1.1 Testing Terminology..............................................................................................................

5
1.1.1 New standard..................................................................................................................5
1.1.2 Error, fault and failure.....................................................................................................5
1.1.3 Reliability........................................................................................................................5
1.2 Why Testing is Necessary......................................................................................................5
1.2.1 Why do software faults occur?.......................................................................................5
1.2.2 What do software faults cost?.........................................................................................6
1.2.3 Testing is necessary.........................................................................................................7
1.2.4 Exhaustive testing...........................................................................................................8
1.2.5 How much testing is enough?.........................................................................................8
1.2.6 Testing measures software quality..................................................................................8
1.2.7 Other factors that influence testing.................................................................................9
1.3 Fundamental Test Process......................................................................................................9
1.3.1 What is the Fundamental Test Process?..........................................................................9
1.3.2 Planning..........................................................................................................................9
1 3.3 A good test case.............................................................................................................10
1.3.4 Specification.................................................................................................................10
1.3.5 Execution......................................................................................................................11
1.3.6 Recording......................................................................................................................11
1.3.7 Check completion.........................................................................................................12
1.3.8 Comparison of the five activities..................................................................................12
1.4 Psychology of Testing..........................................................................................................12
1.4.1 Why test?......................................................................................................................12
1.4.2 Looking for faults is an effective testing approach.......................................................13
1.4.3 Valid and invalid, positive and negative, clean and dirty tests.....................................14
1.4.4 Testing can be perceived as a destructive process........................................................14
1.4.5 Presenting faults to authors and managers....................................................................14
1.4.6 Independent testing is more effective...........................................................................14
1.4.7 Levels of independence.................................................................................................14
1.5 Re-testing and Regression Testing.......................................................................................15
1.5.1 What is re-testing?........................................................................................................15
1.5.2 What is regression testing?...........................................................................................15
1.6 Expected Results..................................................................................................................16
1.6.1 What are expected results?............................................................................................16
1.6.2 Expected results must be defined in advance...............................................................16
1.6.3 Where do expected results come from?........................................................................17
1.6.4 The ‘Oracle Assumption’..............................................................................................17
1.7 Prioritization of Tests...........................................................................................................17
1.7.1 Importance of prioritizing.............................................................................................17
1.7.2 How to prioritize...........................................................................................................17
2. Testing in the Lifecycle..............................................................................................................17
2.1 Models for Testing...............................................................................................................18
2.1.1 Test Design and Test Execution....................................................................................18
2.1.2 The Waterfall Model, Pre-Waterfall, and Damage to Testing.......................................18
2.1.3 The V-Model.................................................................................................................18
2.1.4 When are tests designed: As late as possible?...............................................................19
 2.1.5 When are tests designed: As early as possible!.............................................................19
2.1.6 Verification and Validation............................................................................................20
2.1.7 The impact of early test design on development scheduling........................................21
2.2 Economics of Testing...........................................................................................................21
2.2.1 Testing is expensive?....................................................................................................21
2.2.2 What do software faults cost?.......................................................................................21
2.3 High Level Test Planning.....................................................................................................22
2.3.1 Purpose..........................................................................................................................22
2.3.2 Content of a high level Test Plan..................................................................................22
2.4 Component Testing..............................................................................................................24
2.4.1 What is Component Testing?........................................................................................24
2.4.2 Component Test Strategy..............................................................................................24
2.4.3 Project Component Test Plan........................................................................................26
2.4.4 Component test process................................................................................................26
2.4.5 Test design techniques..................................................................................................26
2.5 Integration Testing in the Small...........................................................................................27
2.5.1 What is Integration Testing in the Small?.....................................................................27
2.5.2 Big Bang integration.....................................................................................................27
2.5.3 Incremental integration.................................................................................................27
2.5.4 Top-down integration and Stubs...................................................................................27
2.5.5 Bottom-up integration and Drivers...............................................................................28
2.5.6 Functional incrementation............................................................................................28
2.5.7 Integration guidelines....................................................................................................28
2.6 System Testing.................................................................................................................29
2.6.1 Functional System Testing............................................................................................29
2.6.2 Requirements-based testing..........................................................................................29
2.6.3 Business process-based testing.....................................................................................29
2.7 Non-Functional System Testing...........................................................................................29
2.7.1 Load, performance & stress testing..............................................................................29
2.7.2 Usability testing............................................................................................................30
2.7.3 Security testing..............................................................................................................30
2.7.4 Configuration and Installation testing...........................................................................30
2.7.5 Reliability testing and other qualities...........................................................................30
2.7.6 Back-up and Recovery testing......................................................................................30
2.7.7 Documentation testing..................................................................................................30
2.8 Integration Testing in the Large...........................................................................................31
2.8.1 What is Integration Testing in the Large?.....................................................................31
2.9 Acceptance Testing..............................................................................................................31
2.9.1 User Acceptance Testing...............................................................................................31
2.9.2 Why users should be involved......................................................................................32
2.9.3 Contract acceptance testing...........................................................................................32
2.9.4 Alpha and Beta testing..................................................................................................32
2.10 Maintenance testing...........................................................................................................32
2.10.1 What is Maintenance testing?.....................................................................................32
2.10.2 Poor or missing specifications....................................................................................33
3. Static Testing..............................................................................................................................33
 3.1 What are Reviews?..............................................................................................................34
3.1.1 Review techniques for individuals................................................................................34
3.1.2 Review techniques for groups.......................................................................................34
3.2 Reviews and the test process................................................................................................34
3.2.1 Benefits of reviews.......................................................................................................34
3.2.2 Reviews are cost-effective............................................................................................34
3.2.3 What can be Inspected?................................................................................................35
3.2.4 What can be reviewed?.................................................................................................35
3.2.5 What to review I Inspect?.............................................................................................35
3.2.6 Costs of reviews............................................................................................................35
3.3 Types of review....................................................................................................................36
3.3.1 Characteristics of different review types......................................................................36
3.3.2 Characteristics of reviews in general............................................................................37
3.3.3 Inspection......................................................................................................................39
3.4 Static analysis.......................................................................................................................41
4. Dynamic Testing Techniques.....................................................................................................43
4.1 About Testing Techniques....................................................................................................43
4.1.1 The need for testing techniques....................................................................................43
4.1.2 What is a testing technique?.........................................................................................43
4.1.3 Advantages of Techniques............................................................................................44
4.2 Black and White Box Testing..............................................................................................44
4.2.1 Types of Testing Technique...........................................................................................44
4.2.2 Static Testing Techniques..............................................................................................45
4.2.3 Functional Testing Techniques (Black Box).................................................................45
4.2.4 Structural Testing Techniques (White Box)..................................................................45
4.2.5 Non-Functional Testing Techniques.............................................................................45
4.2.6 Black Box versus White Box........................................................................................45
4.3 Black Box Test Techniques..................................................................................................46
4.3.1 Techniques Defined in BS 7925-2................................................................................46
4.3.2 Equivalence Partitioning & Boundary Value Analysis.................................................46
4.3.3 State Transition Testing.................................................................................................47
4.4 Non-Systematic Techniques.................................................................................................48
4.4.1 Trail and Error / Ad Hoc...............................................................................................48
4.4.2 Error-Guessing..............................................................................................................48
4.4.3 Lateral Testing...............................................................................................................49
4.4.4 User Testing..................................................................................................................49
4.4.5 Unscripted Testing........................................................................................................49
4.5 White Box Test Techniques..................................................................................................49
4.5.1 What are Coverage Techniques?...................................................................................49
4.5.2 Types of Coverage........................................................................................................50
4.5.3 How to Measure Coverage............................................................................................50
4.5.4 Statement Coverage......................................................................................................51
4.5.5 Branch & Decision Testing / Coverage.........................................................................51
5. Test Management.......................................................................................................................51
5.1 Organization.........................................................................................................................52
5.1.1 The importance of independence..................................................................................52
 5.1.2 Organizational structures for testing.............................................................................52
5.1.3 Usual choices................................................................................................................54
5.1.4 Resourcing issues..........................................................................................................54
5.2 Configuration Management.................................................................................................55
5.2.1 What is configuration management?............................................................................55
5.2.2 Problems resulting from poor configuration management...........................................55
5.2.3 Definition of configuration management......................................................................55
5.2.4 Configuration management in testing...........................................................................56
5.3 Test Estimation, Monitoring and Control............................................................................56
5.3.1 Estimating testing is no different..................................................................................56
5.3.2 Estimating testing is different.......................................................................................57
5.3.3 Monitoring progress......................................................................................................58
5.3.4 Test control........................................................................................................................58
5.4 Incident Management...........................................................................................................59
5.4.1 What is an incident?......................................................................................................59
5.4.2 Monitoring incidents.....................................................................................................59
5.4.3 Reporting incidents.......................................................................................................59
5.4.4 Tracking incidents.........................................................................................................60
5.4.5 Severity versus priority.................................................................................................60
5.5 Standards for Testing...........................................................................................................60
6. Tool Support for Testing (CAST)..............................................................................................60
6.1 Types of CAST Tool............................................................................................................60
6.1.1 Requirements testing tools............................................................................................60
6.1.2 Static analysis tools.......................................................................................................61
6.1.3 Test design tools............................................................................................................61
6.1.4 Data preparation tools...................................................................................................62
6.1.5 Test running tools..........................................................................................................62
6.1.6 Comparison tools..........................................................................................................62
6.1.7 Test harnesses and drivers.............................................................................................63
6.1.8 Performance testing tools..............................................................................................63
6.1.9 Dynamic analysis tools.................................................................................................63
6.1.10 Debugging tools..........................................................................................................63
6.1.11 Test management tools................................................................................................63
6.1.12 Coverage tools............................................................................................................64
6.2 Tool Selection and Implementation.....................................................................................64
6.2.1 Which test activities to automate?................................................................................64
6.2.2 CAST tool requirements...............................................................................................64
6.2.3 CAST readiness............................................................................................................64
6.2.4 Environment constraints and CAST integration...........................................................65
6.2.5 Selection process...........................................................................................................65
6.2.6 Implementation (pilot project and roll out)...................................................................65
1. Principles of Testing

1.1 Testing Terminology

1.1.1 New standard

There is a lot of terminology surrounding testing, but not until recently has there been an
industry standard. A new standard (first published in August 1998) seeks to provide a
standard set of terms: BS 7925-i Glossary of Testing Terms. Although a British Standard, it
is being adopted by the International Standards Organization (ISO) and will hopefully
become an ISO standard within two or three years. A selection of terms from this standard,
plus some others used in this course are in the Glossary at the back of these notes.

1.1.2 Error, fault and failure

Three terms that have specific meanings are error, fault and failure.

Error: a human action that produces an incorrect result.

Fault: a manifestation of an error in software.
Failure: a deviation of the software from its expected delivery or service.

An error is something that a human does, we all make mistakes and when we do whilst
developing software it is known as an error. The result of an error being made is a fault. It is
something that is wrong in the software (source code or documentation - specifications,
manuals, etc.). Faults are also known as a defects or bugs but in this course we will use the
term fault.

When a system or piece of software produces an incorrect result or does not perform the
correct action this is known as a failure. Failures are caused by faults in the software. Note
that software system can contain faults but still never fail (this can occur if the faults are in
those parts of the system that are never used).

1.1.3 Reliability
Another term that should be understood is reliability. A system is said to be reliable when it
performs correctly for long periods of time. However, the same system used by two different
people may appear reliable to one but not to the other. This is because the different people
use the system in different ways.

Reliability: the probability that the software will not cause the failure of the system for a
specified time under specified conditions.

The definition of reliability therefore includes the phrase ‘under specified conditions’. When
reporting on the reliability of a system it is important to explain under what conditions the
system will achieve the specified level of reliability. For example, a system may achieve a
reliability of no more than one failure per month providing no more than 10 people use the
system simultaneously.

1.2 Why Testing is Necessary

1.2.1 Why do software faults occur?

All of the products of software development (specifications, source code, and test
documents) are written by human beings. All human beings are prone to make errors
regardless of how experienced or skilled they are. So a number of faults in software are
inevitable/expected.
Furthermore, many of the computer systems being developed today are so large and
complex that it is not generally possible for any one person to understand every aspect of a
system. Many people have specialized in particular areas or aspects, for example we have
database experts, software design experts, algorithm experts, and network and operating
system experts. Also, there are experts in specific areas of the businesses that the systems
support. Good documentation (specifications and the like) are necessary to communicate all
the relevant information each person needs to complete their parts of a system. However,
people are fallible and so make errors causing faults in the specifications. When dealing with
one part of a system, a systems designer or programmer may think they understand some
other aspect outside of their particular area of expertise but they may not. Assumptions are
often made because it is easier and quicker to assume one thing than to find out what the
real answer should be.

1.2.2 What do software faults cost?

The cost of software faults that are not detected before a system is put into live operation
varies, depending on a number of factors.

Some of the more spectacular software failures are well known. For example the European
Space Agency’s Ariane 5 rocket exploded 37 seconds into its launch because of a software
fault. This fault cost $7billion. Although it had been tested a series of errors meant that out-
of-date test data was used. Another example is the Mariner Space probe. This was meant to
be going to Venus but lost its way as a result of a software fault. The FORTRAN program had
a full stop instead of a comma in a FOR (looping) statement that the compiler accepted as
an assignment statement.

A couple of more recent examples involve Mars probes. In the first example the one
organization’s software calculated the distance to landing in inches while another calculated
the distance in centimeters. This resulted in the loss of the probe at a cost of $125 million.
In the second example, a probe was lost shortly before it should have landed at a cost of
$250 million. It turned out to be spurious signals from hardware sensors were interpreted
by the software as landing, so antennas were deployed, protective covers removed, etc.
This fault had been identified in earlier testing, and changes were made to the hardware
and the software. However, there was no final regression test, due to time and budget
pressure. (The lack of a final test was also responsible for the Hubble telescope being
myopic for its first years of operation.)

One final example is where the expected results were not calculated beforehand and in this
instance cost American Airlines $5Omillion. By way of background to this example, airlines
never want to fly with empty seats - they would prefer to offer discount seats than to fly
empty. However they also do not want to discount the seat when they could receive the full
price. Airlines run complex ‘yield management programs’ to achieve the balance right. But
these programs often change a lot to increase profitability. One such change was
catastrophic for American Airlines. The program gave reasonable -but wrong - results for
the number of discount seats! A spokesman said ‘We’re convinced that if we would have
done more thorough testing, we would have discovered the problem before the software
was ever brought on-line’.

Of course not every software fault causes such huge failure costs (fortunately!). Some
failures are simple spelling mistakes or a misalignment in a report.

There is a vast array of different costs but there is a problem, software is not linear. Small
faults can have large effects. For example a spelling mistake on a screen title said “ABC
SOFTWEAR”, a simple mistake but it cost the company a sale. Another spelling mistake was
when the programmer spelt an insurance company’s name wrong — instead of “Union” they
spelt it “Onion” — only one letter, but this was a major embarrassment.

The consequences of faults in safety critical systems can be much more serious. The follow
examples are all real cases.

• Therac-25 (Canada) - this machine was used in the treatment of cancer patients. It
had the capability of delivering both x-rays and gamma rays. The normal dose of x-
rays is relatively low while gamma ray dosage is much higher. Switching from one
type of ray to the other was a slow process but the operators discovered that if they
typed fast it could be made to switch more quickly. However, it only appeared to
have switched. Six people died as a result of having too high an x-ray dose. This was
a design fault.

• A train driver in Washington DC was killed when an empty train failed to stop in a
stabling siding. The train was fully automated; driver just opens and closes doors.
Driver had been requesting manual operation, as the trains had been overshooting
the stops - this was denied by the humans in “central control” - RISKS Forum, 26 Jan
1998.

• Korean Airlines Flight 801, Boeing 747, 29 survivors (out of 254 passengers and
crew). Came in 1000 ft too low. Combination of “normal” factors: severe weather,
ground proximity system not switched on (or automatically disabled when landing
gear lowered), secondary radar system should have sounded an alarm - either didn’t
or wasn’t noticed. - Computer Weekly, 14 Aug 1987.

• Airbus crashes into trees at an air show at Habsham, Germany. Software protected
engines by preventing them from being accelerated as fast as the pilot required to
avoid hitting some trees after a low fly pass.

The following example is not of a safety critical system, rather a banking system. However,
this did contribute to a death.

• An elderly man bought presents for his grandchildren, and as a result became
overdrawn for the first time in his life. The banks system automatically sent
threatening letters and charged him for being overdrawn. The man committed
suicide. Source: client from major UK bank.

1.2.3 Testing is necessary

Testing is necessary because software is likely to have faults in it and it is better (cheaper,
quicker and more expedient) to find and remove these faults before it is put into live
operation. Failures that occur during live operation are much more expensive to deal with
than failures that occur during testing prior to the release of the software. Of course other
consequences of a system failing during live operation include the possibility of the software
supplier being sued by the customers!

Testing is also necessary so we can learn about the reliability of the software (that is, how
likely it is to fail within a specified time under specified conditions).

Testing is not ‘to prove that the software has no faults’ because it is impossible to prove that
software has no faults. Neither is testing done ‘because testing is included in the project
plan’. Testing should be included in the project plan, but this is not the reason for testing.
(Why is this testing included in the plan - this is the reason for testing.)

Testing is essential in the development of any software system. Testing is needed in order to
assess what the system actually does, and how well it does it, in its final environment. A
system without testing is merely a paper exercise - it may work or it may not, but there is
no way of knowing it without testing it.
1.2.4 Exhaustive testing

Exhaustive testing is defined as exercising every possible combination of inputs and

preconditions. This usually amounts to an impractical number of test cases. For most
systems it is relatively easy to identify millions of possible test cases (not necessarily all
desirable test cases). For example, a system that accepts an input value corresponding to
an employee’s annual salary may be expected to handle figures in the range £0.00 to
£999,999.00. Given that every different input value could be made a separate test case this
means that we can identify 100,000,000 test cases (the number of ip’s in the range) just
from this one input value. Of course, no one would reasonably expect to have to try out
every one of these input values but that is what exhaustive testing would require.
Exhaustive testing would also require each of these 100,000,000 input values to be tried in
different combinations with other inputs and with different preconditions (such as data
values in a database). For even simple systems the number of test cases required for
exhaustive testing is likely to run into millions of millions of millions.

Exhaustive testing is, therefore, impractical (it is not considered impossible, as it may be
achievable for a very small program).

1.2.5 How much testing is enough?

It is possible to do enough testing but determining the how much is enough is difficult.
Simply doing what is planned is not sufficient since it leaves the question as to how much
should be planned. What is enough testing can only be confirmed by assessing the results of
testing. If lots of faults are found with a set of planned tests it is likely that more tests will
be required to assure that the required level of software quality is achieved. On the other
hand, if very few faults are found with the planned set of tests, then (providing the planned
tests can be confirmed as being of good quality) no more tests will be required.

Saying that enough testing is done when the customers or end-users are happy is a bit late,
even though it is a good measure of the success of testing. However, this may not be the
best test stopping criteria to use if you have very demanding end-users who are never
happy!

