Lecture 4

Lecture 4

Process and Method:

An Introduction to the

Rational Unified Process

Traditional Structured Analysis

Described by W. W. Royce, 1970, IEEE WESCON, Managing the development of large software systems.

Decomposition in terms of Function and Data

Modularity available only at the file level

– cf. C language's static keyword (=="file scope")

Data was not encapsulated:

– Global Scope

– File Scope

– Function Scope (automatic, local)

Waterfall Method of Analysis and Design

Waterfall Method

Requirements Analysis– Analysis Specification

• Design Specification– Coding from Design Specification

» Unit Testing» System Testing» UAT Testing» Ship It (????)

Measuring rod is in the form of formal documents (specifications).

Waterfall Process Assumptions Requirements are known up front before design

Requirements rarely change

Users know what they want, and rarely need visualization

Design can be conducted in a pure abstract space, or trial rarely leads to error

The technology will all fit nicely into place when the time comes (the apocalypse)

The system is not so complex. (Drawings are for wimps)

Structured Analysis Problems Reuse is complicated because Data is strewn throughout

many different functions

– Reuse is usually defined as code reuse and is implemented through cutting and pasting of the same code in multiple places. What happens when the logic changes?

• coding changes need to be made in several different places

• changing the function often changes the API which breaks other functions dependent upon that API

• data type changes need to be made each time they are used throughout the application

Waterfall Process Limitations

Big Bang Delivery Theory

The proof of the concept is relegated to the very end of a long singular cycle. Before final integration, only documents have been produced.

Late deployment hides many lurking risks:

– technological (well, I thought they would work together...)

– conceptual (well, I thought that's what they wanted...)

– personnel (took so long, half the team left)

– User doesn't get to see anything real until the very end.

– System Testing doesn't get involved until later in the process.

The Rational Unified Process RUP is a method of managing OO Software Development It can be viewed as a Software Development Framework

which is extensible and features:

– Iterative Development

– Requirements Management

– Component-Based Architectural Vision

– Visual Modeling of Systems

– Quality Management

– Change Control Management

RUP Features

Online Repository of Process Information and Description in HTML format

Templates for all major artifacts, including:– RequisitePro templates (requirements tracking)– Word Templates for Use Cases– Project Templates for Project Management

Process Manuals describing key processes

The Phases

An Iterative Development Process... Recognizes the reality of changing requirements

– Capers Jones’s research on 8000 projects

• 40% of final requirements arrived after the analysis phase, after development had already begun

Promotes early risk mitigation, by breaking down the system into mini-projects and focusing on the riskier elements first

Allows you to “plan a little, design a little, and code a little” Encourages all participants, including testers, integrators, and

technical writers to be involved earlier on Allows the process itself to modulate with each iteration, allowing you

to correct errors sooner and put into practice lessons learned in the prior iteration

Focuses on component architectures, not final big bang deployments

An Incremental Development Process...

Allows for software to evolve, not be produced in one huge effort

Allows software to improve, by giving enough time to the evolutionary process itself

Forces attention on stability, for only a stable foundation can support multiple additions

Allows the system (a small subset of it) to actually run much sooner than with other processes

Allows interim progress to continue through the stubbing of functionality

Allows for the management of risk, by exposing problems earlier on in the development process

Goals and Features of Each Iteration

The primary goal of each iteration is to slowly chip away at the risk facing the project, namely:

– performance risks

– integration risks (different vendors, tools, etc.)

