DESIGN

Quick Summary

What you have learnt so far? Different kinds of software engineering methodology

Waterfall model Spiral model Unified process

Requirements :Use cases Traditional specification: semi-formal

DFD Formal specification: Z, FSM, Petric Nets

Object-oriented specification Class extraction, various diagram to present the class

relationships

Design with formal specification

Example give a Z specification Can you write program codes

for a “push_button” ?

Data and Actions Two aspects of a product

Actions that operate on data Data on which actions operate

The two basic ways of designing a product Operation-oriented design Data-oriented design

Third way Hybrid methods For example, object-oriented design that combines

features of operation-oriented design and data-oriented design

13.1 Design and Abstraction Classical design activities

Architectural design – general and high-level Detailed design Design testing

Architectural design – create modules of the software Input: Specifications document Output: Modular decomposition of the software

Detailed design Each module is designed

Specific algorithms and data structures are selected

13.2 Operation-Oriented Design If system can be presented in a DFD Data flow analysis (DFA)

Can use DFA with most specification methods (Structured Systems Analysis is presented here)

Key point: We have detailed action information from the DFD (ie input to and output from the product)

Figure 13.1

Data Flow Analysis Every product transforms input into output Determine

“Point of highest abstraction of input” “Point of highest abstract of output”

Figure 13.2

Data flow analysis Point of highest abstraction of input – point at

which the input loses the quality of being input and simply becomes internal data operated by the product

Example – input is a membership number the output could be a flag (valid/ not valid) and a discount %

Point of highest abstraction of output – point at which data can be identified as some form of output

Example – an array collecting all information ready to print out

Example

A software to solve circuit related problems

Input: V, I, R, Z etc At point of highest abstraction: a

matrix Output: the unknown V at t = 10ms Other examples based on your own

experiences?

Data Flow Analysis (contd)

Decompose the product into three modules Input, transform, and output

Repeat stepwise until each module performs a single operation: high cohesion Minor modifications may be needed to lower

the coupling Cohesion – degree of interaction within a

module Coupling – degree of interaction between

modules Cohesion - 結合 ; 凝聚 ; 團結力 ; 附著

Cohesion The degree of interaction within a module Seven categories or levels of cohesion (non-linear scale)

Figure 7.4

Coincidental Cohesion A module has coincidental cohesion if it performs multiple,

completely unrelated actions Example modules:

print_next_line, reverse_string_of_characters_comprising_second_parameter, add_7_to_fifth_parameter, convert_fourth_parameter_to_ floating_point

Such modules arise from rules like “Every module will consist of between 35 and 50

statements” so unrelated functions put together in order to fit the size!

Why Is Coincidental Cohesion So Bad?

It degrades maintainability – too many things to consider A module with coincidental cohesion is not reusable – as it is doing

too many things which are difficult to find other programs using all the same features

The problem is easy to fix Break the module into separate modules, each performing one

task

Logical Cohesion

A module has logical cohesion when it performs a series of related actions, one of which is selected by the calling module

Logical Cohesion (contd) Example 1:

function_code = 7;

new_operation (op code, dummy_1, dummy_2, dummy_3);

// dummy_1, dummy_2, and dummy_3 are dummy variables,// not used if function code is equal to 7

Example 2: An object performing all input and output

Example 3: A module that edits insertions, deletions, and modifications of

master file records

Why Is Logical Cohesion So Bad?

The interface is difficult to understand as there may be many irrelevant information (such as dummy variables)

Code for more than one action may be intertwined Difficult to reuse

Why Is Logical Cohesion So Bad? (contd)

A new tape unit is installed then the new codes will be added that may be inter-twined with the part related to another printer such as section numbered 1,2,3,4,6,9,10

Figure 7.5

Temporal Cohesion A module has temporal cohesion when it

performs a series of actions related in time Example:

open_old_master_file, new_master_file, transaction_file, and print_file; initialize_sales_district_table, read_first_transaction_record, read_first_old_master_record (a.k.a. perform_initialization)

Why Is Temporal Cohesion So Bad?

The actions of this module are weakly related to one another, but strongly related to actions in other modules Consider sales_district_table is being initialized in that module but changing the table is via a different module

When changing a data it will involve many modules some of which may be overlooked

Not reusable – same reason no other software will need all the functions

Procedural Cohesion

A module has procedural cohesion if it performs a series of actions related by the procedure to be followed by the product

Example: read_part_number_and_update_repair_record_on_ master_file

Why Is Procedural Cohesion So Bad? The actions are still weakly connected,

so the module is not reusable

Communicational Cohesion

A module has communicational cohesion if it performs a series of actions related by the procedure to be followed by the product, but in addition all the actions operate on the same data

Example 1:update_record_in_database_and_write_it_to_audit_trail

Both working with record Example 2:

calculate_new_coordinates_and_send_them_to_terminal

Both working with coordinates

Why Is Communicational Cohesion So Bad?