Why not stop testing when you have proved that the system works? It is not possible to
prove that a system works without exhaustive testing (which is totally impractical for real
systems).

Have you tested enough when you are confident that the system works correctly? This may
be a reasonable test stopping criterion, but we need to understand how well justified that
confidence is. It is easy to give yourself false confidence in the quality of a system if you do
not do good testing.

Ultimately, the answer to “How much testing is enough?” is “It depends!” (this was first
pointed out by Bill Hetzel in his book “The Complete Guide to Software Testing”). It depends
on risk, the risk of missing faults, of incurring high failure costs, of losing creditability and
market share. All of these suggest that more testing is better. However, it also depends on
the risk of missing a market window and the risk of over-testing (doing ineffective testing)
which suggest that less testing may be better.

We should use risk to determine where to place the emphasis when testing by prioritizing
our test cases. Different criteria can be used to prioritize testing including complexity,
criticality, visibility and reliability.

1.2.6 Testing measures software quality

Testing and quality go hand in hand! Basically we don’t know how good the software is until
we have run some tests. Once we have run some good tests we can state how many faults
we have found (of each severity level) and may also be able to predict how many faults
remain (of each severity level).

1.2.7 Other factors that influence testing

Other factors that affect our decision on how much testing to perform include possible
contractual obligations. For example, a contract between a customer and a software supplier
for a bespoke system may require the supplier to achieve 100% statement coverage
(coverage measures are discussed in session 3). Similarly, legal requirements may impose a
particular degree of thoroughness in testing although it is more likely that any legal
requirements will require detailed records to be kept (this could add to the administration
costs of the testing).

In some industries (such as the pharmaceutical industry and safety critical industries such
as railroad switching and air traffic control) there are standards defined that have the intent
of ensuring rigorous testing.

1.3 Fundamental Test Process

1.3.1 What is the Fundamental Test Process?

This section describes the Fundamental Test Process. This is a test process that is
documented in the standard BS7925-2 Software Component Testing. It therefore relates
most closely to component testing but is considered general enough to apply to all levels of
testing (i.e. component, integration in the small, system, integration in the large, and
acceptance testing).

It is perhaps most applicable to a fairly formal testing environment (such as mission

critical). Most commercial organisations have less rigorous testing processes. However, any
testing effort can use these steps in some form.

The Fundamental Test Process comprises five activities: Planning, Specification,

Execution, Recording, and Checking for Test Completion. The test process always
begins with Test Planning and ends with Checking for Test Completion. Any and all of the
activities may be repeated (or at least revisited) since a number of iterations may be
required before the completion criteria defined during the Test Planning activity are met.
One activity does not have to be finished before another is started; later activities for one
test case may occur before earlier activities for another. The five activities are described in
more detail below.

1.3.2 Planning
The basic philosophy is to plan well. All good testing is based upon good test planning.
There should already be an overall test strategy and possibly a project test plan in place.

This Test Planning activity produces a test plan specific to a level of testing (e.g. system
testing). These test level specific test plans should state how the test strategy and project
test plan apply to that level of testing and state any exceptions to them. When producing a
test plan, clearly define the scope of the testing and state all the assumptions being made.
Identify any other software required before testing can commence (e.g. stubs & drivers,
word processor, spreadsheet package or other 3rd party software) and state the completion
criteria to be used to determine when this level of testing is complete.

Example completion criteria are (some are better than others and using a combination of
criteria is usually better than using just one):
 • 100% statement coverage;
• 100% requirement coverage;
• all screens I dialogue boxes I error messages seen;
• 100% of test cases have been run;
• 100% of high severity faults fixed;
• 80% of low & medium severity faults fixed;
• maximum of 50 known faults remain;
• maximum of 10 high severity faults predicted;
• time has run out;
• testing budget is used up.
A note to the reader: Any text that is ‘‘grayed out’’ (as the text immediately below) is not
directly ~. related to the Foundation Syllabus. We have included it here as background
material — skip over it if i~ you are currently revising for the exam! Hopefully the ideas
will be useful to you after the course.

1 3.3 A good test case

A good test case is effective. That is, it will find a fault. This does not mean that a test case
that does not find a fault is not a good one, since this implies that the fault that the test
case could have found is not present (i.e. it gives us some confidence in the software, and
this in itself has value). Perhaps we should say that a good test case has the potential to
find a fault.

~ A good test case is exemplary; meaning that it does more than one thing for us (is
capable of finding more than one fault).

~ A good test case is evolvable. As the software changes so too will some of the tests need
changing to reflect different functionality, new features, etc. The effort to update test cases
is usually very significant. However, much can be done when designing test cases to reduce
or minimise the amount of maintenance effort needed to update test cases to make them
compatible with later versions of the software.

~ A good test case is economic. A test case that requires 50 people to come into the office
on a Saturday ~ morning and all be poised at their keyboards at 9am is expensive to
perform and it can only be run once a week. A test case that can be run at the touch of a
button and only lasts two seconds is much more economic.

We should design test cases to be economic, evolvable, exemplary and effective. However,
we often need to design a balance between effective/exemplary and evolvable Economic.

1.3.4 Specification

The fundamental test process describes this activity as designing the test cases using the
techniques selected during planning. For each test case, specify its objective, the initial
state of the software, the input sequence and the expected outcome. Since this is a little
vague we have broken down the Test Specification activity into three distinct tasks to
provide a more helpful explanation. (Note that this more detailed explanation of the Test
Specification is not a part of the Foundation syllabus.)

Specification can be considered as three separate tasks:

• identify test conditions — determines ‘what’ is to be tested;
• Design test cases - determine ‘how’ the ‘what’s’(test conditions) are going to be
exercised;
• Build test cases - implementation of the test cases (scripts, data, etc.).
Identify test conditions

Focal testing techniques mostly concentrate on this task, i.e. identifying test conditions.
Sometimes a ~ brainstorming session is also good for this. When we initially brainstorm we
will think of a few conditions - that in turn will trigger more, but remember that the first
50% produced is unlikely to be the best 50%. Therefore, if you need 100 test conditions,
identify 200 (or even 1000) and pick the best 100 test conditions.

Test cases that exercise the most important test conditions will be effective (recall the four
attributes of a good test case).

Design test cases

Designing good test cases is a skill. To be exemplary a test case should exercise several test
conditions but to be economic and evolvable it should not be too big or too complex.
Predicting the expected outcome is its own syllabus topic later on. The term outcome is
used in preference to output because the outcome comprises everything that has been
output and what has been ~ changed, deleted, and not changed. I It 5 frequently necessary
to design not only individual test cases but also whole sets of test cases each I with different
objectives. For example, regression tests, performance tests, and detailed tests of a ~
particular function.

Build test cases

This task involves making the test cases a reality. Writing test procedures or test scripts,
creating or I acquiring the test data and implementing the expected results. These are all
the pre-requisites prior to I test execution.

1.3.5 Execution

The purpose of this activity is to execute all of the test cases (though not necessarily all in
one go). This can be done either manually or with the use of a test execution automation
tool (providing the test cases have been designed and built as automated test cases in the
previous stage).

The order in which the test cases are executed is significant. The most important test cases
should be executed first. In general, the most important test cases are the ones that are
most likely to find the most serious faults but may also be those that concentrate on the
most important parts of the system.

There are a few situations in which we may not wish to execute all of the test cases. When
testing just fault fixes we may select a subset of test cases that focus on the fix and any
likely impacted areas (most likely all the test cases will have been run in a previous test
effort). If too many faults are found by the first few tests we may decide that it is not worth
executing the rest of them (at least until the faults found so far have been fixed). In
practice time pressures may mean that there is time to execute only a subset of the
specified test cases. In this case it is particularly important to have prioritised the test cases
to ensure that at least the most important ones are executed.

If any other ideas for test conditions or test cases occur they should be documented where
they can be considered for inclusion.

1.3.6 Recording
In practice the Test Recording activity is done in parallel with Test Execution. To start with
we need to record the versions of the software under test and the test specification being
used. Then for each test case we should record the actual outcome and the test coverage
levels achieved for those measures specified as test completion criteria in the test plan. In
this way we will be marking off our progress. The test record is also referred to as the “test
log”, but “test record” is the terminology in the syllabus. Note that this has nothing to do
with the recording or capturing of test inputs that some test tools perform!

The actual outcome should be compared against the expected outcome and any discrepancy
found logged and analysed in order to establish where the fault lies. It may be that the test
case was not executed correctly in which case it should be repeated. The fault may lie in the
environment set-up or be the result of using the wrong version of software under test. The
fault may also lie in the specification of the test case: for example, the expected outcome
could be wrong. Of course the fault may also be in the software under test! In these cases
the fault should be fixed and the test case executed again.

The records made should be detailed enough to provide an unambiguous account of the
testing carried out. They may be used to establish that the testing was carried out according
to the plan.

1.3.7 Check completion

This activity has the purpose of checking the records against the completion criteria
specified in the test plan. If these criteria are not met, it will be necessary to go back to the
specification stage to specify more test cases to meet the completion criteria. There are
many different types of coverage measure and different coverage measures apply to
different levels of testing. (Coverage measures are described in session 3.)

1.3.8 Comparison of the five activities.

Comparing these five activities of the Fundamental Test Process it is easy to see that the
first two activities (Test Planning and Test Specification) are intellectually challenging.
Planning how much testing to do, determining appropriate completion criteria, etc. requires
careful analysis and thought. Similarly, specifying test cases (identifying the most important
test conditions and designing good test cases) requires a good understanding of all the
issues involved and skill in balancing them. It is these intellectual tasks that govern the
quality of test cases.

The next two activities (Test Execution and Test Recording) involve predominantly clerical
tasks. Furthermore, executing and recording are activities that are repeated many times
whereas the first two activities, Test Planning and Test Specification are performed only once
(they may be revisited if the completion criteria are not met the first time around but they
are not repeated from scratch). The Test Execution and Test Recording activities can be
largely automated and there are significant benefits in doing so.

1.4 Psychology of Testing

1.4.1 Why test?

There are many reasons for testing including building confidence in the software under test,
demonstrating conformance to requirements or a functional specification, to find faults,
reduce costs, demonstrate a system meets user needs and assessing the quality of the
software. However, one reason that is not valid (but often used erroneously) is to prove that
the software is correct. This is wrong simply because it is not possible to prove that a
software system is correct. It is not possible to prove a system has no faults. It is only
possible to prove that a system has faults - by finding some of them!
How do we measure the quality of software? The number of faults found is a common way
to start. The more faults we find, the worse the quality is, so we can easily measure poor
quality software.

What about the converse: the fewer faults we find, the better the quality of the software?
This is not true, since finding few or no faults can mean one of three things: good software,
poor testing or both poor testing and poor software. Without knowing independently about
the quality of the testing, no justified conclusions can be drawn about the quality of the
software.

If you have a faulty measuring instrument, you cannot draw any justified conclusions about
what you are measuring. If I discovered that I had lost weight, I might be delighted, but if
my scales are broken, I am not actually any thinner - my delight is unjustified.

If we do lots of testing, our confidence will rise, but this is because confidence is a
psychological factor. Consider how differently you might regard a piece of software if you
run all of the easy tests first, so most of the early tests work correctly, compared to running
all of the most difficult tests first, so that most of the early tests fail. The same software
with the same set of tests would give two very different initial impressions to your
assessment of confidence, even though the quality of that software is the same each time.

You cannot have justified confidence in the quality of the software

unless you have confidence in the quality of the testing.

1.4.2 Looking for faults is an effective testing approach

The overall purpose of testing is to give confidence that the system is working well. But this
definition of testing is not effective for test case construction. Glenford Myers shows that the
purpose of testing is to find faults, as is pointed out so effectively in his book “The Art of
Software Testing”, Wiley, 1979.

But finding faults destroys confidence - obviously you cannot have confidence that
something is working when it has just been proved otherwise. Therefore, is the purpose of
testing to destroy confidence, rather than to build it? This is a paradox because both are
true.

The best way to build confidence is to try and destroy it.

If no faults are found, the natural assumption is that the software has no faults; the
assumption that the testing was of poor quality is rarely made but is much more likely. In
fact, you cannot justifiably conclude anything about the quality of the software tested
without some knowledge of the quality of the testing.

Yet when testing reveals no faults, our natural reaction is to be pleased, because we always
like to think the best of ourselves. To be honest, we don’t really want to succeed at finding
faults because we never meant to put them in to begin with. This leads to the “Catch-22” of
testing.

Testing is looking for something you don’t really want to find.

Misunderstanding the testing paradox leads to misconceptions about testing. If you

observed the way people behave, you would think they believed that it is the testing process
that creates faults in software. This is like believing that it is the sunshine streaming in
through a window that creates dust in the air. The sunshine, like testing, only makes visible
what was already there. “Less testing means less debugging” is actually true; the fewer
faults you find, the fewer you would put right before you release the software to the next
stage. However, not finding them does not mean they are not there.

It is sometimes said that we should “build quality in, not test bugs out”. Although it is
certainly a good idea to build quality in, there are misconceptions here as well. There is a
tendency to omit the last two words. But it is not a choice between quality and testing;
testing is a way to build quality in, as will become evident in session 2. It is not true that
faults are “tested out” in any case. Debugging is the process that removes faults; testing
reveals them.

1.4.3 Valid and invalid, positive and negative, clean and dirty tests

Choosing test inputs that the software is expected to handle is an important thing to do. For
example, if a program is expected to accept an integer in the range 1 to 1 000 then test
cases that input an integer in this range are regarded as valid tests in the sense that they
use valid input value. These are also known as positive tests or clean tests.

However, where we choose to use test inputs outside of this range such as 4500, -20 or
even “Hello world”, these test cases are regarded as invalid tests in the sense that they use
invalid input values. These are also known as negative or dirty tests.

1.4.4 Testing can be perceived as a destructive process

Testing can be seen as a destructive process. Looking for faults in something has a negative
connotation; it implies that the something is faulty to start with. The point is, when it comes
to software, it usually is faulty in some way! (Of course developers are inclined to believe
the best of their software, because they created it, it is their baby.)

1.4.5 Presenting faults to authors and managers

Care should be taken when communicating fault information to developers and managers.
In ancient Greece messengers who brought bad news were executed so some things have
improved! However, we must still tread carefully. Dashing up to a developer and saying “You
fool, you did this wrong” is not likely to encourage him or her to investigate the problem.
More likely the developer will go on the defensive, perhaps arguing that it is not a fault but
it is the tester who does not understand it. A more successful approach may be to approach
the developer saying “I don’t understand this, would you mind explaining it to me please?”
In demonstrating it to the tester the developer may then spot the fault and offer to fix it
there and then.

Cem Kaner (co-author of “Testing Computer Software”) says that the best tester is not the
one who finds the most faults but the one who manages to have the most faults fixed. This
requires a good relationship with developers.

1.4.6 Independent testing is more effective

Testing done by someone who has not been involved in the development of the software
under test is likely to be better (i.e. find more faults) than testing done by someone who
has been involved in the software’s development. This is because the person who has been
involved in the development process will have a restricted view of the software, a
developer’s view. In this case any assumptions made during the development process are
likely to be carried over into testing. The other person will be able to view the software
independently of the development process and be able to challenge any assumptions the
developers made.

1.4.7 Levels of independence

There are many levels of independence, that is, ways of achieving a greater or lesser
amount of independence in software testing. It is important to appreciate that the
independence is most required at the specification stage of the testing process. It is here
that the test cases are designed and it is the design of the design test cases (more
specifically, the identification and selection of the test conditions) that governs the quality of
the test cases. People who say that programmers should not test their own code often miss
this point. What they should say is that programmers are not the best people to specify all
of the tests for their own code. However, programmers should contribute to the specification
of the test cases since they can contribute a good technical understanding of the software.

There are good reasons why programmers should execute the tests on their own code.
Perhaps the best of these is that it is cheaper. A programmer who finds a fault in his or her
own code can quickly fix it and execute the test again to confirm the fix. The fact that a
fault was found and fixed does not need to be documented. If someone else were to run the
test, they probably would need to document it as a safe and secure way of communicating
the details to the programmer. The programmer would then need to reproduce the failure
and fix the fault. The fault report would then be updated and make its way back to the
tester who would repeat the test to confirm the fix was correct. This takes much more effort
in total and yet no more has been achieved: one fault has been found and fixed.

We will achieve no independence if only the person who wrote the software specifies the
tests. If another developer from the same team were to specify the tests, a little
independence is achieved. More independence can be achieved if someone outside of the
development team specifies the tests (such as a test team). Further independence can be
achieved by having an outside agency undertake the testing though this in itself introduces
some different problems. Perhaps the greatest level of independence can be achieved by
having a tool generate test cases but these are not likely to be particularly good quality
tests.

Issues of independence are discussed further in session 5.

1.5 Re-testing and Regression Testing

1.5.1 What is re-testing?

When a test fails and we determine the cause of the failure is a software fault, the fault is
reported, we can expect a new version of the software that has had the fault fixed. In this
case we will need to execute the test again to confirm that the fault has indeed been fixed.
This is known as re-testing.

When re-testing, it is important to ensure that the test is executed in exactly the same way
as it was the first time, using the same inputs, data and environment. If the test now
passes does this mean that the software is now correct? Well, we now know that at least
one part of the software is correct — where the fault was. But this is not enough. The fix
may have introduced a different fault elsewhere in the software. The way to detect these
“unexpected side-effects” of fixes is to do regression testing.

1.5.2 What is regression testing?

Like re-testing, regression testing involves executing test cases that have been executed
before. The difference is that for regression testing the test cases probably passed the last
time they were executed (compare this with the test cases executed when re-testing - they
failed the last time).

The term “regression testing” is something of a misnomer. It would be better if it were

called “anti-regression” or “progression” testing because we are executing tests with the
intent of checking that the system has not regressed (that is, it does not now have more
faults in it as a result of some change). More specifically, the purpose of regression testing
is to verify that modifications in the software or the environment have not caused
unintended adverse side effects and that the system still meets it requirements.

It is common for organizations to have what is usually called a regression test suite or
regression test pack. This is a set of test cases that is specifically used for regression
testing. They are designed to collectively exercise most functions (certainly the most
important ones) in a system but not test any one in detail. It is appropriate to have a
regression test suite at every level of testing (component testing, integration testing,
system testing, etc.). All of the test cases in a regression test suite would he executed every
time a new version of software is produced and this makes them ideal candidates for
automation. If the regression test suite is very large it may be more appropriate to select a
subset for execution.

Regression tests are executed whenever the software changes, either as a result of fault
fixes or new or changed functionality. It is also a good idea to execute them when some
aspect of the environment changes, for example when a new version of a database
management system is introduced or a new version of a source code compiler is used.

