Prescriptive or Exploratory Testing: Which Is More Effective? · 2019. 7. 25. · Boeing Australia...
Transcript of Prescriptive or Exploratory Testing: Which Is More Effective? · 2019. 7. 25. · Boeing Australia...
Prescriptive or Exploratory Testing: Which Is More
Effective?
Tafline Murnane,
K.J. Ross & Associates, Australia
Europe’s Premier Software Testing Event
Stockholmsmässan, Sweden
WWW.EUROSTARCONFERENCES.COM
“Testing For Real, Testing For Now”
Prescriptive or Exploratory Testing:
Which is More Effective?
Tafline MurnaneK. J. Ross & Associates Pty. Ltd.
PO Box 131, West Burleigh, 4219, QLD, Australia
M: +61 416 117 931 F: 1300 854 064
http://www.kjross.com.au
Boeing Australia
3
“It can be frustrating, as a tester, to watch someone
who has no experience in testing spend five
minutes using a piece of software and crash it.”
– Ron Patton, Software Testing (2006)
Boeing Australia
4 4
About me…
• Tafline Murnane has worked for the past 3 years as a senior consultant
for K. J. Ross & Associates in Australia (a specialist testing consultancy)
• Previously worked as a developer (before moving to the dark side…)
and as a lecturer and tutor in software testing
• Co-editor of the new ISO/IEC 29119 Software Testing standard
• Completing a PhD in software testing
• Presenter at software testing and engineering conferences both
nationally and internationally
• And is rather passionate about testing…
Melbourne
Boeing Australia
5
Q. How do professional testers perform
black-box testing?
• Prescriptively, using specification-based techniques? e.g.
– Equivalence Partitioning (EP)
– Boundary Value Analysis (BVA)
– Syntax Testing (ST)
• Non-prescriptively, using „experienced-based‟ approaches? e.g.
– Error Guessing
– Exploratory Testing
Boeing Australia
6
Experiment aim
• Compare effectiveness of industry approaches to black-box
testing to „Atomic Rules‟ representation of EP, BVA and ST
– Atomic Rules approach is a very prescriptive representation of EP,
BVA and ST (see following slide…)
• Effectiveness analysed in terms of:
– Usability
• Test technique coverage
• Understanding of test techniques
– Defect-detection percentage
• Percentage of „known‟ defects found
Boeing Australia
7
Defining EP, BVA and ST more prescriptively
through Atomic Rules
• Traditional black-box testing technique descriptions
can be ambiguous
• e.g., the following is an extract from Glenford Myers‟,
The Art of Software Testing (1979):
“If an input condition specifies a „must be‟ situation „first
character of the identifier must be a letter‟, identify one
valid equivalence class (it is a letter) and one invalid class
(it is not a letter)”„is a letter‟ is easy
A | B | C…What constitutes
„not a letter‟???
Boeing Australia
8 8
The Atomic Rules Approach
Equivalence Partitioning
• “Atomic Rules” definition for Equivalence Partitioning:
– EP1: less than lower boundary selection
– EP2: greater than upper boundary selection
– EP3: lower to upper boundary selection
– EP4: integer replacement
– EP5: real number replacement
– EP6: single alpha replacement
– EP7: multiple alpha replacement
– EP8: multiple alphanumeric replacement
– EP9: single non-alphanumeric replacement
– EP10: multiple non-alphanumeric replacement
– EP11: missing item replacement
– EP12: valid list selection
“is a letter”
tested by list
selector rule
“not a letter”
tested by invalid
datatype rules
Tests ranges
Boeing Australia
9 9
The Atomic Rules Approach
Boundary Value Analysis & Syntax TestingBoundary Value Analysis
• BVA1: Lower Boundary - Selection
• BVA2: Lower Boundary Selection
• BVA3: Lower Boundary + Selection
• BVA4: Upper Boundary - Selection
• BVA5: Upper Boundary Selection
• BVA6: Upper Boundary + Selection
• BVA7: First List Item Selection
• BVA8: Last List Item Selection
• BVA9: Missing Item Replacement
Syntax Testing
• ST1: Remove last character
• ST2: Replace last character
• ST3: Add extra character to end
• ST4: Remove first character
• ST5: Replace first character
• ST6: Add extra character to start
• ST7: Uppercase a lowercase letter
• ST8: Lowercase an uppercase letter
• ST9: Null all input
• ST10: Duplicate field
• ST11: Add a field
• ST12: Select each list alternative
• ST13: Select all list alternatives
• ST14: Select all list alternatives in reverse order
The lists of „Atomic Rules‟ act as a
checklist (i.e. Quick Reference
Guide) during black-box testing
BVA1 to BVA6 select values just on, just above and just below the boundaries of equivalence classes (see notes)
Boeing Australia
10 10
Attribute Atomic Rule
Test Method Equivalence Partitioning
Identifier EP1
Name Less than Lower Boundary Selection
DescriptionSelects an equivalence class containing values that lie
below the lower boundary of a field
Source [Myers79]
Rule Type Data-Set Selection Rule (DSSR)
Set Type Range
Valid or Invalid Invalid