Still lack of reusability

Functional Cohesion

A module with functional cohesion performs exactly one action

7.2.6 Functional Cohesion

Example 1: get_temperature_of_furnace

Example 2: compute_orbital_of_electron

Example 3: write_to_diskette

Example 4: calculate_sales_commission

Why Is Functional Cohesion So Good? More reusable

Common functions can be reused easily Corrective maintenance is easier

Fault isolation – easy to focus Fewer regression faults

Easier to extend a product – add more functions

Informational Cohesion

A module has informational cohesion if it performs a number of actions, each with its own entry point, with independent code for each action, all performed on the same data structure

Why Is Informational Cohesion So Good?

Essentially, this is an abstract data type (see later)

Cohesion Example

Figure 7.7

Example

When two or more different levels of cohesion can be assigned to a module is to assign the lowest possible level

Figure 7.8

Coupling The degree of interaction between two modules

Five categories or levels of coupling (non-linear scale)

Content Coupling Two modules are content coupled if one directly

references contents of the other Example 1:

Module p modifies a statement of module q Example 2:

Module p refers to local data of module q in terms of some numerical displacement within q

Example 3: Module p branches into a local label of module q

Why Is Content Coupling So Bad?

Almost any change to module q, even recompiling q with a new compiler or assembler, requires a change to module p

Difficult to reuse only module p

Common Coupling Two modules are common coupled if they have write

access to global data

Example 1 Modules cca and ccb can access and change the

value of global_variable

Figure 7.9

Common Coupling Example 2:

Modules cca and ccb both have access to the same database, and can both read and write the same record

Example 3: FORTRAN common COBOL common (nonstandard) COBOL-80 global

Why Is Common Coupling So Bad? It contradicts the spirit of structured programming

The resulting code is virtually unreadable What causes this loop to terminate if global_variable is a global variable As its value can be changed with many possibilities

Figure 7.10

Why Is Common Coupling So Bad? (contd)

Modules can have side-effects This affects their readability Example: edit_this_transaction (record_7) The entire module source must be read to find out

what it does because other global variables (not just record_7) may be changed by the function

A change during maintenance to the declaration of a global variable in one module necessitates corresponding changes in other modules

Common-coupled modules are difficult to reuse because global variables are missing in the new software system

Why Is Common Coupling So Bad? (contd)

Common coupling between a module p and the rest of the product can change without changing p in any way

For example, if both module p and q can modify global variable gv then there is one instance of common coupling between p and the other modules. But if 10 new modules are designed and implemented, all of which can modify global variable gv, then the number of instances of common coupling between module p and the other modules increases to 11

A module is exposed to more data than necessary This can lead to computer crime as control of data

access is reduced

Control Coupling Two modules are control coupled if

one passes an element of control to the other

Example 1: An operation code is passed to a

module with logical cohesion Example 2:

A control switch passed as an argument

Control Coupling (contd) Module p calls module q

Message: I have failed (a flag) — data

Message: I have failed, so write error message ABC123

— control The return by q is controlling the

action of p (ie to display the corresponding error message)

Why Is Control Coupling So Bad?

The modules are not independent Module q (the called module) must know the

internal structure and logic of module p otherwise cannot pass back the proper control

This affects reusability Modules are usually logical cohesion

7.3.4 Stamp Coupling Some languages allow only simple

variables as parameters part_number satellite_altitude degree_of_multiprogramming

Many languages also support the passing of data structures

part_record satellite_coordinates segment_table

Stamp Coupling (contd)

Two modules are stamp coupled if a data structure is passed as a parameter, but the called module operates on some but not all of the individual components of the data structure

Why Is Stamp Coupling So Bad?

It is not clear, without reading the entire module, which fields of a record (employee_record) are accessed or changed

Examplecalculate_withholding (employee_record)

Difficult to understand Unlikely to be reusable More data than necessary is passed

Uncontrolled data access can lead to computer crime

Why Is Stamp Coupling So Bad? (contd)

However, there is nothing wrong with passing a data structure as a parameter, provided that all the components of the data structure are accessed and/or changed

Examples:invert_matrix (original_matrix, inverted_matrix);

print_inventory_record (warehouse_record);

7.3.5 Data Coupling

Two modules are data coupled if all parameters are homogeneous data items (simple parameters, or data structures all of whose elements are used by called module)

Examples: display_time_of_arrival (flight_number); compute_product (first_number, second_number); get_job_with_highest_priority (job_queue);

Why Is Data Coupling So Good?