Maintenance of a regression test suite should be carried out so it evolves over time in line
with the software. As new functionality is added to a system new regression tests should be
added and as old functionality is changed or removed so too should regression tests be
changed or removed. As new tests are added a regression test suite may become very
large. If all the tests have to be executed manually it may not be possible to execute them
all every time the regression suite is used. In this case a subset of the test cases has to be
chosen. This selection should be made in light of the latest changes that have been made to
the software. Sometimes a regression test suite of automated tests can become so large
that it is not always possible to execute them all. It may be possible and desirable to
eliminate some test cases from a large regression test suite for example if they are
repetitive (tests which exercise the same conditions) or can be combined (if they are always
run together). Another approach is to eliminate test cases that have not found a fault for a
long time (though this approach should be used with some care!).

1.6 Expected Results

1.6.1 What are expected results?

An expected result is what the software is supposed to produce when a test is executed. We
usually refer to the expected outcome (or results) rather than the expected output since
outcome is a more encompassing word. Outcome includes everything that has been created,
changed, or deleted and also includes things that should not have changed. In other words
it is the difference between the state of the system and its environment before and after
executing the test.

1.6.2 Expected results must be defined in advance

It is important that expected results be specified in advance of tests being executed though
not doing so is a fairly common practice. If we do not specify the expected results for a test
the intention would be to verify the actual results by viewing them at the time we execute
the test. This has the advantage of reducing the amount of work we have to do when
specifying the tests (less documentation, less effort). However, this approach has the
disadvantage of being a less reliable. There may be a subconscious desire to see the tests
pass (less work to do - no fault report to write up). It is also a less rigorous approach since
a result that looks plausible may be thought correct even when it isn’t wholly correct.
Calculating an expected result in advance of test execution and then comparing the actual
result with this is a much more reliable approach. It also means that the tester does not
have to have the knowledge that would allow him or her to properly verify test outcomes.
1.6.3 Where do expected results come from?

Expected results should be derived from a specification of what the system should do (for
example a requirement or functional specification). In other words the expected results are
determined by considering the required behavior of the system.

Some approaches to testing lead testers to consider the structure of the system when
designing test cases, in particular by examining the source code. While this is a good
approach to identify test inputs it is vital that the expected result is not derived in this way.
The expected result should be derived from the required behavior of the system, not its
actual behavior. Passing expected results on constructs and values within the source code
will have the effect of generating test cases that test that the software does what the
software does, whereas what we really need is test cases that test that the software does
what the software should do.

1.6.4 The ‘Oracle Assumption’

For most systems it is possible to predict the expected results for any test case but there
are a few systems where this is not the case. For example, systems that predict situations
or perform long and complex calculations that cannot practically be performed manually.
The Oracle Assumption is the name given to the (usually correct) assumption that it is
possible to predict the expected results.

1.7 Prioritization of Tests

1.7.1 Importance of prioritizing

We have already said that we cannot test everything, exhaustive testing (testing all
combinations of inputs and preconditions) is impractical. It is easy enough to identify far
more test cases than we will ever have time to execute so we need an approach to selecting
a subset of them. Selecting test cases at random is not an effective strategy. We need to
use a more intelligent approach that helps identify which tests are most important. In short,
we must prioritize our tests.

Prioritize tests so that whenever you stop testing you have done the best testing
in the time available.

1.7.2 How to prioritize

There are many different criteria that can be used to prioritize tests and they can be used in
combination. Possible ranking criteria include the following:

• tests that would find the most severe failures;

• tests that would find the most visible failures;
• tests that would find the most likely faults;
• ask the end-user to prioritize the requirements (and test those first);
• test first the areas of the software that have had most faults in the past;
• test most those areas of the software that are most complex or critical.

There is further discussion and examples of how to prioritize tests in sessions 4 and 5.

2. Testing in the Lifecycle

This session concentrates on the relationship between testing and all the other development
activities that occur throughout the software development lifecycle. In particular it looks at
the different levels of testing.

2.1 Models for Testing

2.1.1 Test Design and Test Execution

Testing is often considered something which is done after software has been written; after
all, the argument runs, you can’t test something which doesn’t exist, can you? This idea
makes the assumption that testing is merely test execution, the running of tests. Of course,
tests cannot be executed without working software (although static analysis can be carried
out). But testing activities include more than running tests, because before they can be run,
the tests themselves need to be written and designed. As Boris Beizer points Out in his book
“Software Testing Techniques, Van Nostrand Reinhold, 1993, the act of designing a test is
one of the most effective ways of finding faults. If tests need to be designed, when in the
life cycle should the test design activities happen?

The discussion below, on the timing of the test design activities, applies no matter what
software life cycle model is used during development or maintenance. Many descriptions of
life cycles do not make the proper placing of test design activities clear. For example, many
development methods advocate early test planning (which is good), but it is the actual
construction of concrete individual test inputs and expected outcomes which is most
effective at revealing faults in a specification, and this aspect is often not explicit.

2.1.2 The Waterfall Model, Pre-Waterfall, and Damage to Testing

Most software engineers are familiar with a software life cycle model; the waterfall was the
first such model to be generally accepted. Before this, there were informal mental models of
the software development process, but they were fairly simple. The process of producing
software was referred to as programming”, and it was integrated very closely with testing.
The programmers would write some code, try it out, and write some more. After a lot of
iterations, the program would emerge. The point is that testing was very much an integral
part of the software production process.

The main difference in the waterfall model was that the programming steps were spelled
out. Now instead of programming, there are a number of distinct stages such as
requirements analysis, structural or architectural design, detailed design, coding, and then
finally testing. Although the stratification of software production activities is very helpful,
notice what the effect has been on testing. Now it comes last (after the “interesting” part?),
and is no longer an integral part of the whole process. This is a significant change, and has
actually damaged the practice of testing and hence affected the quality of the software
produced in ways that are often not appreciated.

The problems with testing in the classic waterfall model are that testing is very much
product-based, and applied late in the development schedule. The levels of detail of test
activities are not acknowledged, and testing is now vulnerable to schedule pressure, since it
occurs last.

2.1.3 The V-Model

In the V-Model the test activities are spelled out to the same level of detail as the design
activities. Software is designed on the left-hand (downhill) part of the model, and built and
tested on the right-hand (uphill) part of the model. Note that different organizations may
have different names for the development and testing phases. We have used the names
given in the syllabus for the testing phases in our diagram.
The correspondences between the left and right hand activities are shown by the lines
across the middle of the V, showing the test levels from component testing at the bottom,
integration and system testing, and acceptance testing at the top level.
However, even the V-model is often not exploited to its full potential from a testing point of
view.

2.1.4 When are tests designed: As late as possible?

A common misconception is that tests should be designed as late as possible in the life
cycle, i.e. only just before they are needed. The reason for this is supposedly to save time
and effort, and to make progress as quickly as possible. But this is progress only from a
deadline point of view, not progress from a quality point of view, and the quality problems,
whose seeds are sown here, will come back to haunt the product later on.

No test can be designed without a specification, since the specification is the source of what
the correct results of the test should be. Even if that specification is not formally written
down or fully completed, the test design activity will reveal faults in whatever specification
the tests are based on. This applies to code, a part of the system, the system as a whole; or
the user’s view of the system.

If test design is left until the last possible moment, then the faults are found much later
when they are much more expensive to fix. In addition, the faults in the highest levels, the
requirements specification, are found last - these are also the .most critical and most
important faults. The actual effect of this approach is the most costly and time-consuming
approach to testing and software development.

2.1.5 When are tests designed: As early as possible!

If tests are going to be designed anyway, there is no additional effort required to move a
scheduled task to a different place in the schedule.

If tests are designed as early as possible, the inevitable effect of finding faults in the
specification comes early, when those faults are cheapest to fix. In addition, the most
significant faults are found first. This means that those faults are not built in to the next
stage, e.g. major requirement faults are not designed in, so faults are prevented.

An argument against this approach is that if the tests are already designed, they will need
to be maintained. There will be inevitable changes due to subsequent life cycle development
stages that will affect the earlier stages. This is correct, but the cost of maintaining tests
must be compared with the costs of the late testing approach~ not simply be accepted as
negating the good points. In fact, the extent of the test design detail should be determined
in part by the maintenance costs, so that less detail (but always some detail) should be
designed if extensive changes are anticipated.

One of the frequent headaches in software development is a rash of requirement change

requests that come from users very late in the life cycle; a major contributing cause for this
is the user acceptance test design process. When the users only begin to think about their
tests just before the acceptance tests are about to start, they realize the faults and
shortcomings in the requirement specification, and request changes to it. If they had
designed their tests at the same time as they were specifying those requirements, the very
mental activity of test design would have identified those faults before the system had built
them.

The way in which the system will be tested also serves to provide another dimension to the
development; the tests form part of the specification. If you know how it will be tested, you
are much more likely to build something that will pass those tests.
The end result of designing tests as early as possible is that quality is built in, costs are
reduced, and time is saved in test running because fewer faults are found, giving an overall
reduction in cost and effort. This is how testing activities help to build quality into the
software development process.

This can be taken one stage further as tecomxnended by a number of experts ~Beizer,
1990, Hetzel, 1991, Quentin, 1992), by designing tests. Before specifying what is to be
tested. The tests then act as a requirement for what will be built.

2.1.6 Verification and Validation

BS7925-l defines verification as “the process of evaluating a system or component to

determine whether the products of the given development phase satisfy the conditions
imposed at the start of that phase. The ‘conditions imposed at the start of that phase’ are
the key to understanding verification. These conditions should be generic in that they should
apply to any product of that phase and be used to ensure that the development phase has
worked well. They are checks on the quality of the process of producing a product such as
‘documentation must be unambiguous’, ‘document conforms to standard template’, and in
the case of an actual system, ‘has the system been assembled correctly’.

The full definition of the term ‘validation’ as given by BS7925-l is “The determination of the
correctness of the products of software development with respect to the user needs and
requirements”. The key to remembering validation is the phrase ‘with respect to the user
needs and requirements’. This means that the checks may be unique to a particular system
since different systems are developed to meet different user needs. (While this last
statement may be rather obvious it is worth stating when comparing validation with
verification.) While verification is more to do with the process of producing a system,
validation is more concerned with the products produced, i.e. the system itself.

Validation of each of the products of software development typically involves comparing one
product with its parent. For example, (using the terminology given in the V-Model of this
course) to validate a project specification we would compare it with the business
requirement specification. This involves checking completeness and consistency, for example
by checking that the project specification addresses all of the business requirements.

Validating requirements may seem a little tricky given that there is probably no higher level
specification. However, the validation activity is not limited to comparing one document
against another. User requirements can be validated by several other means such as
discussing them with end users and comparing them against your own or someone else’s
knowledge of the user’s business and working practices. Forms of documentation other than
a formal statement of requirements may be used such as contracts, memos or letters
describing individual or partial requirements. Reports of surveys, market research and user
group meetings may also provide a rich source of information against which a formal
requirements document can be validated. In fact many of these different approaches may
from time to time be applicable to the validation of any product of software development
(designs, source code, etc.).

A purpose of executing tests on the system is to ensure that the delivered system has the
functionality defined by the system specification. This best fits as a validation activity (since
it is checking that the system has the functions that are required — i.e. that it is the right
system). Verification at the system test phase is more to do with ensuring that a complete
system has been built. In terms of software this is rarely a large independent task rather it
is subsumed by the validation activities. However, if it were treated as an independent task
it would seek to ensure that the delivered system conforms to the standards defined for all
delivered systems. For example, all systems must include on-line help, display a copyright
notice at startup, conform to user interface standards, conform to product configuration
standards, etc.
Many people have trouble remembering which is which, and what they both mean. Barry
Boehm’s definitions represent a good way to remember them: Verification is “building the
system right”, and Validation is “building the right system”. Thus verification checks the
correctness of the results of one development stage with respect to some pre-defined rules
about what it should produce, while validation checks back against what the users really
want (or what they have specified).

2.1.7 The impact of early test design on development scheduling

Testers are often under the misconception that they are constrained by the order in which
software is built. The worst extreme is to have the last piece of software written be the one
that is needed to start test execution. However, with test design taking place early in the life
cycle, this need not be the case.

By designing the tests early, the order in which the system should ideally be put together
for testing is defined during the architectural or logical design stages. This means that the
order in which software is developed can be specified before it is built. This gives the
greatest opportunity for parallel testing and development activities, enabling development
time scales to be minimized. This can enable total test execution schedules to be shortened
and gives a more even distribution of test effort across the software development life cycle.

2.2 Economics of Testing

2.2.1 Testing is expensive?

We are constantly presented with the statement “testing is expensive” - but when we make
this statement what are we comparing the cost of testing with? If we compare the cost of
testing with the cost of the basic development effort testing may appear expensive.
However, this would be a false picture because the quality of the software that development
delivers has a dramatic impact on the effort required to test it. The more faults there are in
the software the longer testing will take since time will be spent reporting faults and re-
testing them.

Asking the cost of testing is actually the wrong thing to do. It is much more instructive to
ask the cost of not testing i.e. what have we saved the company by finding faults.

A development manager once said to a test manager “If it wasn’t for you we wouldn’t have
any bugs”. (Of cause he meant ‘faults’ not ‘bugs’ but he hadn’t been on this course!)
Another manager said, “Stop testing and you won’t raise any more faults”. Both of these
statements overlook the fact that the faults are already in the software by the time it is
handed over to testing. Testing does not insert the faults it merely reveals them. If they are
not revealed then they cannot be fixed and if they are not fixed they are likely to cause a
much higher cost once the faulty software is released to the end-users.

2.2.2 What do software faults cost?

The cost of faults escalates as we progress from one stage of the development life cycle to
the next. A requirement fault found during a review of the requirement specification will cost
very little to correct since the only thing that needs changing is the requirement
specification document. If a requirement fault is not found until system testing then the cost
of fixing it is much higher. The requirement specification will need to be changed together
with the functional and design specifications and the source code. After these changes some
component and integration testing will need to be repeated and finally some of the system
testing. If the requirement fault is not found until the system has been put into real use
then the cost is even higher since after being fixed and re-tested the new version of the
system will have to be shipped to all the end users affected by it.

Furthermore, faults that are found in the field (i.e. by end-users during real use of the
system) will cost the end-users time and effort. It may be that the fault makes the users’
work more difficult or perhaps impossible to do. The fault could cause a failure that corrupts
the users’ data and this in turn takes time and effort to repair.

The longer a specification fault remains undetected the more likely it is that it will cause
other faults because it may encourage false assumptions. In this way faults can be
multiplied so the cost of one particular fault can be considerably more than the cost of fixing
it.
The cost of testing is generally lower than the cost associated with major faults (such as
poor quality product and/or fixing faults) although few organizations have figures to confirm
this.

2.3 High Level Test Planning

Before planning the following should be set in place (taken from TMap a test management
method I introduced in “Structured Testing of Information Systems” by Martin Pol and Erik
van Veenendaal, I published by KiuWer, 1999). Organizational strategy - which does what.

• Identify people involved (all departments and interfaces involved in the process). This
will depend on your environment. In one organization it may be fairly static whereas
another may vary from project to project.
I • Examine the requirements; identify the test basis documents (i.e. the documents that
are to be used to derive test cases).
• Test organization, responsibilities, reporting lines.
Test deliverables, test plans, specifications, incident reports, summary report.
• Schedule and resources, people and machines.

2.3.1 Purpose

The purpose of high level test planning is to produce a high-level test plan! A high-level test
plan is synonymous with a project test plan and covers all levels of testing. It is a
management document describing the scope of the testing effort, resources required,
schedules, etc.

There is a standard for test documentation. It is ANSI/IEEE 829 “Standard for Software Test
Documentation”. This outlines a whole range of test documents including a test plan. It
describes the information that should be considered for inclusion in a test plan under 16
headings. These are described below.

2.3.2 Content of a high level Test Plan

1. Test Plan Identifier: Some unique reference for this document.

2. Introduction: A guide to what the test plan covers and references to other relevant
documents such as the Quality Assurance and Configuration Management plans.

3. Test Items: The physical things that are to be tested such as executable programs,
data files or databases. The version numbers of these, details of how they will be handed
over to testing (on disc, tape, across the network, etc.) and references to relevant
documentation.
4. Features to be Tested: The logical things that are to be tested, i.e. the functionality
and features.

5. Features not to be Tested: The logical things (functionality / features) that are not to
be tested.

6. Approach: The activities necessary to carry out the testing in sufficient detail to allow
the overall effort to be estimated. The techniques and tools that are to be used and the
completion criteria (such as coverage measures) and constraints such as environment
restrictions and staff availability.

7. Item Pass / Fail Criteria: For each test item the criteria for passing (or failing) that
item such as the number of known (and predicted) outstanding faults.

8. Suspension / Resumption Criteria: The criteria that will be used to determine when
(if) any testing activities should be suspended and resumed. For example, if too many faults
are found with the first few test cases it may be more cost effective to stop testing at the
current level and wait for the faults to be fixed.

9. Test Deliverables: What the testing processes should provide in terms of documents,
reports, etc.

10. Testing Tasks: Specific tasks, special skills required and the inter-dependencies.

11. Environment: Details of the hardware and software that will be needed in order to
execute the tests. Any other facilities (including office space and desks) that may be
required.

12. Responsibilities: Who is responsible for which activities and deliverables?

13. Staffing and Training: Needs Staff required and any training they will need such
as training on the system to be tested (so they can understand how to use it) training in the
business or training in testing techniques or tools.

14. Schedule: Milestones for delivery of software into testing, availability of the
environment and test deliverables.

15. Risks and Contingencies: what could go wrong and what will be done about it to
minimize adverse impacts if anything does go wrong.

16. Approvals: Names and when approved.

This is rather a lot to remember (though in practice you will be able to use the test
documentation standard-IEEE 829 as a checklist). To help you remember what is and is not
included in a test plan, consider the following table that maps all of the headings onto the
acronym SPACE DIRT.

Scope Test Items, Features to be tested, and Features not to be tested.

People Staffing and Training Needs, Schedule, and Responsibilities.

Approach Approach.

Criteria Item Pass/Fail Criteria, Suspension and Resumption Criteria.

Environment Environment.

Deliverables Deliverables
 Incidental Test Plan Identifier, Introduction, Approvals

Risks Risks and Contingencies

Tasks Testing Tasks

2.4 Component Testing

2.4.1 What is Component Testing?

BS792S-l defines a component as “A minimal software item for which a separate

specification is available”. Components are relatively small pieces of software that are, in
effect the building blocks from which the system is formed. They may also be referred to as
modules, units or programs and so this level of testing may also be known as module, unit
or program testing. For some organizations a component can be just a few lines of source
code while for others it can be a small program.

Component testing then is the lowest level of testing (i.e. it is at the bottom on the V-Model
software development life cycle). It is the first level of testing to start executing test cases
(but should be the last to specify test cases). It is the opportunity to test the software in
isolation and therefore in the greatest detail, looking at its functionality and structure, error
handling and interfaces.

Because it is just a component being tested, it is often necessary to have a test hamess or
driver to form an executable program that can be executed. This will usually have to be
developed in parallel with the component or may be created by adapting a driver for
another component. This should be kept as simple as possible to reduce the risk of faults in
the driver obscuring faults in the component being tested. Typically, drivers need to provide
a means of taking test input from the tester or a file, passing it on to the component,
receiving the output from the component and presenting it to the tester for comparison with
the expected outcome.

