Embedded Unit Testing Framework

Erik Bergersen

Master of Science in Computer Science

Submission date: June 2013
Supervisor:
Tor Stlhane, IDI
Co-supervisor:
yvind Grotmol, Energy Micro

Norwegian University of Science and Technology

Department of Computer and Information Science

Problem Description
The aim of this project is to make it easy to do thorough unit testing of C/C++ code written for the EFM32 microcontrollers, by creating a testing framework. The core problem
is that the embedded code interacts with a number of peripherals on the microcontroller,
which are not available when running the tests. Thus, it is essential to create mock objects
or stub functions for all EFM32 modules, which allows the test instrumentation code to
control exactly which response is given to the code-under-test (CUT), and also to capture
the messages sent by the CUT in order to compare it with the expected behavior.
The project will see if unit testing can reduce the number of errors before release of
software, and that the work-hours used on testing can be reduced. Show that Unity and
CMock is suitable as a framework for unit testing embedded software. The project will
also look at problems and barriers that prevent developers from using unit testing during
development of embedded software.

Abstract
This thesis addresses the challenges with unit testing of embedded software. Embedded
software uses peripheral devices that are not available during testing, which results in
higher barriers to use unit testing, compared to normal software. But even with these
barriers there are no problems with unit testing of embedded software.
The study looks at challenges with unit testing of Energy Micros software library, and
solutions to the problems with automated testing of embedded software. Use of automated
testing will reduce the number of errors when code is refactored. Use of unit testing will
also make it possible to take advantage of agile methods like test-driven development. The
main goals for the study is to see if CMock is suitable for generation of mock modules
and if testing with mock modules can reduce the number of errors in software.
Unity is a test harness that supports the needed assertions for testing of embedded software. CMock is able to automatic generate mock modules, but has a few problems with
inline functions and custom types, but it is possible to solve these problems. Even with
these problems CMock is able to generate mock modules that is working, and supports
the testing of embedded software.
An experiment was performed to see if testing of embedded software with mock modules
could reduce the number of errors. The result was significant and showed that the number
of errors was reduced by using mock modules, compared development without testing.
Observations were also done during the experiment, which found some problems with use
of mock modules during testing. The experiment was done by undergraduate students in
computer science.
Use of unit test and mock modules as a substitute for hardware during automated testing
on a native computer, it is possible to reduce the numbers of errors and refactor code
without breaking the existing functionality.

Sammendrag
Denne rapporten tar for seg utfordringene med enhetstesting av embedded programvare.
Embedded programvare bruker perifere enheter som ikke er tilgjengelige under testing,
dette resulterer i hyere barrierer for bruke enhetstesting, sammenlignet med vanlig
programvare. Men selv med disse barrierene er det ingen problemer med enhetstesting av
embedded programvare.
Studien ser p utfordringer knyttet til enhetstesting av Energy Micros programvarebibliotek, og lsninger p problemer med automatisk testing av embedded programvare. Bruk
av automatisert testing vil redusere antall feil etter refaktorering av kildekoden. Bruk av
enhetstesting vil ogs gjre det mulig ta i bruk smidig utvikling, som for eksempel testdrevet utvikling. Hovedmlene for studien er se om CMock er egnet som et rammeverk
for generering av mock moduler og om testing med mock moduler kan redusere antall feil
i programvaren.
Unity er et testrammeverk som sttter de ndvendige sammenligner under testing av
embedded programvare. CMock er i stand til automatisk generere mock moduler, men
har problemer med inline funksjoner og hndtering av strukturer og andre ikke primitive
datatyper, men det er mulig lse disse problemene. Sett bort fra disse problemene
er CMock i stand til generere mock moduler som fungere og kan sttte testingen av
embedded programvare.
Et eksperiment ble utfrt for se om testing av embedded programvare med mock moduler
kunne redusere antall feil. Resultatet av eksperimentet var statistisk signifikant og viste
at antall feil vil bli redusert ved bruk av mock moduler, sammenlignet med utvikling
uten testing. Observasjon ble ogs gjort under forsket, dette oppdaget noen problemer
ved bruk av mock moduler under testing. Eksperimentet ble utfrt av studenter ved
datateknikk.
Ved bruke enhetstester og mock moduler nr hardware ikke er tilgjengelig for automatisk
testing p vanlig datamaskin, er det mulig redusere antall feil og refaktorere kode uten
delegge eksisterende funksjonalitet i programvaren.