– conceptual risks (ferret out analysis and design flaws) Perform a “miniwaterfall” project that ends with a delivery

of something tangible in code, available for scrutiny by the interested parties, which produces validation or correctives

Each iteration is risk-driven The result of a single iteration is an increment--an

incremental improvement of the system, yielding an evolutionary approach

Risk Management

Identification of the risks Iterative/Incremental Development The prototype or pilot project

– Booch’s “Tiger Team”

Early testing and deployment as opposed to late testing in traditional methods

The Development Phases

Inception Phase Elaboration Phase Construction Phase Transition Phase

Inception Phase Overriding goal is obtaining buy-in from all interested

parties Initial requirements capture Cost Benefit Analysis Initial Risk Analysis Project scope definition Defining a candidate architecture Development of a disposable prototype Initial Use Case Model (10% - 20% complete) First pass at a Domain Model

Elaboration Phase Requirements Analysis and Capture

– Use Case Analysis• Use Case (80% written and reviewed by end of phase)

• Use Case Model (80% done)

• Scenarios– Sequence and Collaboration Diagrams

– Class, Activity, Component, State Diagrams

– Glossary (so users and developers can speak common vocabulary)

– Domain Model • to understand the problem: the system’s requirements as they exist

within the context of the problem domain

– Risk Assessment Plan revised

– Architecture Document

Construction Phase Focus is on implementation of the design:

– cumulative increase in functionality

– greater depth of implementation (stubs fleshed out)

– greater stability begins to appear

– implement all details, not only those of central architectural value

– analysis continues, but design and coding predominate

Transition Phase

The transition phase consists of the transfer of the system to the user community

It includes manufacturing, shipping, installation, training, technical support and maintenance

Development team begins to shrink Control is moved to maintenance team Alpha, Beta, and final releases Software updates Integration with existing systems (legacy, existing

versions, etc.)

Elaboration Phase in Detail

Use Case Analysis

– Find and understand 80% of architecturally significant use cases and actors

– Prototype User Interfaces

– Prioritize Use Cases within the Use Case Model

– Detail the architecturally significant Use Cases (write and review them)

Prepare Domain Model of architecturally significant classes, and identify their responsibilities and central interfaces (View of Participating Classes)

Introduction to XP

“When the tests all run, you’re done”

Options

XP is designed around the concept of options– Option to abandon– Option to switch– Option to defer– Option to grow and learn

The Four Variables Management or the Customer chooses 3 of the four variables, the

development team defines the fourth. Cost

– Cost is the amount of capital available, which defines resources. More resources don’t necessarily mean better quality or shorter time (remember Brooks?)

Time– The amount of time available for the project through delivery

Quality– Quality is the degree to which and aplomb with which functionality meets

requirements Scope

– Scope is the amount of work to be done, the totality of the set of requirements. As requirements come and go, scope vacillates.

The Paradigm Shift XP is based on the rejection of a fundamental and long-

standing principle, that it costs less to make changes earlier in the development cycle rather than later. That the graph of cost to change is exponential across time. This fundamental principle has led to several strategies:

– Better safe than sorry

– Functional extravagance

– Design extravagance

– Proliferation of activities that may never provide a return on the investment

The Paradigm Shift Continued The fundamental technical premise of XP is that the graph of cost to

change is not exponential but digressive, and as time goes by, the cost to change is asymptotic. “You make the big decisions as late in the process as possible.” This has several strategies:

– You implement only what you have to, and add functionality later only if necessary

– Design is parsimonious

– Thoreau’s principle: Simplify, Simplify, Simplify.

– Automated tests

– Refactoring

– Learning to drive analogy

– informality

The Four Values Communication

– Communication is bipartite. Developers need to communicate with customers as well as between themselves

Simplicity– “What’s the simplest thing that could possibly work?” Let’s do that.

Feedback– Continuous and instant feedback to all artifacts

– Continuous and instant feedback to the project progression

– Continuous and instant feedback to code Courage

– The courage to change (alter design, throw away code)

– The courage to decide

– The courage to do

– The courage to be

The Basic Principles of XP Rapid feedback

– instant evaluation of all work and deliverables Assume simplicity

– 98% of problems can be solved with “ridiculous simplicity”– What happened to complexity?

• Complexity != complex solutions Incremental change

– Avoid big changes, make smaller changes more often (driving analogy) Embracing change

– Might as well. Heraclitus was right, Parmenides was wrong. You simply will not be stepping into the same river twice.

Quality work– Work ethic– Is Beck a little too hopeful on the human condition?