The programmer who wrote the code most often performs component testing. This is
sensible because it is the most economic approach. A programmer who executes test cases
on his or her own code can usually track down and fix any faults that may be revealed by
the tests relatively quickly. If someone else were to execute the test cases they may have to
document each failure. Eventually the programmer would come to investigate each of the
fault reports, perhaps having to reproduce them in order to determine their causes. Once
fixed, the fixed software would then be re-tested by this other person to confirm each fault
had indeed been fixed. This amounts to more effort and yet the same outcome: faults fixed.
Of course it is important that some independence is brought into the test specification
activity. The programmer should not be the only person to specify test cases (see Session 1
“Independence”).

Both functional and structural test case design techniques are appropriate though the extent
to which they are used should be defined during the test planning activity. This will depend
on the risks involved, for example, how important, critical or complex they are.

2.4.2 Component Test Strategy

The Software Component Testing Standard BS792S-2 requires that a Component Test
Strategy be documented before any of the component test process activities are carried out
(including the component test planning). The component test strategy should include the
following information:

• The test techniques that are to be used for component testing and the rationale for
 their choice;
• The completion criteria for component testing and the rationale for their choice
(typically these will be test coverage measures);
• The degree of independence required during the specification of test cases;
• The approach required (either isolation, top-down, bottom-up, or a combination of
these);
• The environment in which component tests are to be executed (including hardware
and software such as stubs, drivers and other software components);
• The test process to be used, detailing the activities to be performed and the inputs
and outputs of each activity (this must be consistent with the fundamental test
process).

The Component Test Strategy is not necessarily a whole document but could be a part of a
larger document such as a corporate or divisional Testing or Quality Manual. In such cases it
is likely to apply to a number of projects. However, it could be defined for one project and
form part of a specific project Quality Plan or be incorporated into the Project Component
Test Plan.

Exceptions: Any exceptions to this strategy must be documented in the relevant

Component Test Plan together with their justification. Exceptions do not need formal
approval but must be justified.

Design Techniques: The following techniques must be used for all components:
Equivalence Partitioning and Boundary Value Analysis. In addition, for high criticality
components, Decision Testing must also be used.

The rationale for the use of these techniques is that they have proven effective in the past
and are covered by the ISEB Software Testing Foundation Certificate syllabus. All testers on
this project are required to have attained ibis certificate. Decision Testing is more expensive
and therefore reserved for only the most critical components.

Completion Criteria: 100% coverage of valid equivalence partitions. 50% coverage of

valid boundary values providing no boundary faults are found, 100% coverage of valid
boundary values if one or more boundary faults are found. 30% coverage of all invalid
conditions (invalid equivalence partitions and invalid boundary values). For critical
components 100% Decision Coverage must also be achieved.

The rationale for these completion criteria is that 100% coverage of valid equivalence
partitions will ensure systematic coverage of the basic functionality of components is
exercised. The post-project review of component testing on the Warehouse project
recommended 50% coverage of valid boundaries providing no boundary faults are found as
an acceptable way to divert more testing effort onto the most critical components.

Independence Component Test Plans must be reviewed by at least one person other than
the developer/tester responsible for the components in question. Test Specifications must be
reviewed by at least two other people.

Approach: All critical components shall be tested in isolation using stubs and drivers in
place of interfacing components. Non critical components may be integrated using a
bottom-up integration strategy and drivers but the hierarchical depth of untested
components in any one baseline must not exceed three. All specified test cases of
components that are not concerned with an applications user interface shall be automated.

Environment: All automated component test cases shall be run in the standard component
test environment.

Process: The component test process to be used shall conform to the genenc component
test process defined in the Software Component Testing Standard B S792S-2: 1998.
2.4.3 Project Component Test Plan

The Software Component Testing Standard BS792S-2 requires that a Project Component
Test Plan be documented before any of the component test process activities are carried out
(including the component test planning).

The Project Component Test Plan specifies any changes for this project to the Component
Test Strategy and any dependencies between components that affect the order of
component testing. The order of component testing will be affected by the chosen approach
to component testing specified in the Component Test Strategy (isolation, top-down,
bottom-up, or a combination of these) and may also be influenced by overall project
management and work scheduling considerations. Strictly speaking, there are no
dependencies between component tests because all components are tested in isolation.
However, a desire to begin the integration of tested components before all component
testing is complete forces the sequence of component testing to be driven by the
requirements of integration testing in the small.

The Project Component Test Plan is not necessarily a whole document but could be a part of
a larger document such as an overall Project Test Plan.

Dependencies: The dependencies between components of different functional groups govern

the order in which the component tests for a functional group of components should be
performed. This order is shown below.

Graphics (GFX), File Access (FAC), Message Handling (MES), Argument Handling (ARG)

2.4.4 Component test process

The component test process follows the Fundamental Test Process described in Session 1.
The five activities are:

• Component Test Planning;

• Component Test Specification;
• Component Test Execution;
• Component Test Recording; and
• Checking for Component Test Completion.

The component test process always begins with Component Test Planning and ends with
Checking for Component Test Completion. Any and all of the activities may be repeated (or
at least revisited) since a number of iterations may be required before the completion
criteria defined during the Component Test Planning activity are met. One activity does not
have to be finished before another is started; later activities for one test case may occur
before earlier activities for another.

2.4.5 Test design techniques

The Software Component Testing Standard BS792S-2 defines a number of test design and
test measurement techniques that can be used for component testing. These include both
black box and white box techniques. The standard also allows other test design and test
measurement techniques to be defined so you do not have to restrict yourself to the
techniques defined by the standard in order to comply with it. Sessions 3 and 4 have more
details about the test design techniques.
2.5 Integration Testing in the Small

2.5.1 What is Integration Testing in the Small?

Integration testing in the small is bringing together individual components (modules/units)

that have already been tested in isolation. The objective is to test that the ‘set’ of
components function together correctly by concentrating on the interfaces between the
components. We are trying to find faults that couldn’t be found at an individual component
testing level. Although the interfaces should have been tested in component testing,
integration testing in the small makes sure that the things that are communicated are
correct from both sides, not just from one side of the interface. This is an important level of
testing but one that is sadly often overlooked.

As more and more components are combined together then a subsystem may be formed
which has more system-like functionality that can be tested. At this stage it may also be
useful to test nonfunctional aspects such as performance.

For integration testing in the small there are two choices that have to be made:
• How many components to combine in one go?
• In what order to combine components.

The decision over which choices to make are what is called the ‘integration strategy’. There
are two main integration strategies: Big Bang and incremental. These are described in
separate sections below.

2.5.2 Big Bang integration

“Big Bang” integration means putting together all of the components in one go. The
philosophy is that we have already tested all of the components so why not just throw them
all in together and test the lot? The reason normally given for this approach is that it saves
time - or does it?

If we encounter a problem it tends to be harder to locate and fix the faults. If the fault is
found and fixed then re-testing usually takes a lot longer. In the end the Big Bang strategy
does not work - it actually takes longer this way. This approach is based on the [mistaken]
assumption that there will be no faults.

2.5.3 Incremental integration

Incremental integration is where a small number of components are combined at once. At a

minimum, only one new component would be added to the baseline at each integration
step. This has the advantage of much easier fault location and fixing, as well as faster and
easier recovery if things do go badly wrong. (The finger of suspicion would point to the most
recent addition to the baseline.)
However, having decided to use an incremental approach to integration testing we have to
make a second choice: in what order to combine the components. This decision leads to
three different incremental integration strategies: top-down, bottom-up and functional
incrementation.

2.5.4 Top-down integration and Stubs

As its name implies, top-down integration combines components starting with the highest
levels of a hierarchy. Applying this strategy strictly, all components at a given level would be
integrated before any at the next level down would be added.
Because it starts from the top, there will be missing pieces of the hierarchy that have not
yet been integrated into a baseline. In order to test the partial system that comprises the
baseline, stubs are used to substitute for the missing components.

A stub replaces a called component in integration testing in the small. It is a small self-
contained program that may do no more than display its own name and then return. It is a
good idea to keep stubs as simple as possible; otherwise they may end up being as complex
as the components they are replacing.

As with all integration in the small strategies, there are advantages and disadvantages to
the approach. One advantage is that we are working to the same structure as the overall
system and this will be tested most often as we build each baseline. Senior Managers tend
to like this approach because the system can be demonstrated early (but beware that this
can often paint a false impression of the system’s readiness). There are some
disadvantages. Stubs are needed (as in all incremental integration strategies but this
perhaps needs more of them). Creating stubs means extra work though it should save more
effort in the long run. The details of the system are not tested until last and yet these may
be the most important parts of the software.

2.5.5 Bottom-up integration and Drivers

Bottom-up integration is the opposite of top-down. Applying it strictly, all components at the
lowest levels of the hierarchy would be integrated before any of the higher level ones.

Because the calling structure is missing, this strategy requires a way of activating the
baseline, e.g. by calling the component at the top of a baseline. These small programs are
called “drivers” because they drive the baseline. Drivers are also known as test harnesses or
scaffolding. They are usually specifically written for each baseline though there are a few
tools on the market which provide some ‘general purpose’ support. Bottom-up integration
may still need stubs as well though it is likely to use fewer of them.

2.5.6 Functional incrementation

The last integration strategy to be considered is what the syllabus refers to as “functional
incrementation”. We show two examples. Minimum capability is a functional integration
strategy because it is aiming to achieve a basic functionality working with a minimum
number of components integrated.
Thread integration is minimum capability with respect to time; the history or thread of
processing determines the minimum number of components to integrate together.

2.5.7 Integration guidelines

You will need to balance the advantages gained from adding small increments to your
baselines with the effort needed to make that approach work well. For example, if you are
spending more time writing stubs and drivers than you would have spent locating faults in a
larger baseline, and then you should consider having larger increments. However, for critical
components, adding only one component at a time would probably be best.

Keep stubs and drivers as simple as possible. If they are not written correctly they could
invalidate the testing performed.

If the planning for integration testing in the small is done at the right place in the life cycle,
i.e. on the left-hand side of the V-model before any code has been written, then the
integration order determines the order in which the components should be written by
developers. This can save significant time.

2.6 System Testing

System testing has two important aspects, which are distinguished in the syllabus:
functional system testing and non-functional system testing. The non-functional aspects are
often as important as the functional, but are generally less well specified and may therefore
be more difficult to test (but not impossible). If an organization has an independent test
group, it is usually at this level, i.e. it performs system testing.

2.6.1 Functional System Testing

Functional system testing gives us the first opportunity to test the system as a whole and is
in a sense the final baseline of integration testing in the small. Typically we are looking at
end to end functionality from two perspectives. One of these perspectives is based on the
functional requirements and is called requirement-based testing. The other perspective is
based on the business process and is called business process-based testing.

2.6.2 Requirements-based testing

Requirements-based testing uses a specification of the functional requirements for the

system as the basis for designing tests. A good way to start is to use the table of contents
of the requirement specification as an initial test inventory or list of items to test (or not to
test). We should also prioritize the requirements based on risk criteria (if this is not already
done in the specification) and use this to prioritize the tests. This will ensure that the most
important and most critical tests are included in the system testing effort.

2.6.3 Business process-based testing

Business process-based testing uses knowledge of the business profiles (or expected
business profiles). Business profiles describe the birth to death situations involved in the day
to day business use of the system. For example, a personnel and payroll system may have a
business profile along the lines of:
Someone joins company, he or she is paid on a regular basis, he or she leaves the company.

Another business process-based view is given by user profiles. User profiles describe how
much time users spend in different parts of the system. For example, consider a simple
bank system that has just three functions: account maintenance, account queries and
report generation. Users of this system might spend 50% of their time using this system
performing account queries, 40% of their time performing account maintenance and 10% of
their time generating reports. User profile testing would require that 50% of the testing
effort is spent testing account queries, 40% is spent testing account maintenance and 10%
is spent testing report generation.

Use cases are popular in object-oriented development. These are not the same as test
cases, since they tend to be a bit “woolly” but they form a useful basis for test cases from a
business perspective.
Note that we are still looking for faults in system testing, this time in end-to-end
functionality and in things that the system as a whole can do that could not be done by only
a partial baseline.

2.7 Non-Functional System Testing

2.7.1 Load, performance & stress testing

Performance tests include timing tests such as measuring response times to a PC over a
network, but may also include response times for performing a database back-up for
example.

Load tests, or capacity or volume tests are test designed to ensure that the system can
handle what has been specified, in terms of processing throughput, number of terminals
connected, etc.

Stress tests see what happens if we go beyond those limits.

2.7.2 Usability testing

Testing for usability is very important, but cannot be done well by technical people; it needs
to have input from real users.

2.7.3 Security testing

Whatever level of security is specified for the system must be tested, such as passwords,
level of authority, etc.

2.7.4 Configuration and Installation testing

There can be many different aspects to consider here. Different users may have different
hardware configurations such as amount of memory; they may have different software as
well, such as word processor versions or even games. If the system is supposed to work in
different configurations, it must be tested in all or at least a representative set of
configurations. For example, web sites should be tested with different browsers.

Upgrade paths also need to be tested; sometimes an upgrade of one part of the system can
be in conflict with other parts.
How will the new system or software be installed on user sites? The distribution mechanism
should be tested. The final intended environment may even have physical characteristics
that can influence the working of the system.

2.7.5 Reliability testing and other qualities

If a specification says “the system will be reliable”, this statement is untestable. Qualities
such as reliability, maintainability, portability, availability etc. need to be expressed in
measurable terms in order to be testable. Mean Time Between Failures (MTBF) is one way of
quantifying reliability.

A good way of specifying and testing for such qualities is found in Tom Gilb, Principles of
Software Engineering Management, Addison-Wesley, 1988, and is described in an optional
supplement to this course.

2.7.6 Back-up and Recovery testing

Testing recovery is more important that testing of back-ups; in fact, recovery is a test of the
back-up procedures. Recovery tests should be carried out at regular intervals so that the
procedures are rehearsed and somewhat familiar if they are ever needed for a real disaster.

2.7.7 Documentation testing

We produce documentation for two reasons: for users and for maintenance. Both types of
documents can be reviewed or inspected, but they should also be tested. A test of a user
manual is to give it to a potential end user who knows nothing about the system and see if
they can perform some standard tasks.

2.8 Integration Testing in the Large

2.8.1 What is Integration Testing in the Large?

This stage of testing is concerned with the testing of the system with other systems and
networks. There is an analogy of building a house - our finished house (system) now needs
to talk with the outside world:
• Our house needs electricity, gas, water, communications, TV etc to function properly.
So too does our system - it needs to interface with different networks and operating
systems and communications middleware.
• Our house needs to co-exist with other houses and blend in with the community - so
too does our system - it needs to sit alongside other systems such as billing, stock,
personnel systems etc;
• Our new system may need information from outside the organization such as interest
rates, foreign exchange etc and this is obtained via external data interchange (EDI).
A good example of EDI is the way in which our wages are transferred to our Bank
Accounts. Our house receives things from outside organizations such as the Post
Office or delivery trucks;
• Our new system may be required to work with different 3rd Party Packages - not
directly involved with the System under Test.

Different faults will be found during this level of testing and we must be prepared to plan
and execute such tests if they are considered vital for the success of our business. In reality
this level of testing will probably be done in conjunction with system testing rather than as a
separate testing stage. However it is now a visible testing stage and integration testing in
the large is an explicit testing phase in the syllabus.

In terms of planning - it should be planned the same way as Integration testing in the small
(i.e. testing interfaces/connections one at a time). This will reduce the risk of not being able
to locate the faults quickly. Like all testing stages we must identify the risks during the
planning phase - which areas would cause most severity if they were not to work? Perhaps
we are developing a piece of software that is to be used at a number of different locations
throughout the world - then testing the system within a Local Area Network (LAN) and
comparing with the response over a Wide Area Network (WAN) is essential.

When we plan Integration Testing in the Large there are a number of resources we might
need, such as different operating systems, different machine configurations and different
network configurations. These must all be thought through before the testing actually
commences. We must consider what machines we will need and it might be worthwhile
talking to some of the hardware manufacturers as they sometimes offer test sites with
different machine configurations set up.

2.9 Acceptance Testing

2.9.1 User Acceptance Testing

User Acceptance Testing is the final stage of validation. This is the time that customers get
their hands on the system (or should do) and the end product of this is usually a sign-off
from the users.

One of the problems is that this is rather late in the project for users to be involved - any
problems found now are too late to do anything about them. This is one reason why Rapid
Application Development (RAD) has become popular. Users are involved earlier and testing
is done earlier.

However, the users should have been involved in the test specification of the Acceptance
Tests at the start of the project. They should also have been involved in reviews throughout
the project, and there is nothing to say that they cannot be involved in helping to design
System and Integration tests. So there really should be no surprises!

The approach in this stage is a mixture of scripted and unscripted and the model office
concept is sometimes used. This is where a replica of the real environment is set up, for
example a branch office for a bank or building society.

2.9.2 Why users should be involved

It is the end users’ responsibility to perform acceptance testing. Sometimes the users are
tempted to say to the technical staff: “You know more about computers than we do, so you
do the acceptance testing for us”. This is like asking the used car salesman to take a test
drive for you!

The users bring the business perspective to the testing. They understand how the business
actually functions in all of its complexity. They will know of the special cases that always
seem to cause problems. They can also help to identify sensible work-arounds, and they
gain a detailed understanding of the system if they are involved in the acceptance testing.

The differences between system testing and acceptance testing are:

• Done by users, not technical staff;

• Focuses on building confidence rather than finding faults;
• Focuses on business-related cases rather than obscure error handling.

2.9.3 Contract acceptance testing

If a system is the subject of a legally binding contract, there may be aspects directly related
to the contract that need to be tested. It is important to ensure that the contractual
documents are kept up to date; otherwise you may be in breach of a contract while
delivering what the users want (instead of what they specified two years ago). However, it is
not fair for users to expect that the contract can be ignored, so the testing must be against
the contract and any agreed changes.

2.9.4 Alpha and Beta testing

Both alpha and beta testing are normally used by software houses that produce mass-
market shrink-wrapped software packages. This stage of testing is after system testing; it
may include elements of integration testing in the large. The alpha or beta testers are given
a pre-release version of the software and are asked to give feedback on the product. Alpha
and beta testing is done where there are no identifiable “end users” other than the general
public.

The difference between alpha and beta testing is where they are carried out. Alpha testing is
done on the development site - potential customers would be invited in to their offices. Beta
testing is done on customer sites - the software is sent out to them.

2.10 Maintenance testing

2.10.1What is Maintenance testing?

Maintenance Testing is all about preserving the quality we have already achieved. Once the
system is operational, then any further enhancements or fault fixes will be part of
the on-going maintenance of that system — the testing done for these changes to an
existing working system is maintenance testing.

Because the system is already there, when something is changed, there is a lot of the
system that should still work, so maintenance testing involves a lot of regression testing as
well as the testing of the changes.
It is worth noting that there is a different sequence with Maintenance Testing. In
development we start from small components and work up to the full system; in
maintenance testing, we can start from the top with the whole system. This means that we
can make sure that there is no effect on the whole system before testing the individual fix.
We also have different data - there is live data available in maintenance testing, whereas in
development testing we had to build the test data.

