This paper discusses how to decide what percentage of code coverage you need.
By Steve Cornett. Copyright © Bullseye Testing Technology 2006-2011. All rights reserved. Redistribution in whole or in part is prohibited without permission.
Code coverage of 70-80% is a reasonable goal for system test of most projects with most coverage metrics. Use a higher goal for projects specifically organized for high testability or that have high failure costs. Minimum code coverage for unit testing can be 10-20% higher than for system testing.
Empirical studies of real projects found that increasing code coverage above 70-80% is time consuming and therefore leads to a relatively slow bug detection rate. Your goal should depend on the risk assessment and economics of the project. Consider the following factors.
Many projects set no particular minimum percentage required code coverage. Instead they use code coverage analysis only to save time. Measuring code coverage can quickly find those areas overlooked during test planning.
Defer choosing a code coverage goal until you have some measurements in hand. Before measurements are available, testers often overestimate their code coverage by 20-30%.
Although 100% code coverage may appear like a best possible effort, even 100% code coverage is estimated to only expose about half the faults in a system. Low code coverage indicates inadequate testing, but high code coverage guarantees nothing.
In a large system, achieving 100% code coverage is generally not cost effective. Some reasons are listed below.
Generally, the tester should stop increasing code coverage when the tests become contrived. When you focus more and more on making the coverage numbers better, your motivation shifts away from finding bugs.
You can attain higher code coverage during unit testing than in integration testing or system testing. During unit testing, the tester has more facilities available, such as a debugger to manipulate data and conditional compilation to simulate error conditions.
Likewise, higher code coverage is possible during integration testing than in system testing. During integration testing, the test harness often provides more precise control and capability than the system user interface.
Therefore it makes sense to set progressively lower goals for unit testing, integration testing, and system testing. For example, 90% during unit testing, 80% during integration testing, and 70% during system testing.
The information in this paper applies to code coverage metrics that consider control structures independently. Specifically, these are:
Although some of these metrics are less sensitive to control flow than others, they all correlate statistically at a large scale.
The aviation standard DO-178B requires 100% code coverage for safety critical systems. This standard specifies progressively more sensitive code coverage metrics for more critical systems.
|Effect of System Failure||Level||Example||Required Code Coverage|
|Catastrophic||A||Crash||100% modified condition/decision coverage and 100% statement coverage|
|Hazardous||B||Passenger fatality||100% decision coverage and 100% statement coverage|
|Major||C||Passenger injury||100% statement coverage|
|Minor||D||Flight plan change||No code coverage requirement|
|No effect||E||Entertainment system failure||No code coverage requirement|
These requirements consider neither the probability of a failure nor the cost of performing test cases.
The standard IEC 61508:2010 "Functional Safety of Electrical/Electronic/Programmable Electronic Safety-Related Systems" recommends 100% code coverage of several metrics, but the strenuousness of the recommendation relates to the criticality.
|Safety Integrity Level||100% entry points||100% statements||100% branches||100% conditions, MC/DC|
|1 (least critical)||Highly Recommended||Recommended||Recommended||Recommended|
|2||Highly Recommended||Highly Recommended||Recommended||Recommended|
|3||Highly Recommended||Highly Recommended||Highly Recommended||Recommended|
|4 (most critical)||Highly Recommended||Highly Recommended||Highly Recommended||Highly Recommended|
The safety integrity level (SIL) relates to the probability of unsafe failure. Determining the SIL involves a lengthy risk analysis.
This standard recommends but does not require 100% coverage. It specifies you should explain any uncovered code.
This standard does not define the coverage metrics and does not distinguish between condition coverage and MC/DC.
The IEEE Standard for Software Unit Testing section 3.1.2 specifies 100% statement coverage as a completeness requirement. Section A9 recommends 100% branch coverage for code that is critical or has inadequate requirements specification. Although this document is quite old, it was reaffirmed in 2002.
The United States Federal Drug Administration FDA document General Principles of Software Validation recommends structural testing but does not specify any particular code coverage requirement. The document says only:
The amount of structural coverage should be commensurate with the level of risk posed by the software.
Efficient use of code coverage in large-scale software development, Yong Woo Kim, 2003
Code coverage, what does it mean in terms of quality? Williams et al, 2001
An Empirical Study of the Branch Coverage of Different Fault Classes, Melissa Cline and Linda Werner, 1994
Coverage measurement experience during function test, Paul Piwowarski et al, 1993
How Do You Know When You Are Done Testing?, Richard Bender, 2000
How to Misuse Code Coverage, Brian Marick, 1997
Comparing the Effectiveness of Software Testing Strategies, Basili and Selby, 1987
DO-178B, Software Considerations in Airborne Systems and Equipment Certification, 1992
Understanding the Use, Misuse and Abuse of Safety Integrity Levels, Felix Redmil, 2000