The difficulties of content, common, control, and stamp coupling are not present

Maintenance is easier because a change in one module is less likely to cause a regression fault in the other

Coupling Example

Figure 7.11

The number in the linkis the interfaceP calls q with interface (1)

Coupling Example (contd)

Interface description

Figure 7.12

Coupling Example (contd)

Coupling between all pairs of modules

Figure 7.13

Mini Case Study: Word Counting

Example:

Design a product which takes as input a file name, and returns the number of words in that file (like UNIX wc )

Figure 13.3

From programming point of view A file pointer is used to present the file

Mini Case Study: Word Counting (contd)

First refinement – represented using structured diagram

Now refine the two modules of communicational

cohesion Read_and_validate_file_name shows a communicational cohesion-it performs a series of operationsrelated by the sequence of steps

Structure diagram

In the structured diagram (SD), each node represent a module.

A module is a functional abstraction to be implemented by a software module

Usually assume a modules are called by a master module (transaction flow centered)

SD is derived from the DFD

Second refinement

All eight modules now have functional cohesion

Mini Case Study: Word Counting (contd)

Figure 13.5

Word Counting: Detailed Design

The architectural design is complete So proceed to the detailed design Data structures are chosen Algorithms selected

The detailed design is handed to a programmer for implementation

Two formats for representing the detailed design: Tabular Pseudocode (PDL — program design language) PDL is usually used if the programming language is

chosen

Detailed Design: Tabular Format

Figure 13.6(a)

Detailed Design: Tabular Format (contd)

Figure 13.6(b)

Detailed Design: Tabular Format (contd)

Figure 13.6(c)

Detailed Design: Tabular Format (contd)

Figure 13.6(d)

Detailed Design: PDL Format

Figure 13.7

13.3.2 Data Flow Analysis Extensions

In real-world products, there is More than one input stream, and More than one output stream

Data Flow Analysis Extensions

Find the point of highest abstraction for each stream Continue until each module has high cohesion

Adjust the coupling if needed

Figure 13.8

Exercise

Based on the DFD of the library system, design the circulation system using data flow analysis Draw the structured diagram (2

iterations) Detailed design of

“obtained_catalog_information” PDL for “check_out_book”

Exercise

Convert the above Z specification into a PDL

Data-Oriented Design Basic principle

Design the product according to the structure of the data on which it is to operate

Each procedure is given the same structure as the data on which it operates

Three very similar methods Michael Jackson [1975], Warnier [1976], Orr [1981]

Data-oriented design Has never been as popular as action-oriented design With the rise of OOD, data-oriented design has

largely fallen out of fashion

13.6 Object-Oriented Design (OOD) Aim

Design the product in terms of the classes extracted during OOA

If we are using a language without inheritance (e.g., C, Ada 83)

Use abstract data type design

If we are using a language without a type statement (e.g., FORTRAN, COBOL)

Use data encapsulation

Object-Oriented Design Steps OOD consists of two steps:

Step 1. Complete the class diagram Determine the formats of the attributes Assign each method, either to a class or to a client that sends a

message to an object of that class

Step 2. Perform the detailed design

Object-Oriented Design Steps (contd) Step 1. Complete the class diagram

The formats of the attributes can be directly deduced from the analysis artifacts

Example of deriving an attribute: Dates U.S. format (mm/mm/yyyy) European format (dd/mm/yyyy) In both instances, 10 characters are needed

Object-Oriented Design Steps (contd)

Step 1. Complete the class diagram Assign each method, either to a class or to a client

that sends a message to an object of that class Principle A: Information hiding Principle B: If an operation is invoked by many

clients of an object, assign the method to the object, not the clients

Principle C: Responsibility-driven design (class receiving the request is responsible to carry out the operation)

Information hiding

An object is a collection of data and function so entities outside an object does not know what is inside (information is hidden from the outside world!!!)

13.7 Object-Oriented Design: The Elevator Problem Case Study

Step 1. Complete the class diagram

Need to identify methods (functions) of a class Consider the second iteration of the CRC card for the

elevator controller

OOD: Elevator Problem Case Study

CRC card

Figure 12.9 (again)

OOD: Elevator Problem Case Study (contd)

Responsibilities 8. Start timer 10. Check requests, and 11. Update requests

are assigned to the elevator controller

Because they are carried out by the elevator controller – according to responsibility-driven design

OOD: Elevator Problem Case Study (contd)

The remaining eight responsibilities have the form “Send a message to another class to tell it do

something”

These should be assigned to that other class Responsibility-driven design Safety considerations

Methods open doors, close doors are assigned to class Elevator Doors Class

Methods turn off button, turn on button are assigned to classes Floor Button Class and Elevator Problem Class