A breadth test is a shallow but broad test over the whole system, often used as a regression
suite. Depth tests explore specific areas such as changes and fixes. Impact analysis
investigates the likely effects of changes, so that the testing can be deeper in the riskier
areas.

2.10.2Poor or missing specifications

It is often argued that Maintenance Testing is the hardest type of testing to do because:

• there are no specifications;

• any documentation is out-of-date;
• lack of regression test scripts;
• knowledge base is limited due to age of the system (and programmers!).

If you do not have good specifications, it can be argued that you cannot test. The
specification is the oracle that tells the tester what the system should do.

So what do we do? Although this is a difficult situation, it is very common, and there are
ways to deal with it. Make contact with those who know the system, i.e. the users. Find out
from them what the system does do, if not what it should do. Anything that you do leam:
document. Document your assumptions as well so that other people have a better place to
start than you did. Track what it is costing the company in not having good, well maintained
specs

To find out what the system should do, you will need some form of oracle. This could be the
way the system works now - many Year 2000 tests used the current system as the oracle
for date-changed code. Another suggestion is to look in user manuals or guides (if they
exist). Finally, you may need to go back to the experts and “pick their brains”. You can
validate what is already there but not verify it (nothing to verify against).

3. Static Testing

This session looks at Static Testing techniques. These techniques are referred to as ‘static’
because the software is not executed; rather the specifications, documentation and source
code that comprise the software are examined in varying degrees of detail.

There are two basic types of static testing. One of these is people-based and the other is
tool-based. People-based techniques are generally known as “reviews” but there are a
variety of different ways in which reviews can be performed. The tool-based techniques
examine source code and are known as “static analysis”. Both of these basic types are
described in separate sections below.
3.1 What are Reviews?

“Reviews” is the generic name given to people-based static techniques. More or less any
activity that involves one or more people examining something could be called a review.
There are a variety of different ways in which reviews are carried out across different
organizations and in many cases within a single organization. Some are very formal, some
are very informal, and many lie somewhere between the two. The chances are that you
have been involved in reviews of one form or another.

One person can perform a review of his or her own work or of someone else’s work.
However, it is generally recognized that reviews performed by only one person are not as
effective as reviews conducted by a group of people all examining the same document (or
whatever it is that is being reviewed).

3.1.1 Review techniques for individuals

Desk checking and proof reading are two techniques that can be used by individuals to
review a document such as a specification or a piece of source code. They are basically the
same processes: the reviewer double-checks the document or source code on their own.
Data stepping is a slightly different process for reviewing source code: the reviewer follows
a set of data values through the source code to ensure that the values are correct at each
step of the processing.

3.1.2 Review techniques for groups

The static techniques that involve groups of people are generically referred to as reviews.
Reviews can vary a lot from very informal to highly formal, as will be discussed in more
detail shortly. Two examples of types of review are walkthroughs and Inspection. A
walkthrough is a form of review that is typically used to educate a group of people about a
technical document. Typically the author “walks” the group through the ideas to explain
them and so that the attendees understand the content. Inspection is the most formal of all
the formal review techniques. Its main focus during the process is to find faults, and it is the
most effective review technique in finding them (although the other types of review also find
some faults). Inspection is discussed in more detail below.

3.2 Reviews and the test process

3.2.1 Benefits of reviews

There are many benefits from reviews in general. They can improve software development
productivity and reduce development timescales. They can also reduce testing time and
cost. They can lead to lifetime cost reductions throughout the maintenance of a system over
its useful life. All this is achieved (where it is achieved) by finding and fixing faults in the
products of development phases before they are used in subsequent phases. In other
words, reviews find faults in specifications and other documents (including source code)
which can then be fixed before those specifications are used in the next phase of
development.
Reviews generally reduce fault levels and lead to increased quality. This can also result in
improved customer relations.

3.2.2 Reviews are cost-effective

There are a number of published figures to substantiate the cost-effectiveness of reviews.

Freedman and Weinberg quote a ten times reduction in faults that come into testing with a
50% to 80% reduction in testing cost. Yourdon in his book on Structured Walkthroughs
found that faults were reduced by a factor of ten. Gilb and Graham give a number of
documented benefits for software Inspection, including 25% reduction in schedules, a 28
times reduction in maintenance cost, and finding 80% of defects in a single pass (with a
mature Inspection process) and 95% in multiple passes.

3.2.3 What can be Inspected?

Anything written down can be inspected. Many people have the impression that Inspection
applies mainly to code (probably because Fagan’s original article was on “Design and code
inspection”). However, although Inspection is performed on code, it gives more value if it is
performed on more “upstream” documents in the software development process. It can be
applied to contracts, budgets, and even marketing material, as well as to policies,
strategies, business plans, user manuals, procedures and training material. Inspection also
applies to all types of system development documentation, such as requirements, feasibility
studies and designs. It is also very appropriate to apply to all types of test documentation
such as test plans, test designs and test cases. In fact even with Fagan’s original method, it
was found to be very effective applied to testware.

3.2.4 What can be reviewed?

Anything that can be inspected can also be reviewed, but reviews can apply to more things
than just those ideas that are written down. Reviews can be done on visions, strategic plans
and “big picture” ideas. Project progress can be reviewed to assess whether work is
proceeding according to the plans. A review is also the place where major decisions may be
made, for example about whether or not to develop a given feature.

Reviews and Inspections are complementary. Inspection excludes discussion and solution
optimizing, but these activities are often very important. Any type of review that tries to
combine more than one objective tends not to work as well as those with a single focus. It
works better to use Inspection to find faults and to use reviews to discuss, come to a
consensus and make decisions.

3.2.5 What to review I Inspect?

Looking at the ‘V’ life cycle diagram that was discussed in Session 2, reviews and
Inspections apply to everything on the left-hand side of the V-model. Note that the reviews
apply not only to the products of development but also to the test documentation that is
produced early in the life cycle. We have found that reviewing the business needs alongside
the Acceptance Tests works really well. It clarifies issues that might otherwise have been
overlooked. This is yet another way to find faults as early as possible in the life cycle so that
they can be removed at the least cost.

3.2.6 Costs of reviews

You cannot gain the benefits of reviews without investing in doing them, and this does have
a cost. As a rough guide, something between 5% and 15% of project effort would typically
be spent on reviews. If Inspections are being introduced into an organization, and then 15%
is a recommended guideline. Once the Inspection process is mature; this may go down to
around 5%. Note that 10% is half a day a week.
Remember that the cost of reviews always needs to be balanced against the cost of not
doing them, and finding the faults (which are already there) much later when it will be much
more expensive to fix them. The costs of reviews are mainly in people’s time, i.e. it is an
effort cost, but the cost varies depending on the type of review. The leader or moderator of
the review may need to spend time in planning the review (this would not be done for an
informal review, but is required for Inspection). The studying of the documents to be
reviewed by each participant on their own is normally the main cost (although in practice
this may not be done as thoroughly as it should). If a meeting is held, the cost is the length
of the meeting times the numbers of people present. The fixing of any faults found or the
resolution of issues found may or may not be followed up by the leader. In the more formal
review techniques, metrics or statistics are recorded and analyzed to ensure the continued
effectiveness and efficiency of the review process. Process improvement should also be a
part of any review process, so that lessons learned in a review can be folded back into
development and testing processes. (Inspection formally includes process improvement;
most other forms of review do not.)

3.3 Types of review

We have now established that reviews are an important part of software testing. Testers
should be involved in reviewing the development documents that tests are based on, and
should also review their own test documentation.
In this section, we will look at different types of reviews, and the activities that are done to
a greater or lesser extent in all of them. We will also look at the Inspection process in a bit
more detail, as it is the most effective of all review types.

3.3.1 Characteristics of different review types

3.3.1.1Informal review

As its name implies, this is very much an ad hoc process. Normally it simply consists of
someone giving their document to someone else and asking them to look it over. A
document may be distributed to a number of people, and the author of the document would
hope to receive back some helpful comments. It is a very cheap form of review because
there is no monitoring of metrics, no meeting and no follow-up. It is generally perceived to
be useful, and compared to not doing any reviews at all, it is. However, it is probably the
least effective form of review (although no one can prove that since no measurements are
ever done!)

3.3.1.2Technical review or Peer review

A technical review may have varying degrees of formality. This type of review does focus on
technical issues and technical documents. A peer review would exclude managers from the
review. The success of this type of review typically depends on the individuals involved -
they can be very effective and useful, but sometimes they are very wasteful (especially if
the meetings are not well disciplined), and can be rather subjective. Often this level of
review will have some documentation, even if just a list of issues raised. Sometimes metrics
will be kept. This type of review can find important faults, but can also be used to resolve
difficult technical problems, for example deciding on the best way to implement a design.

3.3.1.3Walkthrough

A walkthrough is typically led by the author of a document, for the purpose of educating the
participants about the content so that everyone understands the same thing. A walkthrough
may include “dry runs” of business scenarios to show how the system would handle certain
specific situations. For technical documents, it is often a peer group technique.

3.3.1.4Inspection
An Inspection is the most formal of the formal review techniques. There are strict entry and
exit criteria to the Inspection process, it is led by a trained Leader or moderator (not the
author), and there are defined roles for searching for faults based on defined rules and
checklists. Metrics are a required part of the process. More detail on Inspection is given later
in this session.

3.3.2 Characteristics of reviews in general

3.3.2.1Objectives and goals

The objectives and goals of reviews in general normally include the verification and
validation of documents against specifications and standards.
Some types of review also have an objective of achieving a consensus among the attendees
(but not Inspection).
Some types of review have process improvement as a goal (this is formally included in
Inspection).

3.3.2.2 Activities

There are a number of activities that may take place for any review.
The planning stage is part of all except informal reviews.
In Inspection (and sometimes technical or peer reviews), an overview or kickoff meeting is
held to put everyone “in the picture” about what is to be reviewed and how the review is to
be conducted. This premeeting may be a walkthrough in its own right.
The preparation or individual checking is usually where the greatest value is gained from a
review process. Each person spends time on the review document (and related documents),
becoming familiar with it and/or looking for faults. In some reviews, this part of the process
is optional (at least in practice). In Inspection it is required.
Most reviews include a meeting of the reviewers. Informal reviews probably do not, and
Inspection does not hold a meeting if it would not add economic value to the process.
Sometimes the meeting time is the only time people actually look at the document.
Sometimes the meetings run on for hours and discuss trivial issues. The best reviews (of
any level of formality) ensure that value is gained from the process, including the meeting.

The more formal review techniques include follow-up of the faults or issues found to ensure
that action has been taken on everything raised (Inspection does, as do some forms of
technical or peer review).

The more formal review techniques collect metrics on cost (time spent) and benefits
achieved.

3.3.2.3Roles and responsibilities

For any of the formal reviews (i.e. not informal reviews), there is someone responsible for
the review of a document (the individual review cycle). This may be the author of the
document (walkthrough) or an independent Leader or moderator (formal reviews and
Inspection). The responsibility of the Leader is to ensure that the review process works. He
or she may distribute documents, choose reviewers, mentor the reviewers, call and read the
meeting, perform follow-up and record relevant metrics.

The author of the document being reviewed or inspected is generally included in the review,
although there are some variants that exclude the author. The author actually has the most
to gain from the review in terms of learning how to do their work better (if the review is
conducted in the right spirit!).
The reviewers or Inspectors are the people who bring the added value to the process by
helping the author to improve his or her document. In some types of review, individual
checkers are given specific types of fault to look for to make the process more effective.

Managers have an important role to play in reviews. Even if they are excluded from some
types of peer review, they can (and should) review management level documents with their
peers. They also need to understand the economics of reviews and the value that they
bring. They need to ensure that the reviews are done properly, i.e. that adequate time is
allowed for reviews in project schedules.

There may be other roles in addition to these, for example an organization-wide coordinator
who would keep and monitor metrics, or someone to “own” the review process itself - this
person would be responsible for updating forms, checklists, etc.

3.3.2.4Deliverables

The main deliverable from a review is the changes to the document that was reviewed. The
author of the document normally edits these. For Inspection, the changes must conform to
the accepted rules for the document. In some types of review, the reviewers suggest
improvements to the document itself. Generally the author can either accept or reject the
changes suggested.

If the author does not have the authority to change a related document (e.g. if the review
found that a correct design conflicted with an incorrect requirement specification), then a
change request may be raised to change the other document(s).
For Inspection and possibly other types of review, process improvement suggestions are a
deliverable. This includes improvements to the review or Inspection process itself and also
improvements to the development process that produced the document just reviewed.
(Note that these are improvements to processes, not to reviewed documents.)

The final deliverable (for the more formal types of review, including Inspection) is the
metrics about the costs, faults found, and benefits achieved by the review or Inspection
process.

3.3.2.5 Pitfalls

Reviews are not always successful. They are sometimes not very effective, so faults that
could have been found slip through the net. They are sometimes very inefficient, so that
people feel that they are wasting their time. Often insufficient thought has gone into the
definition of the review process itself - it just evolves over time.

One of the most common causes for poor quality in the review process is lack of training,
and this is more critical the more formal the review.

Another problem with reviews is having to deal with documents that are of poor quality.
Entry criteria to the review or Inspection process can ensure that reviewers’ time is not
wasted on documents that are not worthy of the review effort.

A lack of management support is a frequent problem. If managers say that they want
reviews to take place but don’t allow any time in the schedules for the, this is only “lip
service” not commitment to quality.
Long-term, it can be disheartening to become expert at detecting faults if the same faults
keep on being injected into all newly written documents. Process improvements are the key
to long-term effectiveness and efficiency.
3.3.3 Inspection

3.3.3.1Typical reviews versus Inspection

There are a number of differences between the way most people practice reviews and the
Inspection process as described in Software Inspection by Gilb and Graham, Addison-
Wesley, 1993.

In a typical review, the document is given out in advance, there are typically dozens of
pages to review, and the instructions are simply “Please review this.”

In Inspection, it is not just the document under review that is given out in advance, but also
source or predecessor documents. The number of pages to focus the Inspection on is closely
controlled, so that Inspectors (checkers) check a limited area in depth - a chunk or sample
of the whole document. The instructions given to checkers are designed so that each
individual checker will find the maximum number of unique faults. Special defect-hunting
roles are defined, and Inspectors are trained in how to be most effective at finding faults.

In typical reviews, sometimes the reviewers have time to look through the document before
the meeting, and some do not. The meeting is often difficult to arrange and may last for
hours.

In Inspection, it is an entry criterion to the meeting that each checker has done the
individual checking. The meeting is highly focused and efficient. If it is not economic, then a
meeting may not be held at all, and it is limited to two hours.

In a typical review, there is often a lot of discussion, some about technical issues but much
about trivia (details/finer parts). Comments are often mainly subjective, along the lines of “I
don’t like the way you did this” or “Why didn’t you do it this way?”

In Inspection, the process is objective. The only thing that is permissible to raise as an issue
is a potential violation of an agreed Rule (the Rule sets are what the document should
conform to). Discussion is severely curtailed in an Inspection meeting or postponed until the
end. The Leader’s role is very important to keep the meetings on track and focused and to
keep pulling people away from trivia and pointless discussion.
Many people keep on doing reviews even if they don’t know whether it is worthwhile or not.
Every activity in the Inspection process is done only if its economic value is continuously
proven.
3.3.3.2 Inspection is more

Inspection contains many mechanisms that are additional to those found in other formal
reviews. These include the following:

• Entry criteria, to ensure that we don’t waste time inspecting an unworthy document;

• Training for maximum effectiveness and efficiency;

• Optimum checking rate to get the greatest value out of the time spent by looking deep;

• Prioritizing the words: Inspect the most important documents and their most important
parts;

• Standards are used in the Inspection process (there are a number of Inspection
standards also);
• The Leader checks that editing has been completed (follow-up);

• Process improvement is built in to the Inspection process;

• Exit criteria ensure that the document is worthy of being used “downstream” and that
the Inspection process was carried out correctly.

One of the most powerful exit criteria is the quantified estimate of the remaining defects per
page. This may be say 3 per page initially, but can be brought down by orders of magnitude
over time.

3.3.3.3 Inspection is better

Typical reviews are probably only 10% to 20% effective at detecting existing faults. The
return on investment is usually not known because no one keeps track even of their cost.
When Inspection is still being learned, its effectiveness is around 30% to 40% (this is
demonstrated in Inspection training courses). Once Inspection is well established and
mature, this process can find up to 80% of faults in a single pass, 95% in multiple passes.
The return on investment ranges from 6 hours to 30 hours for every hour spent.

3.3.3.4The Inspection process

The diagram shows a product document infected with faults. The document must pass
through the entry gate before it is allowed to start the Inspection process. The Inspection
Leader performs the planning activities. A Kickoff meeting is held to “set the scene” about
the documents and the process.

The Individual Checking is where most of the benefits are gained. 80% or more of the faults
found will be found in this stage.

A meeting is held (if economic). The editing of the document is done by the author or the
person now responsible for the document. This involves redoing some of the activities that
produced the document initially, and it also may require Change Requests to documents not
under the control of the editor. Process improvement suggestions may be raised at any
time, for improvements either to the Inspection process or to the development process.

The document must pass through the Exit gate before it is allowed to leave the Inspection
process. There are two aspects to investigate here: is the product document now ready
(e.g. has some action been taken on all issues logged), and was the Inspection process
carried out properly? For example, if the checking rate was too fast, then the checking has
not been done properly.

A gleaming new improved document is the result of the process, but there is still a “blob” on
it. It is not economic to be 100% effective in Inspection. At least with Inspection you
consciously predict the levels of remaining faults rather than fallaciously assuming that we
have found them all!

3.3.3.5 How the checking rate enables deep checking in Inspection

There is a dramatic difference of Inspection to normal reviews, and that is in the depth of
checking. This is illustrated by the picture of a document. Initially there are no faults visible.
Typically in reviews, the time and size of document determine the checking rate. So for
example if you have 2 hours available for a review and the document is 100 pages long,
then the checking rate will be 50 pages per hour. (Any two of these three factors determine
the third.)
This is equivalent to “skimming the surface” of the document. We will find some faults - in
this example we have found one major and two minor faults. Our typical reaction is now to
think: “This review was worthwhile wasn’t it - it found a major fault. Now we can fix that
and the two other minor faults, and the document will now be OK.” Think: are we missing
anything here?
Inspection is different. We do not take any more time, but it is the optimum rate for the
type of document that is used to determine the size of the document that will be checked in
detail. So if the optimum rate is one page per hour and we have two hours, then the size of
the sample or chunk will be 2 pages. (Note that the optimum rate needs to be established
over time for different types of document and will depend on a number of factors, and it is
based on prioritized words (logical page rather than physical page). Of course it doesn’t take
an hour just to read a single page, but the checking done in Inspection includes comparing
each paragraph or sentence on the target page with all source documents, checking each
paragraph or phrase against relevant rule sets, both generic and specific, working through
checklists for different role assignments, as well as the time to read around the target page
to set the context. If checking is done to this level of thoroughness, it is not at all difficult to
spend an hour on one page!)