Subordinate Principles Teach learning Small initial investment Play to win Concrete experiments Open, honest communication Work with people’s instincts, not against them Accepted not foisted responsibility Local adaptation (of process) Travel light (the nomadic team) Honest measurement (no lying)

The Four Basic Activities

Coding Testing Listening Designing

Dominance of Coding and Testing Code is unambiguous and constant. It offers no opinions. Code is a another language for communication (as in pair

programming) Tests allow for a secondary view into the code, from another angle Tests verify that “what was meant” was actually implemented Tests can validate performance as well as functionality You are responsible for writing multiple unit tests, you write a simple

test for every possible way to “break” your code. Automated tests can prolong the longevity of the code, and provide

continuous validation. A testing mentality promotes more self-assured programming style, as

successful tests yield confidence in the code.

The Practices Planning – quickly determine the scope of the next iteration.

Customers do the planning based on feedback from the developers.– “Software development is always an evolving dialog between the possible

and the desirable.” Small Releases – take baby steps in each iteration. Rank iterations

according to those which deliver the most valuable business requirements.

Metaphor – define a simple story of how the system will work. It should be enlightening.

Simple design – few classes and methods, no duplicated logic Testing – Developers write unit tests, Customers write functional tests Refactoring – revisiting code with rules that simplify the code. “When

the system requires that you duplicate code, it’s asking for refactoring.”

Pair Programming Collective Ownership – anyone can change any code at any time.

The Practices, cont.

Continuous Integration – code is integrated every half or full day at most. Integration is putting new code with the current system.

Sane work week On-site customer – customer needs to be around Coding standards that all coders follow

Pair Programming One programmer writes the code, at the low level. He/she “has the

ball”, or at least the keyboard. The other programmer looks at the code being written from a higher

strategic level:– What additional tests could break this?

– Can this be done more simply? (designing)

– Have I seen this before? (Refactoring)

– Did the guy with the ball just introduce a bug?

– Is this the best approach to this problem?

– Did the guy with the ball forget something?

– Does a question need to be answered by the Customer? Coding standards help reduce the need for reformatting code and

bickering about style. Pairs write tests together too, following the same principles.

“Problems” With Pair Programming

What happens on a geographically distributed development team?

Management will object to “waste”, you only get half as much done, or we’ll need twice as many programmers.

Pairs will naturally “self-select” in a Darwinian sense, militating against teaching learning.

Project Planning

Three phases:– Exploration– Commitment– Steering

Exploration Phase

Write a story (think “simplified” Use Case) Estimate a story: how long will it take to

code this? Split a story: if a part of a story is more

important than another, split it into two stories

Commitment Phase

Business chooses the scope and delivery date of the next iteration

Four movements:– Sort by value (must have, should have, nice to have)

– Sort by (estimation) risk

– Set velocity – how quickly do we expect to move on this?

– Choose Scope – Ok, given the above, what are we to deliver and when is it due?

Steering Phase Four movements:

– Iteration• Iterations run 1 to 3 weeks generally. • Each iteration selects one or more stories to

implement. Each iteration must yield a system that runs end-to-end, however embryonically.

– Recovery: if development has overstated velocity, re-evaluate the set of stories (deliverables)

– New story: If business realizes it’s got a new story, the new story is estimated, ranked, and added.

– Reestimate: If development feels the plan is inadequate, it can reestimate the remaining stories and reset the estimated velocity.

Iteration Planning Task planning Three Phases:

– Exploration Phase

• Write a task by breaking down the stories into tasks

• Split a task if necessary

– Commitment Phase

• Accept a task

• Estimate a task

– Steering Phase

• Implement a task

• Record Progress

• Recovery – what to do if overworked: manage scope

• Verify story with functional tests

What about Design Strategy?

Start with a test. A simple test. Design and implement just enough to get

that test running, and make sure you don’t break another test.

Add functionality and repeat Refactor. “The definition of the best design is the

simplest design that runs all the test cases.”

Use Case Analysis

What is a Use Case?

– A sequence of actions a system performs that yields a valuable result for a particular actor.

What is an Actor?