Figure 13.11

Detailed Class Diagram: Elevator Problem

Detailed Design: Elevator Problem

Detailed design of elevatorEventLoop

is constructed from the statechart

Figure 13.12

Dynamic Modeling: The Elevator Problem Case Study

Produce a UML statechart

State, event, and predicate are distributed over the statechart

Figure 12.7

13.8 Object-Oriented Design: The MSG Foundation Case Study

Step 1. Complete the class diagram

The final class diagram is shown in the next slide Date Class is needed for C++ Java has built-it functions for handling dates

Final Class Diagram: MSG Foundation

Figure 13.13

Class Diagram with Attributes: MSG Foundation

Figure 13.14

Assigning Methods to Classes: MSG Foundation

Example: setAssetNumber, getAssetNumber From the inheritance tree, these

accessor/mutator methods should be assigned to Asset Class

So that they can be inherited by both subclasses of Asset Class (Investment Class and Mortgage Class)

Figure 13.15

Assigning Methods to Classes: MSG Foundation

Detailed Design: MSG Foundation

Determine what each method does

Represent the detailed design in an appropriate format PDL (pseudocode) here

Method EstimateFundsForWeek::computeEstimatedFunds

Figure 13.16

Method Mortgage::totalWeeklyNetPayments

Figure 13.17

13.9 The Design Workflow Summary of the design workflow:

The analysis workflow documents (input for design workflow) are iterated and integrated until the programmers can utilize them

Major tasks to achieve – identify the methods to be assigned to classes as well as performing detailed design

Decisions to be made include: What programming language to use Reuse of existing software components Portability – work under different software/hardware

platform

The Design Workflow (contd) The idea of decomposing a large workflow into

independent smaller workflows (packages) is carried forward to the design workflow

The objective is to break up the upcoming implementation workflow into manageable pieces Subsystems

It does not make sense to break up the MSG Foundation case study into subsystems — it is too small

The Design Workflow (contd)

Why the product is broken into subsystems: It is easier to implement a number of

smaller subsystems than one large system

If the subsystems are independent, they can be implemented by programming teams working in parallel

The software product as a whole can then be delivered sooner

The Design Workflow (contd)

The architecture of a software product includes The various components How they fit together The allocation of components to subsystems

The task of designing the architecture is specialized It is performed by a software architect

The Design Workflow (contd)

The architect needs to make trade-offs Every software product must satisfy its functional

requirements (the use cases) It also must satisfy its nonfunctional requirements,

including Portability, reliability, robustness, maintainability,

and security It must do all these things within budget and time

constraints

The architect must assist the client by laying out the trade-offs – balance between the requirements and the budget

The Design Workflow (contd) It is usually impossible to satisfy all the requirements,

functional and nonfunctional, within the cost and time constraints

Some sort of compromises have to be made The client has to

Relax some of the requirements; Increase the budget; and/or Move the delivery deadline

The software architect must provide expert knowledge to assist the client to make the proper decision !!!!

The Design Workflow (contd)

The architecture of a software product is critical The requirements workflow can be fixed during

the analysis workflow The analysis workflow can be fixed during the

design workflow The design workflow can be fixed during the

implementation workflow But there is no way to recover from a suboptimal

architecture The architecture must immediately be

redesigned

13.10 The Test Workflow: Design

Testing the design Design reviews (testing) must be

performed The design must correctly reflect the

specifications The design itself must be correct

13.11 The Test Workflow: The MSG Foundation Case Study

A design inspection must be performed All aspects of the design must be

checked

Even if no faults are found, the design may be changed during the implementation workflow

13.12 Formal Techniques for Detailed Design

Implementing a complete product and then proving it correct is hard

However, use of formal techniques during detailed design can help

Correctness proving can be applied to module-sized pieces The design should have fewer faults if it is developed in

parallel with a correctness proof If the same programmer does the detailed design and

implementation The programmer will have a positive attitude to the

detailed design This should lead to fewer faults

13.15 Metrics for Design Measures of design quality

Cohesion Coupling Fault statistics

Cyclomatic complexity (M) – counting the number of binary decisions (branches) plus 1; the lower the value of M, the better

CC = 1 if no decision is made Cyclomatic complexity is problematic

Data complexity is ignored (the value M is related to the control complexity)

It is not used much with the object-oriented paradigm

Cyclomatic Complexity

What is the cyclomaticcomplexity for the flow graph?

Metrics for Design (contd)

Metrics have been put forward for the object-oriented paradigm They have been challenged on both

theoretical and experimental grounds There is no suitable metric for OOD

13.16 Challenges of the Design Phase The design team should not do too much

The detailed design should not become code The design team should not do too little

It is essential for the design team to produce a complete detailed design