How does this depth-oriented approach affect the faults found? On the picture, we have
gone deep in the

Inspection on a limited number of pages. We have found the major one found in the other
review plus two (other) minors, but we have also found a deep-seated major fault, which we
would never have seen or even suspected if we had not spent the time to go deep. There is
no guarantee that the most dangerous faults are I lying near the surface!

When the author comes to fix this deep-seated fault, he or she can look through the rest of
the document for similar faults, and all of them can then be corrected. So in this example
we will have corrected 5 major I faults instead of one. This gives tremendous leverage to
the Inspection process - you can fix faults you didn’t find!

3.3.3.6Inspection surprises

To summarize the Inspection process, there are a number of things about Inspection which
surprise people. The fundamental importance of the Rules is what makes Inspection
objective rather than a subjective review. The Rules are democratically agreed as applying
(this helping to defuse author defensiveness) and by definition a fault is a Rule violation.

The slow checking rates are surprising, but the value to be gained by depth gives far greater
long-term gains than surface—skimming review that miss major deep—seated problems.

The strict entry and exit criteria help to ensure that Inspection gives value for money.

The logging rates are much faster than in typical reviews (30 to 60 seconds; typical reviews
log one thing in 3 to 10 minutes). This ensures that the meeting is very efficient. One
reason this works is that the final responsibility for all changes is fully given to the author,
who has total responsibility for final classification of faults as well as the content of all fixes.

More information on Inspection can be found in the book Software Inspection, Tom Gilb and
Dorothy I Graham, Addison-Wesley, 1993, ISBN 0-201-63181-4.

3.4 Static analysis

3.4.1.1What can static analysis do?

Static analysis is a form of automated testing. It can check for violations of standards and
can find things that may or may not be faults. Static analysis is descended from compiler
technology. In fact, many compilers may have static analysis facilities available for
developers to use if they wish. There are also a number of stand-alone static analysis tools
for various different computer programming languages. Like a compiler, the static analysis
tool analyses the code without executing it, and can alert the developer to various things
such as unreachable code, undeclared variables, etc.

Static analysis tools can also compute various metrics about code such as cyclomatic
complexity.

3.4.1.2Data flow analysis

Data flow analysis is the study of program variables. A variable is basically a location in the
computer’s memory that has a name so that the programmer can refer to it more
conveniently in the source code. When a value is put into this location, we say that the
variable is “defined”. When that value is accessed, we say that it is “used”.

For example, in the statement “x = y + z”, the variables y and z are used because the
values that they contain are being accessed and added together. The result of this addition
is then put into the memory location called “x”, so x is defined.

The significance of this is that static analysis tools can perform a number of simple checks.
One of these checks is to ensure that every variable is defined before it is used. If a variable
is not defined before it is used, the value that it contains may be different every time the
program is executed and in any case is unlikely to contain the correct value. This is an
example of a data flow error (it should be called a “fault” but the tools refer to them as
“data flow errors”). Another check that a static analysis tool can make is to ensure that
every time a variable is defined it is used somewhere later on in the program. If it isn’t,
then why was it defined in the first place? This is known as a data flow anomaly and
although it can be a perfectly harmless fault, it can also indicate that something more
serious is wrong.

3.4 1.3 Control flow analysis

Control flow analysis can find infinite loops, inaccessible code, and many other suspicious
aspects. However, not all of the things found are necessarily faults; defensive programming
may result in code that is technically unreachable.

3.4.1.4Cyclomatic complexity

Cyclomatic complexity is related to the number of decisions in a program or control flow

graph. The easiest way to compute it is to count the number of decisions (diamond-shaped
boxes) on a control flow graph and add 1. Working from code, count the total number of IF’s
and any loop constructs (DO, FOR, WHILE, REPEAT) and add 1. The cyclomatic complexity
does reflect to some extent how complex a code fragment is, but it is not the whole story.

3.4.1.5 Other static metrics

Lines of code (LOC or KLOC for 1000’s of LOC) are a measure of the size of a code module.
A very detailed measurement devised by Halstead counts operators (such as “+“, “-“,
multiply and divide, but this measure is not much used now. Fan-in is related to the number
of modules that call (in to) a given module. Modules with high fan-in are found at the
bottom of hierarchies, or in libraries where they are frequently called. Modules with high
fan-out are typically at the top of hierarchies, because they call out to many modules (e.g.
the main menu). Any module with both high fan-in and high fan-out probably needs re-
designing.

Nesting levels relate to how deeply nested statements are within other IF statements. This
is a good metric to have in addition to cyclomatic complexity, since highly nested code is
harder to understand than linear code, but cyclomatic complexity does not distinguish them.
Other metrics include the number of function calls and a number of metrics specific to
object-oriented code.

3.4.1.6 Limitations and advantages

Static analysis has its limitations. It cannot distinguish “fail-safe” code from real faults or
anomalies, and may create a lot of spurious failure messages. Static analysis tools do not
execute the code, so they are not a substitute for dynamic testing, and they are not related
to real operating conditions.

However, static analysis tools can find faults that are difficult to see and they give objective
quality information about the code. We feel that all developers should use static analysis
tools, since the information they can give can find faults very early when they are very
cheap to fix.

4. Dynamic Testing Techniques

4.1 About Testing Techniques

4.1.1 The need for testing techniques

In Session 1 we explained that testing everything is known as exhaustive testing (defined as

exercising every combination of inputs and preconditions) and demonstrated that it is an
impractical goal. Therefore, as we cannot test everything we have to select a subset of all
possible tests. In practice the subset we select is a very tiny subset and yet it has to have a
high probability of finding most of the faults in a system.

Experience and experiments have shown us that selecting a subset at random is neither
very effective nor very efficient (even if it is tool supported). We have to select tests using
some intelligent thought process. Test techniques are such thought processes.

4.1.2 What is a testing technique?

A testing technique is a thought process that helps us select a good set of tests from the
total number of all possible tests for a given system. Different techniques offer different
ways of looking at the software under test, possibly challenging assumptions made about it.
Each technique provides a set of rules or guidelines for the tester to follow in identifying test
conditions and test cases. They are based on either a behavioral or a structural model of the
system In other words, they are based on an understanding of the system’s behavior
(functions and non-functional attributes such as performance or ease of use - what the
system does) or its structure, how it does it.

There are a lot of different testing techniques and those that have been published have
been found to be successful at identifying tests that find faults. The use of testing
techniques is ‘best practice’ though they should not be used to the exclusion of any other-
approach.
Put simply, a testing technique is a means of identifying good tests. Recall from Session 1
that a good test case is four things:
• effective - has potential to find faults;
• exemplary- represents other test cases;
• evolvable - easy to maintain;
• economic - doesn’t cost much to use.

4.1.3 Advantages of Techniques

Different people using the same technique on the same system will almost certainly arrive
at different test cases but they will have a similar probability of finding faults. This is
because the technique will guide them into having a similar or the same view of the system
and to make similar or the same assumptions.

Using a technique also gives the tester some independence of thought. Developers who
apply test techniques may find faults that they would not have found if they had only tested
intuitively.

Using techniques makes testing more effective

This means that more faults will be found with less effort. Because a technique focuses on a
particular type of fault it becomes more likely that the tests will find more of that type of
fault. By selecting appropriate testing techniques it is possible to control more accurately
what is being tested and so reduce the chance of overlap between different test cases. You
also have more confidence that you are testing the things that are most in need of being
tested.

Using techniques makes testing more efficient

This means that faults will be found with less effort. Given two sets of tests that find the
same faults in the same system, if one set costs 25% less to produce then this is better
even though they both find the same faults. Good testing not only maximizes effectiveness
but also maximizes efficiency, i.e. minimizes cost.

Systematic techniques are measurable, meaning that it is possible to quantify the extent of
their use making it possible to gain an objective assessment of the thoroughness of testing
with respect to the use of each testing technique. This is useful for comparison of one test
effort to another and for providing confidence in the adequacy of testing.

4.2 Black and White Box Testing

4.2.1 Types of Testing Technique

There are many different types of software testing technique, each with its own strengths
and weaknesses. Each individual technique is good at finding particular types of fault and
relatively poor at finding other types. For example, a technique that explores the upper and
lower limits of a single input range is more likely to find boundary value faults than faults
associated with combinations of inputs. Similarly, testing performed at different stages in
the software development life cycle is going to find different types of faults; component
testing is more likely to find coding faults than system design faults.

Each testing technique falls into one of a number of different categories. Broadly speaking
there are two main categories, static and dynamic. However, dynamic techniques are
subdivided into two more categories behavioral (black box) and structural (white box).
Behavioral techniques can be further subdivided into functional and non-functional
techniques. Each of these is summarized below.
4.2.2 Static Testing Techniques

As we saw in Session 3, static testing techniques do not execute the code being examined,
and are generally used before any tests are executed on the software. They could be called
non-execution techniques. Most static testing techniques can be used to ‘test’ any form of
document including source code, design, functional and requirement specifications.
However, ‘static analysis’ is a tool supported version that concentrates on testing formal
languages and so is most often used to statically ‘test’ source code.

4.2.3 Functional Testing Techniques (Black Box)

Functional testing techniques are also known as ‘black-box’ and input / output-driven testing
techniques because they view the software as a black box with inputs and outputs, but have
no knowledge of how it is structured inside the box. In essence, the tester is concentrating
on the function of the software, that is, what it does, not how it does it.

4.2.4 Structural Testing Techniques (White Box)

Structural testing techniques use the internal structure of the software to derive test cases.
They are commonly called ‘white-box’ or ‘glass-box’ techniques (implying you can see into
the system) since they require knowledge of how the software is implemented, that is, how
it works. For example, a structural technique may be concerned with exercising loops in the
software. Different test cases may be derived to exercise the loop once, twice, and many
times. This may be done regardless of the functionality of the software.

4.2.5 Non-Functional Testing Techniques

Non-functional aspects of a system (also known as ‘quality’ aspects) include performance,

usability portability, maintainability, etc. This category of technique is concerned with
examining how well the system does something, not what it does or how it does it.
Techniques to test these non-functional aspects are less procedural and less formalized than
those of other categories as the actual tests are more dependent on the type of system,
what it does and the resources available for the tests.

How to specify non-functional tests is outside the scope of the syllabus for this course but
an approach to doing so is outlined in the supplementary section at the back of notes. The
approach uses quality attribute templates, a technique from Tom Gilb’s book “Principles of
Software Engineering Management”, Addison-Wesley, 1988.

Non-functional testing at system level is part of the syllabus and was covered in Session 2
(but techniques for deriving non-functional tests are not covered).

4.2.6 Black Box versus White Box

Black box techniques are appropriate at all stages of testing (Component Testing through to
User Acceptance Testing). While individual components form part of the structure of a
system, when performing Component Testing it is possible to view the component itself as a
black box, that is, design test cases based on its functionality without regard for its
structure. Similarly, white box techniques can be used at all stages of testing but are
typically used most predominately at Component and Integration Testing in the Small.
4.3 Black Box Test Techniques

4.3.1 Techniques Defined in BS 7925-2

The Software Component Testing Standard BS792S-2 defines the following black-box testing
techniques:
• Equivalence Partitioning;
• Boundary Value Analysis;
• State Transition Testing;
• Cause-Effect Graphing;
• Syntax Testing;
• Random Testing.

The standard also defines how other techniques can be specified. This is important since it
means that anyone wishing to conform to the Software Component Testing Standard is not
restricted to using the techniques that the standard defines.

4.3.2 Equivalence Partitioning & Boundary Value Analysis

4.3.2.1 Equivalence partitioning

Equivalence Partitioning is a good all-round functional black-box technique. It can be applied

at any level of testing and is often a good technique to use first. It is a common sense
approach to testing, so much so that most testers practice it informally even though they
may not realize it. However, while it is better to use the technique informally than not at all,
it is much better to use the technique in a formal way to attain the full benefits that it can
deliver.

The idea behind the technique is to divide or partition a set of test conditions into groups or
sets that can be considered the same or equivalent, hence ‘equivalence partitioning’.
Equivalence partitions are also known as equivalence classes, the two terms mean exactly
the same thing.

The benefit of doing this is that we need test only one condition from each partition. This is
because we are assuming that all the conditions in one partition will be treated in the same
way by the software. If one condition in a partition works, we assume all of the conditions in
that partition will work and so there is no point in testing any of these others. Conversely, if
one of the conditions in a partition does not work, then we assume that none of the
conditions in that partition will work so again there is no point in testing any more in that
partition. Of course these are simplifying assumptions that may not always be right but
writing them down at least gives other the chance to challenge the assumptions and
hopefully help identify more accurate equivalence partitions.

For example, a savings account in a bank earns a different rate of interest depending on the
balance in the account. In order to test the software that calculates the interest due we can
identify the ranges of balance values that each earns a different rate of interest. For
example, if a balance in the range £0 to £100 has a 3% interest rate, a balance between
£100 and £1,000 has a 5% interest rate, and balances of £1,000 and over have a 7%
interest rate, we would initially identify three equivalence partitions: £0 - £100, £100.01 -
£999.99, and £1,000 and above. When designing the test cases for this software we would
ensure that these three equivalence partitions were each covered once. So we might choose
to calculate the interest on balances of £50, £260 and £1,348. Had we not have identified
these partitions it is possible that at least one of them could have been missed at the
expense of testing another one several times over (such as with the balances of £30, £140,
£250, and £400).
4.3.2.2 Boundary value analysis

Boundary value analysis is based on testing on and around the boundaries between
partitions. If you have done “range checking”, you were probably using the boundary value
analysis technique, even if you weren’t aware of it. Note that we have both valid boundaries
(in the valid partitions) and invalid boundaries (in the invalid partitions).

4.3.2.3 Design Test Cases

Having identified the conditions that you wish to test, the next step is to design the test
cases. The more test conditions that can be covered in a single test case, the fewer the test
cases that are needed.
Generally, each test case for invalid conditions should cover only one condition. This is
because programs typically stop processing input as soon as they encounter the first fault.
However, if it is known that the software under test is required to process all input
regardless of its validity it is sensible to continue as before and design test cases that cover
as many invalid conditions in one go as possible. In either case, there should be separate
test cases covering valid and invalid conditions. The test cases to cover the boundary
conditions are done in a similar way.

4.3.2.4 Why do both EP and BVA?

Technically, because every boundary is in some partition, if you did only boundary value
analysis (BVA) you would also have tested every equivalence partition (EP). However this
approach will cause problems when the value fails — was it only the boundary value that
failed or did the whole partition fail? Also by testing only boundaries we would probably not
give the users too much confidence as we are using extreme values rather than normal
values. The boundaries may be more difficult (and therefore more costly) to set up as well.

We recommend that you test the partitions separately from boundaries - this means
choosing partition values that are NOT boundary values.

What partitions and boundaries you exercise and which first depends on your objectives. If
your goal is the most thorough approach, then follow the traditional approach and test valid
partitions, then invalid partitions, then valid boundaries and finally invalid boundaries.
However if you are under time pressure and cannot test everything (and who isn’t), then
your objective will help you decide what to test. If you are after user confidence with
minimum tests, you may do valid partitions only. If you want to find as many faults as
possible as quickly as possible, you may start with invalid boundaries.

4.3.3 State Transition Testing

This technique is not specifically required by the ISEB syllabus, although we do need to
cover one additional black box technique. Therefore the details of this technique should not
be in the exam, but only an awareness of another functional technique.

State transition testing is used where some aspect of the system can be described in what is
called a “finite state machine”. This simply means that the system can be in a (finite)
number of different states, and the transitions from one state to another are determined by
the rules of the “machine”. This is the model on which the system and the tests are based.
Any system where you get a different output for the same input, depending on what has
happened before, is a finite state system. For example, if you request to withdraw £100
from a bank ATM, you may be given cash. Later you may make exactly the same request
but be refused the money (because your balance is insufficient). This later refusal is
because the state of your bank account had changed from having sufficient funds to cover
the withdrawal to having insufficient funds. The transaction that caused your account to
change its state was probably the earlier withdrawal. Another example is a word processor.
If a document is open, you are able to Close it. If no document is open, then “Close” is not
available. After you choose “Close” once, you cannot choose it again for the same document
unless you open that document. A document thus has two states: open and closed.

A state transition model has four basic parts:

• The states that the software may occupy (open/closed or funded/insufficient funds);
• The transitions from one state to another (not all transitions are allowed);
• The events that cause a transition (withdrawing money, closing a file);
• The actions that result from a transition (an error message, or being given your cash).

Note that a transition does not need to change to a different state; it could stay in the same
state. In fact, trying to input an invalid input would be likely to produce an error message as
the action, but the transition would be back to the same state the system was in before.

Deriving test cases from the state transition model is a black box approach. Measuring how
much you have tested (covered) is getting close to a white box perspective. However, state
transition testing is generally regarded as a black box technique.

You can design tests to test every transition shown in the model. If every (valid) transition
is tested, this is known as “0-switch” coverage. You could also test a series of transitions
through more than one state. If you covered all of the pairs of two valid transitions, you
would have “1-switch” coverage, covering the sets of 3 transitions would give “2-switch”
coverage, etc.

However, deriving tests only from the model may omit the negative tests, where we could
try to generate invalid transitions. In order to see the total number of combinations of
states and transitions, both valid and invalid, a state table can be used.

4.4 Non-Systematic Techniques

Although it is true that testing should be rigorous, thorough and systematic, this is not all
there is to testing. There is a definite role for non-systematic techniques.

4.4.1 Trail and Error / Ad Hoc

Many people confuse ad hoc testing with error guessing. Ad hoc testing is unplanned and
usually done before rigorous testing. It is good as a quick ‘test readiness’ assessment. In
other words, perform a few things on the latest version of software to determine if it ready
for more thorough testing.

4.4.2 Error-Guessing

Error guessing is a technique that should always be used after other more formal techniques
have been applied. The success of error guessing is very much dependent on the skill of the
tester, as good testers know where the faults are most likely to lurk. Some people seem to
be naturally good at testing and others are good testers because they have a lot of
experience either as a tester or working with a particular system and so are able to pin
point its weaknesses. This why an error guessing approach is best used after more formal
techniques have been applied. In using other techniques the tester is likely to gain a better
understanding of the system, what it does and how it works. With a better understanding
anyone is likely to be more able to think of ways in which the system may not work
properly.

There are no rules for error guessing. The tester is encouraged to think of situations in
which the software may not be able to cope. Typical conditions to try include divide by zero,
blank (or no) input, empty files and the wrong kind of data (e.g. alphabetic characters
where numeric are required). If anyone ever says of a system or the environment in which
it is to operate “That could never happen”, it might be a good idea to test that condition as
such assumptions about what will and will not happen in the live environment are often the
cause of failures.

Error guessing is sometimes known by other names including experience-driven testing and
heuristic testing.

