Find defects
„ Training
„ Communications
„ Hostage taking
„ Individual:
„ Preparation, or
„ Detection
„ Initial:
„ Check criteria
„ Plan
„ Overview
Process
„
„ Group:
„ Detection, or
„ Collection
„ Inspection record
„ Data collection
„ Exit:
„ Change
„ Follow-up
„ Document & data
handling
Kristian Sandahl, IDA
krisa@ida.liu.se
Kristian Sandahl, IDA
krisa@ida.liu.se
„ Number of inspectors
Kristian Sandahl, IDA
krisa@ida.liu.se
„ Severity
„ Decision for entire document:
„
„ Number of hours
„ Classes of defects
„ Description
Pass with changes
„ Reinspect
„ Number of defects
„ Location
Data collection
„ Scribe
„ Inspector
„ Moderator
„ Author
Roles
„ Identification
Inspection record
Kristian Sandahl, IDA
krisa@ida.liu.se
testing in a tech-report 1985.
„ Graham and Gilb publish the book Software inspections 1993. This
describes the standard process of today.
„ Presentation of the Porter-Votta experiment in Sorrento 1994 starts a
boom for replications.
„ Sauer et al compare experimental data with behavioural research in a
tech-report 1996
„ Basili and Selby show the advantage of inspections compared to
IBM systems journal
„ Fagan publishes results from code and design inspections 1976 in
„ The best way of finding many defects in code and
other documents
„ Experimentally proven in replicated studies
„ Goals:
Development over the years
Systematic inspections
Kristian Sandahl, IDA
krisa@ida.liu.se
Kristian Sandahl, IDA
krisa@ida.liu.se
Id
Kristian Sandahl, IDA
krisa@ida.liu.se
„ Majority values:
„ up to 3.5 h preparation per document
„ up to 3 h inspection time
„ up to 4000 lines of code
„ 2 to 6 people involved
„ 90 percentile = 30
„ Median number of defects = 8
Kristian Sandahl, IDA
krisa@ida.liu.se
„
„
„
„
Kristian Sandahl, IDA
krisa@ida.liu.se
line (12 minutes per page)
Size of document have negative effect on DFR, max
recommendation 5000 lines
A certain project is better than two of the others
4 inspectors seems best (not significant)
Analysis performed by Henrik Berg, LiTH-MAT-Ex-199908
„ Preparation time per code line typically 0.005 hours per
„ Usage-based
„ Perspective-based
„ Scenario
„ Checklist
„ 214 code inspections from 4 projects at Ericcson
Kristian Sandahl, IDA
krisa@ida.liu.se
Class.
Regression wrt defect detection
ratio
Description
Reading techniques
Practical investigation
Loc.
Our inspection record
inspector 3
defects, frequent defects, or
random defects
„ Popular mind map:
The Ishikawa diagram
„ Parameters:
„ Defect category
„ Visible consequences
„ Did-detect
„ Introduced
Main
„ Should-detect
cause
„ Reason
„ Performed regularly for severe
inspector 1
inspector 3
Main
cause
Main
cause
Main
cause
Main
cause
Main
cause
Kristian Sandahl, IDA
krisa@ida.liu.se
Problem
Kristian Sandahl, IDA
krisa@ida.liu.se
Fewer remaining defects
inspector 2
Root-cause analysis
Many remaining defects
inspector 2
inspector 1
Capture-recapture
Kristian Sandahl, IDA
krisa@ida.liu.se
False positives
Kristian Sandahl, IDA
krisa@ida.liu.se
Inspection
data
Coding
Kristian Sandahl, IDA
krisa@ida.liu.se
Inspection
data
Test-cases
Inspection Inspection Inspection
data
data
data
Design
number of inspectors, compare with testing
„ Enthusiasts: Vary things, such as PI – DD,
you got from that
Kristian Sandahl, IDA
krisa@ida.liu.se
„ Record the result as we did, reflect on the value
„ Improvement: reduce variation, increase precision
Analysis
„ Be at least 3 inspectors
„ Control – adjust the process
formal document
„ Practice inspection on code from a lab, or a
„ Appraisal – defect detection
„ Assurance – prediction of defects
Home assignment no 9
Inspections in quality assurance
Defect list
Two experts
„ Audit
„ Review
„ Structured walk-through
„ Good data recording
„ Peer review
Repository
„ Desk check
Weaker methods
„ Good reading techniques
Inspectors
”Optimal” method
„ Calender time
„ Person-hours
Cost of quality
Kristian Sandahl, IDA
krisa@ida.liu.se