Envision , Evolve, Unleash

Principles of Testing

Contents
Envision , Evolve, Unleash

Why testing is necessary

Fundamental test process
Psychology of testing
Re-testing and regression testing
Expected results
Prioritisation of tests

What is a bug?

Failure: deviation of the software from its

expected behaviour

Quality Definitions

Definitions:
Meeting the customers requirements the first
time and every time
Conformance to a set of customer
requirements that, if met, results in a product
that is fit for its intended use

Reliability versus faults

Reliability: the probability that software will not

cause the failure of the system for a specified
time under specified conditions
Can a system be fault-free? (zero faults, right first
time)
Can a software system be reliable but still have
faults?
Is a fault-free software application always
reliable?

Why do faults occur in software?

software is written by human beings

who know something, but not everything
who have skills, but arent perfect
who do make mistakes (errors)

under increasing pressure to deliver to strict

deadlines
no time to check but assumptions may be wrong
systems may be incomplete

if you have ever written software ...

What do software faults cost?

huge sums
Ariane 5 ($7billion)
Mariner space probe to Venus ($250m)
American Airlines ($50m)

very little or nothing at all

minor inconvenience
no visible or physical detrimental impact

software is not linear:

small input may have very large effect

Safety-critical systems

software faults can cause death or injury

radiation treatment kills patients (Therac-25)
train driver killed
aircraft crashes (Airbus & Korean Airlines)
bank system overdraft letters cause suicide

So why is testing necessary?

because software is likely to have faults
to learn about the reliability of the software
to fill the time between delivery of the software
and the release date
to prove that the software has no faults
because testing is included in the project plan
because failures can be very expensive
to avoid being sued by customers
to stay in business

Why not just "test everything"?

Avr. 4 menus
3 options / menu
system has
20 screens

Average: 10 fields / screen

2 types input / field
(date as Jan 3 or 3/1)
(number as integer or decimal)
Around 100 possible values

Total for 'exhaustive' testing:

20 x 4 x 3 x 10 x 2 x 100 = 480,000 tests
If 1 second per test, 8000 mins, 133 hrs, 17.7 days
(not counting finger trouble, faults or retest)

10 secs = 34 wks, 1 min = 4 yrs, 10 min = 40 yrs

Exhaustive testing?

What is exhaustive testing?

when all the testers are exhausted
when all the planned tests have been executed
exercising all combinations of inputs and
preconditions

How much time will exhaustive testing take?

infinite time
not much time
impractical amount of time

How much testing is enough?

its never enough
when you have done what you planned
when your customer/user is happy
when you have proved that the system works
correctly
when you are confident that the system works
correctly
it depends on the risks for your system

How much testing?

It depends on RISK
risk of missing important faults
risk of incurring failure costs
risk of releasing untested or under-tested software
risk of losing credibility and market share
risk of missing a market window
risk of over-testing, ineffective testing

So little time, so much to test ..

test time will always be limited

use RISK to determine:
what to test first
what to test most
how thoroughly to test each item

i.e. where to
place emphasis

what not to test (this time)

use

RISK to
allocate the time available for testing by
testing ...

prioritising

Most important principle

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

Testing and quality

testing measures software quality

testing can find faults; when they are
removed, software quality (and possibly
reliability) is improved
what does testing test?
system function, correctness of operation
non-functional qualities: reliability, usability,
maintainability, reusability, testability, etc.

Other factors that influence testing

contractual requirements
legal requirements
industry-specific requirements
e.g. pharmaceutical industry (FDA), compiler
standard tests, safety-critical or safety-related
such as railroad switching, air traffic control
ItIt is
isdifficult
difficultto
todetermine
determine
how
how much
muchtesting
testing is
isenough
enough
but
but ititis
isnot
notimpossible
impossible

Envision , Evolve, Unleash

Fundamental test process

The test process

Planning (detailed level)

specification

execution

recording

check
completion

Test planning

how the test strategy and project test plan

apply to the software under test
document any exceptions to the test strategy
e.g. only one test case design technique needed
for this functional area because it is less critical

other software needed for the tests, such as

stubs and drivers, and environment details
set test completion criteria

Test specification

Planning (detailed level)

specification

execution

Identify conditions
Design test cases
Build tests

recording

check
completion

A good test case

effective
exemplary
evolvable

Finds faults

Represents others

Easy to maintain

economic
Cheap to use

Test specification

test specification can be broken down into three

distinct tasks:
1. identify:
2. design:
3. build:

determine what is to be tested (identify

test conditions) and prioritise
determine how the what is to be tested
(i.e. design test cases)
implement the tests (data, scripts, etc.)