– A user or outside system that interacts with the system being designed in order to obtain some value from that interaction

Use Cases describe scenarios that describe the interaction between users of the system and the system itself.

Use Cases describe WHAT the system will do, but never HOW it will be done.

What’s in a Use Case? Define the start state and any preconditions that accompany it Define when the Use Case starts Define the order of activity in the Main Flow of Events Define any Alternative Flows of Events Define any Exceptional Flows of Events Define any Post Conditions and the end state Mention any design issues as an appendix Accompanying diagrams: State, Activity, Sequence Diagrams View of Participating Objects (relevant Analysis Model Classes) Logical View: A View of the Actors involved with this Use Case, and

any Use Cases used or extended by this Use Case

Use Cases Describe Function not Form

Use Cases describe WHAT the system will do, but never HOW it will be done. Use Cases are Analysis Products, not Design Products.

Use Cases Describe Function not Form

Use Cases describe WHAT the system should do, but never HOW it will be done

Use cases are Analysis products, not design products

Benefits of Use Cases Use cases are the primary vehicle for requirements capture in

RUP Use cases are described using the language of the customer

(language of the domain which is defined in the glossary) Use cases provide a contractual delivery process (RUP is Use

Case Driven) Use cases provide an easily-understood communication

mechanism When requirements are traced, they make it difficult for

requirements to fall through the cracks Use cases provide a concise summary of what the system should

do at an abstract (low modification cost) level.

Difficulties with Use Cases As functional decompositions, it is often difficult to make

the transition from functional description to object description to class design

Reuse at the class level can be hindered by each developer “taking a Use Case and running with it”. Since Ucs do not talk about classes, developers often wind up in a vacuum during object analysis, and can often wind up doing things their own way, making reuse difficult

Use Cases make stating non-functional requirements difficult (where do you say that X must execute at Y/sec?)

Testing functionality is straightforward, but unit testing the particular implementations and non-functional requirements is not obvious

Use Case Model Survey The Use Case Model Survey is to illustrate, in

graphical form, the universe of Use Cases that the system is contracted to deliver.

Each Use Case in the system appears in the Survey with a short description of its main function.– Participants:

• Domain Expert

• Architect

• Analyst/Designer (Use Case author)

• Testing Engineer

Sample Use Case Model Survey

Analysis Model In Analysis, we analyze and refine the requirements described in the

Use Cases in order to achieve a more precise view of the requirements, without being overwhelmed with the details

Again, the Analysis Model is still focusing on WHAT we’re going to do, not HOW we’re going to do it (Design Model). But what we’re going to do is drawn from the point of view of the developer, not from the point of view of the customer

Whereas Use Cases are described in the language of the customer, the Analysis Model is described in the language of the developer:– Boundary Classes

– Entity Classes

– Control Classes

Why spend time on the Analysis Model, why not just “face the cliff”?

By performing analysis, designers can inexpensively come to a better understanding of the requirements of the system

By providing such an abstract overview, newcomers can understand the overall architecture of the system efficiently, from a ‘bird’s eye view’, without having to get bogged down with implementation details.

The Analysis Model is a simple abstraction of what the system is going to do from the point of view of the developers. By “speaking the developer’s language”, comprehension is improved and by abstracting, simplicity is achieved

Nevertheless, the cost of maintaining the AM through construction is weighed against the value of having it all along.

Boundary Classes Boundary classes are used in the Analysis Model to model interactions

between the system and its actors (users or external systems) Boundary classes are often implemented in some GUI format (dialogs,

widgets, beans, etc.) Boundary classes can often be abstractions of external APIs (in the

case of an external system actor) Every boundary class must be associated with at least one actor:

Entity Classes

Entity classes are used within the Analysis Model to model persistent information

Often, entity classes are created from objects within the business object model or domain model

Control Classes The Great Et Cetera Control classes model abstractions that coordinate, sequence, transact,

and otherwise control other objects In Smalltalk MVC mechanism, these are controllers Control classes are often encapsulated interactions between other

objects, as they handle and coordinate actions and control flows.