclusion
Introduction
•
•
•
What is Test Driven Development (TDD)?
Problems with TDD?
o Desired "behavior" of an system is specified in terms
of code, not always the right way to capture behavior
o Other stakeholders in Software lifecycle find it hard
to be involved in the whole process
Introduction
•
•
What is Behaviour Driven Development (BDD)?
o BDD is TDD done right!!!
o BDD was conceived by Dan North in 2003
o BDD uses natural language to describe the "desired
behavior" of the system, that can be understood by
both the developer and the customer
Demo of an existing BDD application using Cucumber
which is a BDD framework predominantly used with
Ruby
Problem Statement
•
In existing BDD frameworks, developer needs to
manually write the glue code which maps "behaviors" to
the implementation code
•
This mapping involves writing code for interpreting
behaviors in terms of test-cases
•
Writing this glue code is tedious and error prone, if the
written test code does not depict the specified behavior
• Can this intermediate step of writing glue code be
automated?
Proposed Solution:
SmartBDD
•
•
•
Automatically generate test code from "behavior"
description, eliminating the intermediate step of writing
test code
This allows the direct execution of "behaviors" on
application code
SmartBDD makes BDD available to the developer
community, by reducing the learning curve, time and
effort
Target Audience
•
•
•
Developers, who want to practice BDD using a
simplified framework
Students, SmartBDD abstracts the process of magically
generating test code, making BDD more accessible, by
allowing students to follow a test-first approach towards
software development.
Stakeholders like customers who may have not followed
a test-first approach, can use the system to test
application code.
Related Work
•
•
•
There are very few published studies on BDD, most of
which take a relatively narrow view of BDD and only
treat it as a specific technique of software development
There are tools which use UML, XML schema etc. as
input and generate an template for the source code.
Various Frameworks that make BDD accessible
 JBehave
 Cucumber
 JDave
User Scenarios
SmartBDD UI - Eclipse Plugin
Scenario: specification
•
•
•
•
•
GIVEN, an initial context
WHEN, occurrence of an event
THEN, expected outcome
Parameters in Quotes, " "
Connectives - And and Not (only in then condition)
Scenario: example
Scenario 1: Account Deposit
Given a bank account with a balance of "100" $
When a customer deposits "20" $
Then the balance of the account should be "120" $
Architecture
Natural Language
Processing
•
Pre-processing:
o Parsing scenario text to computation object-oriented
class
o Syntax validation
Natural Language
Processing
•
Language Augmentation
o Prioritizing/weighting words: based on Part of
Speech (POS), e.g., verb, noun, adjective
o Filtering out unimportant words, e.g.,"a", "the", "is"
o Clarifying language meaning using:
 Synonyms
 Stemming
 Grammatical structure
o Using: openNLP, StanfordNLP, WordNet, Snowball
Probabilistic Matcher
• Assumptions: naming conventions, e.g., camel casing
• Scenarios - implementation coding mapping
o
o
o
o
Mapping unit: clause
Clause - class mapping
 Name matching
 POS (e.g.,nouns)
Clause - constructor mapping
 Name matching
 #parameters
 Given clauses
Clause - method mapping
 Name matching, POS, #parameters
Code Information Extractor
•
•
Is responsible for extracting information from the
implementation code at run-time for analysis about the
class that is loaded, to keep track of all the classes in a
project and the necessary details about them.
We built a wrapper over the Java Reflection library, to
extract Class, Constructor, Method, Variables and
Parameter information. It is also used to infer the return
type and number of arguments for a method.
Code Generation
•
Code generator interacts with all the above
components, integrates the information from
augmentation engine and information extractor and
uses probabilistic values from the matcher Engine, to
generate the code on-the-fly.
•
The code is generated using a sophisticated on-the-fly
generator using the CodeModel library.
Code Generation:
Natural Language Interpretation
•
The code generator makes certain assumptions about
the interpretation of the natural language.
•
Given condition specifies object creation. You should be
able to create multiple objects as required by the
scenario.
•
The When condition specifies an action which is a call
to a particular method.
•
The postcondition i.e. Then, contains a sort of assertion
to verify whether the case failed or passed.
Demo Time!!!
•
•
HTTP Requester example with BDD approach
Multiple classes in same scenario - Stack and Queue
example
Contributions
•
• Make BDD more accessible to developers, specifically
for students by automating the whole process of testcode generation.
•
Creating loosely coupled components, which will help in
porting SmartBDD to other languages.
•
Unique Idea about probabilistic matching, and how a
good model will provide insight into code generation
ability.
•
Simplified UI for development.
Planned Evaluation
•
•
•
Make SmartBDD open-source and try to involve
developers to use the framework and provide feedback.
Deploy SmartBDD in a CS1 course for Java in Fall
2013.
Things to evaluate
o Compare with other BDD tools and evaluate if our
tool is making a difference based on student
feedback.
o Does the code generated perform the behavior that
the student expects?
o Is the tool generally useful? Does it reduce
complexity and developer time? Other Metrics.
Future Work
•
•
•
Lots of things still to do!!!
NLP Augmentation engine:
o Our grammar, still requires parameters to be passed
in quotes, we want to infer this information from
scenarios.
o Make inference about order of parameters that are
passed.
o Augment with root words and synonyms.
o Make as much inference as possible.
Code information extractor
o extract details from comments, local variables etc.
Future Work
•
•
•
Probabilistic matcher
o tune it for better results with more accurate
probability value.
o How to handle same probability value.
Code generation
o Passing objects
o Referring return values from previous method calls.
o Better assertion capability.
o Provide conditional statements, looping constructs.
o Acceptance testing involving interaction between
multiple classes.
UI Improvements, usability and effectiveness.
Conclusion
•
SmartBDD enables automatic code generation from
"behaviors"
•
SmartBDD makes BDD accessible and hopefully easy
for students to use.
•
Though it may not be industry ready, this tool should fit
the need for students in CS1 courses to help them
follow sound software engineering practices, by using
BDD to unit test their code.
Acknowledgements
•
•
•
We would like to thank,
Dr. Stephen H. Edwards
 Idea
 Insightful and
Timely Feedback
All students in CS-6604 course
who provided us with
useful feedback,
and had to bear with us!
Questions?
Thank You!!!