Task 1: identify conditions

(determine what is to be tested and prioritise)
list the conditions that we would like to test:
use the test design techniques specified in the test plan
there may be many conditions for each system function or
attribute
e.g.
life assurance for a winter sportsman
number items ordered > 99
date = 29-Feb-2004

prioritise the test conditions

must ensure most important conditions are covered

Selecting test conditions

Importance

First set

Best set

Time

Task 2: design test cases

(determine how the what is to be tested)

design test input and test data

each test exercises one or more test conditions

determine expected results

predict the outcome of each test case, what is
output, what is changed and what is not changed

design sets of tests

different test sets for different objectives such as
regression, building confidence, and finding faults

Designing test cases

Importance

Most important
test conditions
Least important
test conditions
Test cases

Time

Task 3: build test cases

(implement the test cases)

prepare test scripts

less system knowledge tester has the more
detailed the scripts will have to be
scripts for tools have to specify every detail

prepare test data

data that must exist in files and databases at the
start of the tests

prepare expected results

should be defined before the test is executed

Test execution

Planning (detailed level)

specification

execution

recording

check
completion

Execution

Execute prescribed test cases

most important ones first
would not execute all test cases if
testing only fault fixes
too many faults found by early test cases
time pressure

can be performed manually or automated

Test recording

Planning (detailed level)

specification

execution

recording

check
completion

Test recording 1
The test record contains:
identities and versions (unambiguously) of
software under test
test specifications

Follow the plan

mark off progress on test script

document actual outcomes from the test
capture any other ideas you have for new test cases
note that these records are used to establish that all test
activities have been carried out as specified

Test recording 2
Compare actual outcome with expected outcome.
Log discrepancies accordingly:

software fault
test fault (e.g. expected results wrong)
environment or version fault
test run incorrectly

Log coverage levels achieved (for measures

specified as test completion criteria)
After the fault has been fixed, repeat the required
test activities (execute, design, plan)

Check test completion

Planning (detailed level)

specification

execution

recording

check
completion

Check test completion

Test completion criteria were specified in the

test plan
If not met, need to repeat test activities, e.g.
test specification to design more tests
Coverage too low
specification

execution

recording

check
completion

Coverage
OK

Test completion criteria

Completion or exit criteria apply to all levels of

testing - to determine when to stop
coverage, using a measurement technique, e.g.
branch coverage for unit testing
user requirements
most frequently used transactions

faults found (e.g. versus expected)

cost or time

Envision , Evolve, Unleash

Re-testing and regression testing

Re-testing after faults are fixed

Run a test, it fails, fault reported

New version of software with fault fixed
Re-run the same test (i.e. re-test)
must be exactly repeatable
same environment, versions (except for the
software which has been intentionally changed!)
same inputs and preconditions

If test now passes, fault has been fixed

correctly - or has it?

Re-testing (re-running failed tests)

New faults introduced by the first
fault fix not found during re-testing

x
x
x

Fault now fixed

Re-test to check

Regression test

to look for any unexpected side-effects

x
x
x

Cant guarantee
to find them all

Regression testing 1

misnomer: "anti-regression" or "progression"

standard set of tests - regression test pack
at any level (unit, integration, system,
acceptance)
well worth automating
a developing asset but needs to be
maintained

Regression testing 2

Regression tests are performed

after software changes, including faults fixed
when the environment changes, even if
application functionality stays the same
for emergency fixes (possibly a subset)

Regression test suites

evolve over time
are run often
may become rather large

Regression testing 3

Maintenance of the regression test pack

eliminate repetitive tests (tests which test the
same test condition)
combine test cases (e.g. if they are always run
together)
select a different subset of the full regression suite
to run each time a regression test is needed
eliminate tests which have not found a fault for a
long time (e.g. old fault fix tests)

Regression testing and automation

Test execution tools (e.g. capture replay) are
regression testing tools - they re-execute tests which
have already been executed
Once automated, regression tests can be run as
often as desired (e.g. every night)
Automating tests is not trivial (generally takes 2 to 10
times longer to automate a test than to run it
manually
Dont automate everything - plan what to automate
first, only automate if worthwhile

Envision , Evolve, Unleash

Expected results

Expected results

Should be predicted in advance as part of the

test design process
Oracle Assumption assumes that correct outcome
can be predicted.

Why not just look at what the software does

and assess it at the time?
subconscious desire for the test to pass - less work
to do, no incident report to write up
it looks plausible, so it must be OK - less rigorous
than calculating in advance and comparing

A test
inputs

expected
outputs

A Program:
Read A
IF (A = 8) THEN
PRINT (10)
ELSE
PRINT (2*A)

6?

10?

Envision , Evolve, Unleash

Prioritisation of tests

Prioritising tests

We cant test everything

There is never enough time to do all the
testing you would like
So what testing should you do?

Most important principle

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

How to prioritise?

Possible ranking criteria (all risk based)

test where a failure would be most severe
test where failures would be most visible
test where failures are most likely
ask the customer to prioritise the requirements
what is most critical to the customers business
areas changed most often
areas with most problems in the past
most complex areas, or technically critical

Summary
Envision , Evolve, Unleash

Testing is necessary because people make errors

The test process: planning, specification, execution,
recording, checking completion
Independence & relationships are important in
testing
Re-test fixes; regression test for the unexpected
Expected results from a specification in advance
Prioritise to do the best testing in the time you have