4.4.3 Lateral Testing

Lateral testing is an approach to non-systematic testing that uses lateral thinking

techniques. Lateral thinking employs alternative methods of thinking. Rather than progress
in logical manner from one thought to another, the aim is to make a leap into very different
areas of thought.

4.4.4 User Testing

Asking users to “try out” the new software is a form of non-systematic testing. This should
only be done once the software is known to be reasonably stable, otherwise the ~users will
lose confidence in the software.

4.4.5 Unscripted Testing

Unscripted testing means testing without a specification of pre-planned tests.

4.5 White Box Test Techniques

White box techniques are normally used after an initial set of tests has been derived using
black box techniques. They are most often used to measure “coverage” - how much of the
structure has been exercised or covered by a set of tests.
Coverage measurement is best done using tools, and there are a number of such tools on
the market. These tools can help to increase productivity and quality. They increase quality
by ensuring that more structural aspects are tested, so faults on those structural paths can
be found. They increase productivity and efficiency by highlighting tests that may be
redundant, i.e. testing the same structure as other tests (although this is not necessarily a
bad thing!)

4.5.1 What are Coverage Techniques?

Coverage techniques serve two purposes: test measurement and test case design. They are
often used in the first instance to assess the amount of testing performed by tests derived
from functional techniques. They are then used to design additional tests with the aim of
increasing the test coverage.
Coverage techniques are a good way of generating additional test cases that are different
from existing tests and in any case they help ensure breadth of testing in the sense that
test cases that achieve 100% coverage in any measure will be exercising all parts of the
software. There is also danger in these techniques. 100% coverage does not mean 100%
tested. Coverage techniques measure only one dimension of a multi-dimension concept. Two
different test cases may achieve exactly the same coverage but the input data of one may
find an error that the input data of the other doesn’t. Further more, coverage techniques
measure coverage of the software code that has been written, they cannot say anything
about the software that has not been written. If a function has not been implemented only
functional testing techniques will reveal the omission.
In common with all structural testing techniques, coverage techniques are best used on
areas of software code where more thorough testing is required. Safety critical code, code
that is vital to the correct operation of a system, and complex pieces of code are all
examples of where structural techniques are particularly worth applying. They should always
be used in addition to functional testing techniques rather than as an alternative to them.

Test coverage can be measured based on a number of different structural elements in

software. The simplest of these is statement coverage which measures the number of
executable statements executed by a set of tests and is usually expressed in terms of the
percentage of all executable statements in the software under test. In fact, all coverage
techniques yield a result which is the number of elements covered expressed as a
percentage of the total number of elements. Statement coverage is the simplest and
perhaps the weakest of all coverage techniques. The adjectives ‘weak’ and ‘strong’ applied
to coverage techniques refer to their likelihood in finding errors. The stronger a technique
the more errors you can expect to find with test cases designed using that technique with
the same measure of coverage.

4.5.2 Types of Coverage

There are a lot of structural elements that can be used for coverage. Each technique uses a
different element; the most popular are described in later sections. Besides statement
coverage, there are number of different types of control flow coverage techniques most of
which are tool supported. These include branch or decision coverage, LCSAJ (linear code
sequence and jump) coverage, condition coverage and condition combination coverage. Any
representation of a system is in effect a model against which coverage may be assessed.
Call tree coverage is another example for which tool support is commercially available.

Another popular, but often misunderstood, coverage measure is path coverage. Path
coverage is usually taken to mean branch or decision coverage because both these
techniques seek to cover 100% of the ‘paths’ through the code. However, strictly speaking
for any code that contains a loop, path coverage is impossible since a path that travels
round the loop say 3 times is different from the path that travels round the same loop 4
times. This is true even if the rest of the paths are identical. So if it is possible to travel
round the loop an unlimited number of times then there are an unlimited number of paths
through that piece of code. For this reason it is more correct to talk about ‘independent path
segment coverage’ though the shorter term ‘path coverage’ is frequently used.

There is currently very little tool support available for data flow coverage techniques, though
tool support is growing. Data flow coverage techniques include definitions, uses, and
definition-use pairs.

Other, more specific, coverage measures include things like database structural elements
(records, fields, and sub-fields) and files. State transition coverage is also possible. It is
worth checking for any new tools, as the test tool market can develop quite rapidly.

4.5.3 How to Measure Coverage

For most practical purposes coverage measurement is something that requires tool support.
However, knowledge of steps needed to measure coverage is useful in understanding the
relative merits of each technique.

1. Decide on the structural element to be used.

2. Count the structural elements.
3. Instrument the code.
4. Run the tests for which coverage measure is required.
5. Using the output from the instrumentation, determine the percentage of elements
 exercised.

Instrumenting the code (step 3) implies inserting code along-side each structural element in
order to record that the associated structural element has been exercised. Determine the
actual coverage measure (step 5) is then a matter of analyzing the recorded information.

When a specific coverage measure is required or desired but not attained, additional test
cases have to be designed with the aim of exercising some or all of the structural elements
not yet reached. These are then run through the instrumented code and a new coverage
measure determined. This is repeated until the required coverage measure is achieved.

Finally the instrumentation should be removed. However, in practice the instrumentation

should be done to a copy of the source code such that it can be deleted once you have
finished measuring coverage. This avoids any errors that could be made when removing
instrumentation. In any case all the tests ought to be re-run again on the uninstrurnented
code.

4.5.4 Statement Coverage

Statement coverage is the number of executable statements exercised by a test or test

suite. This is calculated by:
Number of statements exercised
Statement Coverage = Total number of statements x 100%

Typical ad hoc testing achieves 60% to 75% statement coverage.

4.5.5 Branch & Decision Testing / Coverage

Branch coverage is the number of branches (decisions) exercised by a test or test suite. This
is calculated by:
Number of branches exercised
Branch Coverage = Total number of branches x 100%

Typical ad hoc testing achieves 40% to 60% branch coverage. Branch coverage is stronger
than statement coverage since it may require more test cases to achieve 100% coverage.
For example, consider the code segment shown below.
if a> b
c=O
endif
To achieve 100% statement coverage of this code segment just one test case is required
which ensures variable ‘a’ contains a value that is greater than the value of variable ‘b’.
However, branch coverage requires each decision to have had both a true and false
outcome. Therefore, to achieve 100% branch coverage, a second test case is necessary.
This will ensure that the decision statement ‘if a> b’ has a false outcome.

Note that 100% branch coverage guarantees 100% statement coverage.

Branch and decision coverage are actually slightly different for less than 100% coverage,
but at 100% coverage they give exactly the same results.

5. Test Management

This section looks at the process of managing the tests and test processes. Organization
issues such as types of test teams, responsibilities, etc. will be important if we are to see
effective testing within our organization.
Strong disciplines such as Configuration Management not only assist the developer but also
should be seen as a complete lifecycle discipline that includes ‘testware’.

Test management activities include estimation, monitoring, control and the recording and
tracking of incidents. At any time we should know how well we are doing and understand
what controlling actions we can take to keep testing on target.

5.1 Organization

5.1.1 The importance of independence

It is important to realize that companies will have different requirements when it comes to
organizational structures for testing. The different stages of testing will be performed within
organizations with varying degrees of independence using a variety of different approaches.

We have already seen that independence is important for effective testing. Greater
independence gives a more objective view of the document being reviewed/inspected.
Authors are less likely to take things personally and reviewers less vulnerable to pressure
the more independent the review process becomes.

If we were to plot the number of faults found over a period of time, it would probably rise
steeply at first, but then begin to decline. If we were to release the product to the end users
at that point, we may perhaps expect the number of faults found to continue to diminish,
but instead the number found increases. The reason for this is that the users have a
different ‘worldview’. Therefore if our aim is to find as many faults as possible then we need
to have as many of these different worldviews as possible, and as early as possible in the
life cycle.

There are however advantages of both familiarity and independence and one should not
replace familiarity with independence — we need both. With independence we are providing
that different, more objective assessment of the software and perhaps find faults others
wouldn’t. However the programmer knows and understands the software and will know
where problems are most likely to occur.

5.1.2 Organizational structures for testing

We must recognize that whilst independence is important there are varying degrees of
independence which have advantages and disadvantages associated.
Each of the broad options is discussed in turn below.

5.1.2.1Developers Only

This is where the programmer will test his or her own code. They know the code best and
are more familiar with it, so they may find problems that a less technical tester would miss.
They can also find and fix faults very cheaply at this stage.
However, they might not be the best people to try to break it. There is a tendency to see
what you meant instead of what is actually there, so they may miss things that an
independent mind would see. It may also be a rather subjective assessment of their own
work; they want to show how good they are, not how easily their software can be made to
fall over!

5.1.2.2Development Team Responsibility (buddy testing)

In this regime the developers work together and tests will be designed (and also probably
run) by another developer other than the one who wrote the code. This is sometimes
referred to as ‘buddy testing’. As the name suggests — this is usually operated under
‘friendly’ conditions.

It should be considered more advantageous than the previous option as a certain level of
independence is introduced whilst maintaining a technical perspective. However, as already
intimated — a technical perspective only is certainly insufficient when it comes to testing a
system.

Also, for buddy testing to work effectively, time must be allocated to design (and run) the
tests by the ‘buddy’. This could take longer due to learning curves, work priority and
pressure of their own work.

5.1.2.3Testers on the Development Team

One of the members of a development team is assigned the responsibility for testing. This
person may already be an experienced tester (with or without development experience) and
could be brought into the team specifically to take the testing responsibility. Although
working along-side the developers, this person may not have a detailed knowledge of the
system from a technical perspective. This gives greater independence in their testing and
yet encourages a team spirit in which developers and tester are working toward the same
goal.

However, testers in this situation might find themselves undermined and unreasonable
pressure placed on them to do all the testing because it is deemed to be ‘their job’. They
may be corruptible by peer pressure and where it is only one tester on the team it provides
only a single view.

5.1.2.4Dedicated team of testers

This team is often referred to as the Independent Test Group (ITG) or Independent Test Unit
(ITU) and will usually be totally independent of development with a different reporting
structure. These teams usually are looked upon as the ‘testing experts’ and will have a high
level of testing experience.

There can, however, be a high degree of over-reliance on this team to perform all necessary
tests including those that should be undertaken by developers (particularly component
testing). Alternatively little or no component testing is performed leaving the independent
testers to find coding faults that could have been found and fixed more cheaply had
component testing been performed. In these situations the independent testing team
becomes a bottleneck and test responsibility, rather than shared, is left to the ITU.

Whilst being a completely separate department has its advantages, it does have drawbacks
— namely confrontation and an ‘over-the-wall’ mentality. Both of which need to be resolved
if we are to see an effective and efficient test regime.

5.1.2.5Internal Test Consultants

Ideally, a team of internal test consultants would comprise of highly specialized experts in
the testing world. In practice they tend to be selected on the basis of having the most
experience or perhaps the most enthusiasm for testing (not bad criteria but not necessarily
the same as the ideal!). They offer advice and guidance on various testing issues such as
procedures, test design, test automation, reviews, etc. and may perform ‘health checks’ on
testing throughout the company. They do not undertake the testing themselves but can
assure consistency of testing across different projects.
Advice and guidance is often required and they are usually in a better position to influence
and challenge company procedures. However, they are not given authority and someone still
has to undertake the testing.

5.1.2.6Outside Organization

Some companies provide a testing service. This can be undertaken on their site or they may
send a number of testers to manage and perform the testing on the developer site. These
companies usually specialize in a certain industry (such as insurance, finance, banking, etc.)
and so can provide in depth specialist business knowledge.

As they are outside organizations, they will not be drawn into internal politics. They can be
expensive and any experience they gain will be lost from the project once their testing is
complete.

5.1.3 Usual choices

The most common levels of independence at each of stages of testing are the following:

• Component testing - performed by programmers (in some cases, buddy testing);

• Integration testing in the small - poorly defined and seldom done well by anyone;
• System testing - performed by an independent test team;
• Acceptance testing — performed by users or user representatives.

5.1.4 Resourcing issues

Independence is important in testing, as an independent mind will see things that may be
missed by the person who developed the software. However, familiarity also has benefits. It
is not a question of achieving one or the other, but of achieving a good balance. Different
levels of testing can use different approaches to achieve independence. For example, the
use of test design techniques gives independence of thought. A test strategy should state
what levels of independence are required for each level of testing.

A good mix of skills is important within a project. We must however consider the skill set for
the team. The following are a useful reminder of the sort of skill set we will need in order to
facilitate good testing.
• Technique Specialists — specialize in the use of test design techniques such as
Equivalence Partitioning and Boundary Value Analysis. They can then become the
source of knowledge, advice and guidance for the rest of the team;
• Automation Specialists — These have a keen understanding and desire to specialize in
the test automation arena. They can develop automation standards and promote good
automation practices. They generally need programming skills, since test scripts for
automated tools are programming languages;
• Database Experts — Everyone in the team should have an understanding of the
underlying database, but not everyone needs know the intricate details of the database
environment.
• Business Skills — Having key people in the test team with business knowledge is
essential for testing from a business perspective;
• Usability Experts — We have already seen that testing for usability is often poorly done
because it is often poorly specified. Test (and development) teams could benefit by
having specialized expertise in this area;
• Test Environment Experts — Maintaining test environments is crucial for successful
testing and this task should not be underestimated in terms of its complexity and
sensitivity;
• Test Managers — Essential people who encourage, motivate and protect the rest of the
 test team.

5.2 Configuration Management

5.2.1 What is configuration management?

Our systems are made up of a number of items (or things). Configuration Management is all
about effective and efficient management and control of these items.

During the lifetime of the system many of the items will change. They will change for a
number of reasons; new features, fault fixes, environment changes, etc. We might also have
different items for different customers, such as version A contains modules 1,2,3,4 & 5 and
version B contains modules 1,2,3,6 & 7. We may need different modules depending on the
environments they run under (such as Windows NT and Windows 2000).

An indication of a good Configuration Management system is to ask ourselves whether we

can go back two releases of our software and perform some specific tests with relative ease.

5.2.2 Problems resulting from poor configuration management

Often organizations do not appreciate the need for good configuration management until
they experience one or more of the problems that can occur without it. Some problems that
commonly occur as a result of poor configuration management systems include:
• the inability to reproduce a fault reported by a customer;
• two programmers have the same module out for update and one overwrites the other’s
change;
• unable to match object code with source code;
• do not know which fixes belong to which versions of the software;
• faults that have been fixed reappear in a later release;
• a fault fix to an old version needs testing urgently, but tests have been updated.

5.2.3 Definition of configuration management

A good definition of configuration management is given in the ANSI/IEEE Standard 729-

1983, Software Engineering Terminology. This says that configuration management is:
• “the process of identifying and defining Configuration Items in a system,
• controlling the release and change of these items throughout the system life cycle,
• recording and reporting the status of configuration items and change requests, and
• verifying the completeness and correctness of configuration items.”

This definition neatly breaks down configuration management into four key areas:
• configuration identification;
• configuration control;
• configuration status accounting; and
• configuration audit.
Configuration identification is the process of identifying and defining Configuration Items in
a system. Configuration Items are those items that have their own version number such
that when an item is changed, a new version is created with a different version number. So
configuration identification is about identifying what are to be the configuration items in a
system, how these will be structured (where they will be stored in relation to each other)
the version numbering system, selection criteria, naming conventions, and baselines. A
baseline is a set of different configuration items (one version of each) that has a version
number itself Thus, if program X comprises modules A and B, we could define a baseline for
version 1.1 of program X that comprises version 1.1 of module A and version 1.1 of module
B. If module B changes, a new version (say 1.2) of module B is created. We may then have
a new version of program X, say baseline 2.0 that comprises version 1.1 of module A and
version 1.2 of module B.

Configuration control is about the provision and management of a controlled library

containing all the configuration items. This will govern how new and updated configuration
items can be submitted into and copied out of the library. Configuration control also
determines how fault reporting and change control is handled (since fault fixes usually
involve new versions of configuration items being created).

Status accounting enables traceability and impact analysis. A database holds all the
information relating to the current and past states of all configuration items. For example,
this would be able to tell us which configuration items are being updated, who has them and
for what purpose.

Configuration auditing is the process of ensuring that all configuration management

procedures have been followed and of verifying the current state of any and all configuration
items is as it is supposed to be. We should be able to ensure that a delivered system is a
complete system (i.e. all necessary configuration items have been included and extraneous
items have not been included).

5.2.4 Configuration management in testing

Just about everything used in testing can reasonably be placed under the control of a
configuration management system. That is not to say that everything should. For example,
actual test results may not be though in some industries (e.g. pharmaceutical) it can be a
legal requirement to do so.

Note that configuration management can be done using specialized configuration

management tools or can be done manually.

5.3 Test Estimation, Monitoring and Control

5.3.1 Estimating testing is no different

Test estimation in many ways is no different to estimating other activities such as software
design or programming. We must break down the activities into well defined tasks that, in
turn, can be estimated.

There are a number of methods that can be employed to estimate the testing effort
required, these are described below.
• Guessing (Finger in the Air — F.I.A. approach): This is not a good method to base
our final estimate on, because it can be easily questioned and very often challenged.
Whilst it is not advisable to rely solely on this method, it does have its use and can be
reasonably accurate depending on past project experience and the expertise of the
estimator.
• Past project knowledge: To base our estimates on previous, similar projects is a
reasonable thing to do. This is only effective if we have recorded such data from
previous projects. Again, it can be easily challenged and estimates may be reduced as
a result.
• Work Breakdown Structures (WBS): Here we identify tasks that make up the test
activities and estimate each one in turn. Testers who could verify that the estimate is
 realistic can then review each task estimate. If not then estimates would be re-worked.
Should the estimates not be approved, then each task in turn can be questioned as to
relevance, criticality, urgency and importance — the estimates can then be adjusted
accordingly.

Whatever method is used, test estimation will always be required in advance and should be
included in the high level test plan. The estimates should be reviewed and if necessary
revised throughout the project once further information is obtained.

5.3.2 Estimating testing is different

Whilst test estimation is similar to estimating any other activity in some respects, in others
ways it is very different. The main reasons are as follows:
• it is not an independent activity — testing is completely reliant on development
delivering the software to an agreed date and to an agreed quality standard. Should
the quality of the software not be as good as we expected then we will spend more
time reporting a greater number of faults and retesting them;
• it is reliant on attaining the agreed system — that there are no new surprise features
added by the developers. If extra features are added, then these will need testing and
this will affect the schedules;
• it is reliant on a stable test environment — if the test environment is volatile then this
again will affect the schedules.
It is very important that rework is planned for. (Note that the syllabus for this course does
not cover how planning for rework can be done.)

5.3.2.1Estimating iterations

There is one major difference in estimating testing compared to estimating other tasks.
Most activities, once they are done, that’s it, and they are finished and complete. However,
testing, once it is done it does not stay “done” - it has to be done again and again.
Successful tests will find faults, but once they are fixed, re tests and regression tests are
needed. This can result in a number of test iterations or test cycles. Three or four test
iterations are typical. It may not be necessary to perform all of the tests with every iteration
(often it is not possible because of time constraints) but it is certainly desirable to do so
with the last iteration (though this too is more often an ideal than a reality).