Original Datatype Integer, Real, Alpha, Non-alphanumeric
Test Datatype Same as original
Test Data Length Same as original
# Fields Populated 1
# Tests Derived 0
Examples …
The Atomic Rules Schema
• Each Atomic Rule is defined in a characterisation schema
Boeing Australia
11 11
Two Previous University Experiments
Myers‟
representation
of EP & BVA
Atomic Rules
representation
of EP & BVA
Compare learnability & usability
• Limitations:
- involved only novice testers, not industry practitioners
- test cases designed for fictional specifications, thus defect-
detection effectiveness could not be assessed
• Nonetheless, these university experiments provided good
preparation for the industry experiment…
Boeing Australia
12
The Industry Experiment
“Practitioner”
approach to
black-box
testing
Compare usability &
defect-detection effectiveness
Atomic Rules
representation
of EP, BVA &
ST
Boeing Australia
13 13
Two Day Experiment
11 participants5 test leads
5 testers
1 manager new to testing
Average 2.5 years experience in testing
Programs tested Address Parser
Batch Processor
Data collectedTest cases
Defect reports
Feedback on participant experiences
Materials providedSlides
Quick reference guide
Boeing Australia
14 14
Group 2
Atomic Rules Testing of
Address Parser
Experiment Plan
Introductory Presentation
Group 1
Practitioner Testing of
Address Parser
Group 2
Practitioner Testing of
Batch Processor
Atomic Rules Presentation
Group 1
Atomic Rules Testing of
Batch Processor
Wrap-Up Presentation
Each participant
tested a different
program during both
phases of testing
Boeing Australia
15 15
Two Programs Tested
The Address Parser
100 Main Street Melbourne 3000.
FLAT 002 , 100 High St Sydney 2000.
UNIT 014 / 220 Main St North Ryde 2113.
C/o High Street North North Sydney 2000.
C/- St Helena Road Greensborough 3088.
The Batch Processor
sbatch 11BBB
A11A 10
ebatch 11BBB
sbatch 22ABC
B22B -10
C99E 20, A11A -40
ebatch 22ABC
lbatch
Expected the Address Parser to
be a familiar conceptExpected the Batch Processor to
be an unfamiliar concept
Boeing Australia
16 16
Research Questions
1. How did the practitioners perform black-box testing?
2. How many defects did they detect?
3. How many defects were detectable by Atomic Rules?
4. Were the results affected by domain knowledge?
5. Were the results affected by tester experience?
6. Were the results affected by current job?
7. What was their understanding of EP, BVA & ST?
• Standard tests for significance were used– result is statistically significant if it is unlikely
to have occurred by chance
– using t-tests, chi-square tests, etc.
• Threats to validity were thoroughly explored
Boeing Australia
17 17
1. How did the practitioners perform
black-box testing?Q. How would you explain your approach to test case design on day 1?
1
1
1
1
2
3
3
4
7
0 1 2 3 4 5 6 7 8
Random
Duplication / addition
Syntax
Output
Boundary
Experience-based / ad hoc
Valid
Invalid
Specification-based
Self-e
xpla
nation
Number of Testers
One participant also developed
a random number generator in
MS Excel!
Despite these responses
from participants, test
case design patterns
indicated that all
participants carried out
Exploratory Testing
Boeing Australia
18 18
1. How did the practitioners perform
black-box testing? (continued…)
Technique Approach
Technique
Coverage
Std
Dev
Significant
Difference?
EPPractitioner 13% 3.6
NoAtomic Rules 14% 8.6
BVAPractitioner 3% 1.7
YesAtomic Rules 9% 6.8
STPractitioner 2% 0.8
YesAtomic Rules 9% 10
Out of 501 tests derived during
„practitioner‟ testing, only eight (1.6%)
could not be derived by Atomic Rules
from EP, BVA and ST
Low % coverage, as large quantity of
test cases needed to be designed to
achieve 100% coverage (i.e. required
more than 3.5 hours for test design)
Q. How did test technique coverage compare between day 1 and 2?(number tests derived / total number tests derivable by the technique)
Boeing Australia
19 19
2. How many defects did they detect?
• Practitioner approach: 25%
• Atomic Rules approach: 35%
(Statistically significant difference was found)
The „prescriptive‟
Atomic Rules
approach was
more effective
=Defect Detection
Percentage (DDP)number faults detected
total number known faults
Given more time, higher
percentages may have been
achieved for both prescriptive
and non-prescriptive testing
Boeing Australia
20 20
• Address Parser
– EP 92%
– BVA 67%
– ST 67%
• Batch Processor
– EP 54%
– BVA 58%
– ST 67%
3. How many defects were detectable by
the Atomic Rules approach?• Percentage of defects detectable was identified by applying all Atomic Rules
from EP, BVA and ST to all applicable input fields
• When various test data values could be selected (e.g. selecting a value from
an equivalence class during EP), „mid-point‟ values were chosen
What program characteristics
make a black-box testing
technique more (or less)
effective?