Past history is often a good guide to the likely number of iterations. For example, if the
release of the system underwent five test iterations then there is a good chance that the
next release will need at least four and possibly six.
Other measures that can be used to help estimate test effort more accurately include:
• The number of faults that are likely to be found during testing;
• The percentage of nested faults (faults that can only be found after another one
has been fixed); and the percentage of faults fixed incorrectly (together with the
time wait for each release).

An added complication to estimating iterations is that not all iterations will use all of the
tests. Some may contain only checks for correct fixes, for example, while others may be a
complete regression test of all tests in a suite.

5.3.2.2 Time to report faults

One of the important tasks that testers must do is to report incidents or faults found. It is
important that faults are reported in such a way that a developer can quickly reproduce
them, otherwise they will not be able to fix the fault. But how much time is spent in writing
up fault reports? The more time you spend reporting faults the less testing you will actually
do! However, if a developer cannot reproduce a fault he or she will have wasted time trying.
The tester will then have to spend more time recalling the information that he or she should
have reported in the first place.

The more faults there are in the software the more faults the testers will have to report.
How many can they be reasonably expected to report before this fault reporting time (and
future retesting time) has a significant de~mental impact on the planned testing effort?
When estimating test effort it is important to consider how faults are likely to be found and
so how much time will be spent reporting and retesting them.

5.3.3 Monitoring progress

Once we have started testing we must monitor our test progress and take corrective action
should things go wrong. Effective monitoring and control is vital in the test management
process.
Recording the number of tests run against the number of tests passed and the number of
tests planned is one good way to show test progress. This is a powerful visual aid when
plotted on a graph. This type of graph is called an ‘S-curve’ because its shape resembles the
letter S (somewhat elongated). S-curves give early warning of problems. For example, if the
number of tests passing falls significantly below the number run there may be a number of
different reasons. Perhaps there are lots of faults being found or there are only one or two
faults that are affecting many tests. In either case more development effort needs to be
resourced to fix the faults. A faulty test environment could cause the problem, so in this
case more development resource may be needed to resolve it.

Other useful measures include the number of incidents or faults raised together with either
the number resolved and / or the number of faults expected.

5.3.4 Test control

Test control is about management actions and decisions that affect the testing process,
tasks and people with a view to making a testing effort achieve its objectives. This may be
the original or a modified plan. Modifying an original plan in the light of new knowledge (i.e.
what testing has revealed so far) is a frequently necessary and prudent step.

The use of entry and exit criteria are perhaps one of the simplest and yet highly effective
control mechanisms available to managers. Entry criteria are conditions that must be
meeting before the associated activity can start. Similarly, exit criteria are conditions that
must be met before an activity can be declared complete.

The exit criteria of one activity are often the same as the entry criteria of the next activity.
For example, the exit criteria for component testing might be that all components have been
tested sufficiently to achieve 100% statement coverage and all known faults have been
fixed. These could also be the entry criteria for the next testing activity (supposedly
integration in the small, but could also apply to system testing). However, the next activity
might have additional entry criteria. For example, entry criteria for system testing might
include something about the availability of the test environment.

Tightening (or loosening) an entry or exit criteria is just one of the actions a test manager
can take. Reallocation of resources such as acquiring more testers or developers, and
moving people from one task to another to focus attention on more important areas is often
an effective controlling action.

There are some factors that testing can affect indirectly, such as which faults are fixed first.
However, one thing that cannot be affected by testing is the number of faults that are
already in the software being tested. The testing only affects whether or not those faults are
found.

Once a controlling action has been taken some form of feedback is essential in order to see
the effect of the action.

Neither the testers nor the test manager should make the decision about when to release a
product. The testers and test manager are responsible for supplying accurate and objective
information about the software quality so that whoever does make the release decision
makes it on the basis of solid facts.

5.4 Incident Management

5.4.1 What is an incident?

An incident is any event that occurs during testing that requires subsequent investigation or
correction. Usually, the event is a mismatch between the actual and expected results of a
test (a failure occurs). The cause of this can be one of a number of things:
• a fault in the software;
• a fault in the test (e.g. expected result was wrong);
• the environment was wrong;
• the test was run incorrectly (e.g. entered the wrong input); or
• a documentation or specification fault (i.e. what the specification says is wrong).

5.4.2 Monitoring incidents

Incident reports can be analyzed to monitor and improve the test process. For example, if a
significant number of incidents reported during system testing turn out to be coding faults
that could have been found by component testing then this tells us that the component
testing process should be improved.

5.4.3 Reporting incidents

Whilst we can log incidents at any stage throughout the lifecycle, it is advisable to only log
incidents after hand-over from development or at least when someone other than the
author of the software performs the testing. This is largely because the benefit of a
developer logging incidents on his or her code before they hand it to anyone else is vastly
outweighed by the cost of doing. It is much cheaper for a developer to simple fix the
problem and retest it than it is for him or her to stop and log the problem before fixing it.
There is also a psychological cost to the developer — effectively having to log all the faults
in their own code is not a big motivator.

There is a distinct danger that not enough time is spent in logging an incident. It is the
tester’s responsibility to raise incidents factually and with enough detail for the developer to
do their job efficiently. Otherwise we could end up with the situation that the fault cannot be
reproduced by the developer so the incident report is returned with a request for more
information, or worse still, the incident is ignored. Spending extra time logging sufficient
information such that it can be reliably and quickly reproduced will be of great benefit.

Information we should record typically includes:

• test id (the test that failed);
• details of the test environment (e.g. whether run on Windows NT or Windows 2000,
etc.);
 • id and version of the software under test;
• both actual and expected results (for comparison purposes and to help developers track
the fault);
• severity (impact of the failure to the customer/user);
• priority (the urgency of fixing the fault);
• name of the tester or automated test information; and
• any other relevant information needed so that the developer can reproduce and fix the
fault.

Basically anything that the developer needs to know in order to reproduce the fault with
ease. We should not tell them how to code the changes though!
There might be other information that can be recorded that will help you and your
organization with metrics and monitoring of progress (for example, the effort spent on
handling the incident).

5.4.4 Tracking incidents

Incidents should be tracked from inception (introduction) through the various stages to their
final resolution. For any incident logged we must be in a position of knowing its exact
status, whether it be waiting for further action, be with a developer for fixing, with the
tester for re-testing, or has been re-tested and cleared.

5.4.5 Severity versus priority

It is useful to distinguish severity and priority, because they are different aspects. Severity
is related to the impact of a failure caused by a fault. Priority is related to the urgency of
fixing a fault. For example, if a fault is holding up a series of automated tests, it could have
high priority even if the impact on the user is low.

5.5 Standards for Testing

There are a number of different standards for testing from the Quality Management
standard (ISO 9000 series) which tells us that testing should be done, to industry specific
standards that detail the level of testing to be performed. The most recent testing standards
are those adopted by the ISEB — namely BS792S-l and BS792S-2. The latter tells us how
we should perform the testing.

6. Tool Support for Testing (CAST)

6.1 Types of CAST Tool

Tool support is available for testing in every stage of the software development life cycle.
However, this does not mean to say that all testing activities can be automated or indeed
made automatic. Tool support for many testing activities usually facilitates greater
productivity and greater accuracy but still requires manual participation throughout.

6.1.1 Requirements testing tools

Requirements testing tools are a relatively new type of tool that helps with the task of
analyzing requirements. These tools can work on requirement specifications written in a
formal structured language or just plain English. Although they cannot help validate
requirements (i.e. tell you if the requirements are what the end user actually wants) they
can help with verifying the requirements (i.e. checking conformance to standards for
requirements specifications). A modern word processing application can be seen as a very
basic requirement-testing tool since one of the functions of these tools is to check grammar.
Ambiguity in a requirement specification often leads to serious faults in the delivered system
and these ambiguities are sometimes caused by poor grammar. Proper requirements testing
tools offer a much richer set of functionality than mere grammar checking (though this too
is one of their functions). For example, they can check for consistent use of key terms
throughout a specification and derive a list of possible test conditions for acceptance testing
(though this should sensibly be used as a starting point for further development or as an
additional source of ideas for cross checking purposes).

Possible pitfalls of these tools include false confidence. The fact that the tool does not find
anything wrong with a requirement specification does not imply it is perfect but someone is
likely to see it that way! They do require manual intervention, they are not automatic and
they certainly cannot correct all the (potential) faults that they find.

Perhaps the most obvious pitfall is that the requirements have to be written down. Many
organizations fail to produce a complete requirement specification and for them this type of
tool will have limited value (unless it proves to be the catalyst for more complete
requirements specifications).

6.1.2 Static analysis tools

Static analysis tools analyze source code without executing it. They are a type of super
compiler that will highlight a much wider range of real or potential problems than compilers
do. Static analysis tools detect all of certain types of fault much more effectively and
cheaply that can be achieved by any other means. For example, they can highlight
unreachable code, some infinite loops, use of a variable prior to its definition, and
redefinition of a variable without an intervening use. These and many more potential faults
can be difficult to see when reading source code but can be picked up within seconds by a
static analysis tool.

Such tools also calculate various metrics for the code such as McCabe’s cyclomatic
complexity, Halstead metrics and many more. These can be used, for example, to direct
testing effort to where it is most needed.

Although an extremely valuable type of testing tool, it is one that is not used by many
organizations. The pitfalls are more psychological than real, for example, a static analysis
tool may highlight something that is not going to cause a failure of the software. This is
because it is a static view of the software.

6.1.3 Test design tools

Test design tools help to derive test inputs. They are sometimes referred to as test case
generators though this is a rather exaggerated claim. A proper test case includes the
expected outcome (i.e. what the result of running the test case should be). No tool will ever
be able to generate the expected outcome (other than for the most simple and possibly
least needed test cases). Thus we prefer to call them partial test case generators.

Test design tools usually work from a formal specification, an actual user interface or from
source code. In the first case the specification has to be in a formal language that the test
design tool understands or for some tools a CASE (Computer Aided Software Engineering)
tool can hold it. A CASE tool captures much of the information required by the test design
tool as the system is being designed and therefore saves the need to re-specify the design
information in a different format just for the test tool. Where a test design tool uses the
user interface of an application the user interface has to be implemented before the test
design tools can be used. It is also a fairly restricted set of test inputs that it can generate
since they concentrate on testing the user interface rather than the underlying functionality
of the software. This is still useful though. When it is the source code that is used to
generate test inputs, it is usually in conjunction with a coverage tool, to identify the branch
conditions needed to send a test down a particular path.

6.1.4 Data preparation tools

Data preparation tools manipulate existing data or generate new test data. Where new data
is generated the tool uses a set of instructions or rules supplied by you that describe the
format and content of the data to be generated. For example, if you require a lot of names
and addresses to populate a database you would specify the valid set of characters and the
maximum and minimum lengths of each field and let the tool generate as many records as
you require. The names and addresses it generates will not be sensible English names but
they will conform to the rules you laid down and so will be valid for the purposes of testing.

Starting with actual data and manipulating it to ensure data privacy and/or reduce its size
can generate more realistic test data.

This type of tool makes it possible to generate large volumes of data (as required for
volume, performance and stress testing for example) when it is needed. This makes it more
manageable since the large volumes do not necessarily have to be kept since they can be
regenerated whenever required. On the downside, the technical set up for complex test data
may be rather difficult or at least very tedious.

6.1.5 Test running tools

Test running tools enable tests to be executed automatically and in some cases enables the
test outputs to be compared to the expected outputs. They are most often used to
automate regression testing and usually offer some form of capture/replay facility to record
a test being performed manually so it can then replay the same key strokes. The recording
is captured in a test script that can be edited (or written from scratch) and is used by the
tool to re-perform the test.

These tools are applicable to test execution at any level: unit, integration, system or
acceptance testing. The benefits include faster test execution and unattended test execution
reducing manual effort and permitting more tests to be run in less time.

The pitfalls are enormous and have caused as many as half of all test automation projects to
fail in the long term. The cost of automating a test case is usually far more (between 2 and
10 times) than the cost of running the same test case manually. The cost of maintaining the
automated test cases (updating them to work on new versions of the software) can also
become larger than the manual execution cost. It is possible to avoid these pitfalls but it is
not necessarily easy to do so. For useful advice with this type of tool, read the book
Software Test Automation: effective use of test execution tools” by Mark Fewster and
Dorothy Graham, Addison Wesley, 1999.

The syllabus separates test running tools into two categories: character-based and GUI
(Graphical User Interface) tools.

6.1.6 Comparison tools

Test running tools usually offer some form of dynamic comparison facilities that enable the
output to the screen during the execution of a test case to be compared with the expected
output. However, they are not as good at comparing other types of test outcome such as
changes to a database and generated report files. For this a stand-alone comparison tool
can be used.
These tools offer vastly improved speed and accuracy over manual methods. They will
highlight all differences they find, even the ones you are not interested in unless you can
specify some form of mask or filter to hide those expected differences such as dates and
times. Specifying masks may not be an easy task.

6.1.7 Test harnesses and drivers

Not all software can be turned into an executable program. For example, a library function
that a programmer may use as a building block within his or her program should be tested
separately first. This requires a harness or driver, a separate piece of source code that is
used to pass test data into the function and receive the output from it.

At unit testing and the early stages of integration testing these are usually custom-built
though there are a few commercial tools that can provide some support (though they are
likely to be language specific).

At later stages of testing such as system and acceptance testing, harnesses (also called
simulators) may be required to simulate hardware systems that are not available or cannot
be used until the software is demonstrably shown to be reliable. For example, software that
controls some aspect of an aircraft needs to be known to work before it is installed in a real
aircraft!

6.1.8 Performance testing tools

If performance measurement is something you need to do, then a performance testing tool
is a must. Such tools are able to provide a lot of very accurate measures of response times,
service times and the like. Other tools in this category are able to simulate loads including
multiple users, heavy network traffic and database accesses. Although they are not always
easy to set up, simulating particular loads is usually much more cost effective than using a
lot of people and/or hardware, or “testing” your new web site by public failure to cope!

6.1.9 Dynamic analysis tools

Dynamic analysis tools assess the system while the software is running. For example, tools
that can detect memory leaks are dynamic analysis tools.

A memory leak occurs if a program does not release blocks of memory when it should, so
the block has leaked out of the pool of memory blocks available to all programs. Eventually
the faulty program will end up owning all of memory, nothing can run, the system hangs up
and must be re-booted.

6.1.10Debugging tools

Debugging tools are traditionally used by programmers to help investigate problems in their
source code. They allow the code to be executed one instruction at a time and the value of
variables to be examined and set. This latter facility makes them particularly good for
testing since specific conditions can be simulated within the source code by using the
debugger tool to set variables to particular values.

6.1.11Test management tools

Test management tools help throughout the software development lifecycle. This category
covers tool support for test planning and monitoring, but also incident management (fault
tracking) tools. For example, some test management tools help with decomposition of the
system functionality into test cases and are able to track and cross-reference from
requirements through to test cases and back again. In this way, if a requirement changes it
is possible for the test management tool to highlight those test cases that will need to be
updated and rerun. Similarly if a test case fails, the tools will be able to highlight the
requirement(s) affected.

Many test management tools are integrated with (or provide an interface to) other testing
tools, particularly test running tools. This can be exceedingly helpful since it becomes
possible to cause automated tests to be executed from within the test management tool,
where you have all the information on the success or failure of the previous tests.

6.1.12Coverage tools

Coverage tools assess how much of the software under test has been exercised by a set of
tests. They can do this by a number of methods but the most common is for the tool to
instrument the source code in a “precompiler pass”. This involves the tool in inserting new
instructions within the original code such that when it is executed the new code writes to a
data file recording the fact that it has been executed. After a set of test cases have been
executed the tool then examines this data file to determine which parts of the original
source code have been executed and (more importantly) which parts have not been
executed.

Coverage tools are most commonly used at unit test level. For example, branch coverage is
often a requirement for testing safety-critical or safety-related systems. However, some
coverage tools can also measure the coverage of design level constructs such as call trees.

It is important not to set some arbitrary coverage measure as a target without a good
understanding of the consequences of doing so. Achieving 100% branch coverage may
seem like a good idea but it can be a very expensive goal, and may not be the best testing
that could be done in the circumstances. Coverage has to be justified against other means
of achieving the desired level of quality of software.

6.2 Tool Selection and Implementation

This section gives a brief overview of the tool selection and implementation process as
described by the syllabus. More information on tool selection and implementation can be
found in Chapters 10 and 11 of the book “Software Test Automation” by Mark Fewster and
Dorothy Graham, published by Addison-Wesley, 1999,ISBN0-201-331403.

6.2.1 Which test activities to automate?

Given the wide range of different types of testing tool there is a wide range of testing
activities that can be tool supported (but not necessarily fully automated). When considering
tool support identifies all the activities where tool support could be beneficial and prioritize
them to identify the most important. Although test running tools are the most popular, this
may not be the best place to introduce tool support first.

6.2.2 CAST tool requirements

Testing tools often provide a rich set of functions and features. However, the most
functionally rich tool may not be best for any particular situation. Having a tool that matches
the existing test process is usually more important. If a tool does not adequately support
your test process it could be more of a hindrance than a help.

6.2.3 CAST readiness

CAST readiness is the term given to a test process that is ready for tool support, i.e. the
ease with which CAST tools can be implemented. If you have a chaotic testing process it is
unlikely that any testing tool will help. The benefits of tools usually depend on a systematic
and disciplined test process. If the test process is poor it may be possible to improve it in
parallel with the introduction of a testing tool but care should be taken. In our experience
about half of the organizations that have bought a test tool have abandoned it. Introducing
a testing tool is not an easy thing to do well and if it is not done well there is a strong
chance that testing will cost more or be less effective as a result of the tool.

6.2.4 Environment constraints and CAST integration

The environment into which it will go will almost certainly influence the choice of a testing
tool. Some tools can only work on certain hardware platforms, operating systems or
programming languages, and some require their own hardware.

If one or more testing tools are already in use it may be necessary to consider their
integration with any new tool. This issue may also affect other types of tool, for example,
configuration management tools and CASE (Computer Aided Software Engineering) tools.
Some vendors offer a variety of integrated test tools where compatibility between the tools
is assured although this does not necessarily mean that if one of their tools suites your
needs all the other will as well.

6.2.5 Selection process

After it has been decided which testing activity will offer most benefit from tool support the
job of selecting a suitable test tool should be considered as a process comprising four
stages:
• produce a candidate tool shortlist;
• arrange demonstrations;
• evaluate the selected tool(s);
• review and select tool.

6.2.6 Implementation (pilot project and roll out)

After a tool has been selected the task of implementing should start with a pilot project. The
aim of this is to ensure that the tool can be used to achieve the planned benefits. Objectives
of the pilot project include gaining experience with the tool in a controlled way and on a
small scale, identifying changes to the testing process to accommodate the tool to its best
advantage and to assess the actual costs and benefits of a full implementation.

Roll out of the tool on a larger scale should only be attempted after a successful pilot
project. This will require a strong commitment from new tool users and projects as there is
an initial overhead is using any new tool.