High DDP for EP! Was it due to
Atomic Rules representation of EP
being more thorough than Myers‟?
Or was it caused by the types of
defects present?
Boeing Australia
21 21
4. Were the results affected by domain
knowledge?
Approach Program Defects Detected
Std
Dev
Significant
Difference?
PractitionerAddress Parser 34% 4
YesBatch Processor 18% 6
Atomic RulesAddress Parser 30% 8
NoBatch Processor 40% 15
Results suggest that testers either need
domain knowledge OR a prescriptive
testing technique to be effective
(i.e. Atomic Rules fills the gap)
• After the experiment, the testers explained they were more familiar with
the Address Parser, as they regularly tested addresses as part of their jobs
and since it was based on real-world concepts
• Did this have an impact on percentage of defects detected?
Significant difference
found during Practitioner
testing, but not during
Atomic Rules testing
Boeing Australia
22 22
5. Were results affected by years of
experience in testing?
• Compared percentage defects detected and test technique
coverage by years of experience
Approach n Experience
Technique
Coverage
Std
Dev
Sig
Diff?
Defects
Detected
Std
Dev
Sig
Diff?
Practitioner6 < 2 years 7% 1.5
No22% 10
No5 2 – 10 years 6% 1.7 28% 10
Atomic
Rules
6 < 2 years 9% 5.8No
35% 8No
5 2 – 10 years 13% 6.3 35% 18
Number of years of testing experience did
not affect defect-detection effectiveness or
test technique coverage
No significant differences
were found
No significant differences
were found
Boeing Australia
23 23
6. Were results affected by the tester‟s
current job?• Compared percentage defects detected and test technique
coverage by current job
Approachn Current Job
Technique
Coverage
Std
Dev
Sig
Diff?
Defects
Detected
Std
Dev
Sig
Diff?
Practitioner5 Test Lead 5% 1.8
No19% 0.1
Yes5 Tester 7% 1.4 30% 0.1
Atomic
Rules
5 Test Lead 10% 6.4No
42% 0.2No
5 Tester 11% 7.2 30% .1
Did the Atomic Rules approach fill
the gap in recent test practice for
the Test Leads?
No significant differences
were foundOnly significant difference found
was Defects Detected during
Practitioner testing
Boeing Australia
24 24
7. What was their understanding of the
techniques?• Did their understanding of EP, BVA and ST improve after learning the Atomic
Rules representation of these techniques?
Understanding of Black-Box Testing Techniques
9
8
9
3
2
3
2
2
2
5
4
5
0
1
0
3
5
3
0
0
0
0
0
0
0 1 2 3 4 5 6 7 8 9 10 11
EP Initial
BVA Initial
ST Initial
EP Final
BVA Final
ST Final
Number of Testers
Novice Intermediate Advanced Expert
Results indicate that
their understanding
significantly improved
Boeing Australia
25 25
Other Findings
• Three participants (27%) felt the Quick Reference Guide
was useful for checking test coverage
• Five participants (46%) felt matrices (mapping input fields
to Atomic Rules) would have supported test design during
Atomic Rules testing (for planning & coverage)
• Eight participants (73%) felt participation would impact on
how they perform black-box testing in future
• All enjoyed the challenge of learning a new approach to
black-box testing and/or found the
Atomic Rules approach „interesting‟
Boeing Australia
26
Recommendations for Industry
• Exploratory testing is useful and can be effective!
– Useful for gaining familiarity with a system
– Can be as effective as prescriptive testing, particularly when testers
have domain knowledge in the system
– Training in Exploratory Testing could significantly improve abilities
• Prescriptive black-box testing techniques have their place!
– Can be more effective than Exploratory Testing, particularly when
testers do not have domain knowledge in a system
– Important to have precise definitions of techniques like EP, BVA
and ST, to ensure testing is thorough and auditable
– Testers in industry do enjoy learning and using
prescriptive approaches to black-box testing
Boeing Australia
27
Future Work
• Since the experiment involved a limited number of testers, it would be ideal to repeat the experiment with more practitioners & industry-developed programs
• Additional aspects to assess:
– Why are particular black-box testing techniques more effective against certain types of programs?
– How important is domain knowledge in developing complete and effective test sets?
– Do practitioners retain knowledge they gain through black-box testing training?
– Is automation of test data generation useful for industry?
• e.g. populate parameters in automated test suites that are derived through other means, such as model-based testing or keyword-driven frameworks
Boeing Australia
28
Thank you for your interest in this presentation!
Questions?
Many thanks to the participants of this experiment and to K. J. Ross & Associates & La Trobe University
for their support of this research
Tafline MurnaneK. J. Ross & Associates Pty. Ltd.
PO Box 131, West Burleigh, QLD, 4219
Australia
P: 1300 854 063 F: 1300 854 064
M: +61 416 117 931
http://www.kjross.com.au