PREFACE

The report was written as final thesis for the master degree in computer science at Norwegian University of Science and Technology (NTNU). The thesis was carried out in the
spring of 2013 and looked at unit testing for embedded software.
The aim for the project is to do easy unit testing of C/C++ code, which are written for
EFM microcontroller, by studying Unity and CMock as unit test framework for embedded
software. The project was given by yvind Grotmol at Energy Micro.
I would like to thank my supervisor Tor Stlhane for his continuous input and guidance
during the project. I would also like to thank yvind Grotmol and Marius Granns
at Energy Micro for the help during the project. I appreciate that XCOM14 could get
participants to my experiment.

Trondheim, June 10, 2013

Erik Bergersen

ii

CONTENTS

Preface

Contents

iii

List of Figures

vii

List of Tables

ix

1 Introduction

1.1

Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2

Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.3

Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.4

Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.5

Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2 Background

2.1

EFM32 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

Unit Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3

Testing of Embedded Systems . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.4

Existing Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3 Embedded Unit Test Framework

19

3.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3.2

Dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3.3

Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3.4

Testing of Peripherals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.5

Use of Mock Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

iii

3.6

Testing of Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.7

Testing of Drivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.8

Automated Build System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

3.9

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4 Experiment

47

4.1

Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.2

Planning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.3

Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.4

Analysis & Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.5

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5 Discussion

65

5.1

Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5.2

Mock Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.3

Embedded Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5.4

Test Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.5

Agile Development and Test-driven Development . . . . . . . . . . . . . . 70

5.6

Regression Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

6 Conclusion

73

6.1

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

6.2

Further Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

Bibliography

75

Abbreviations

77

A Experiment Guidelines

79

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
A.2 Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
A.3 Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
A.4 When Finished . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
B Experiment Test Cases

93

B.1 Task 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
B.2 Task 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
B.3 Task 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
B.4 Task 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
iv

B.5 Task 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

C Mock Preperation Script

103

C.1 Script . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

C.2 Config-file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

vi

LIST OF FIGURES

2.1

Bit-banding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

Software Layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3

TDD State Machine [16] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3.1

Blink Dependecies

3.2

Blink Dependecies with Mock Modules . . . . . . . . . . . . . . . . . . . . 21

3.3

STK3700 Hardware Layout [12] . . . . . . . . . . . . . . . . . . . . . . . . 35

4.1

Boxplot of Test Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

4.2

Distribution of Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

A.1 Connect with energyAware Commander . . . . . . . . . . . . . . . . . . . . 81

A.2 Upload with energyAware Commander . . . . . . . . . . . . . . . . . . . . 81
A.3 STK3700 Hardware Layout [12] . . . . . . . . . . . . . . . . . . . . . . . . 82
A.4 Text Pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

vii

viii

LIST OF TABLES

4.1

Errors per Participant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.2

Errors per Test Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

4.3

T-test Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

B.1 Test case TC1-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

B.2 Test case TC1-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
B.3 Test case TC1-3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
B.4 Test case TC1-4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
B.5 Test case TC1-5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
B.6 Test case TC2-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
B.7 Test case TC2-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
B.8 Test case TC2-3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
B.9 Test case TC2-4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
B.10 Test case TC2-5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
B.11 Test case TC2-6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
B.12 Test case TC3-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
B.13 Test case TC3-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
B.14 Test case TC3-3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
B.15 Test case TC3-4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
B.16 Test case TC3-5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
B.17 Test case TC4-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
B.18 Test case TC4-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
B.19 Test case TC4-3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
B.20 Test case TC4-4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
B.21 Test case TC5-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
ix

B.22 Test case TC5-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

B.23 Test case TC5-3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
B.24 Test case TC5-4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
B.25 Test case TC5-5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

CHAPTER

1
INTRODUCTION

This chapter is an introduction to the thesis. The chapter contains; problem definition,
motivation, context and an outline of the report.

1.1

Problem Definition

The task was given by yvind Grotmol in Energy Micro. The projects aim is to look
at an embedded unit test framework, which make it easy to do unit testing of C/C++
code. This code is written for Energy Micros microcontrollers. The company needs a
unit test framework to take advantage of agile methods during development of example
projects for Energy Micros development kits, and development of the software library for
the microcontrollers.
One problem with unit testing in the embedded space is that the code interacts with
peripherals on the microcontroller that not can be accessed during tests. To be able to
test this logic there is a need to create stub functions and mock objects that make it
possible to test instrumentation code to check which response is given during the test,
and to capture the messages in order to compare it with the expected behavior.
The thesis will look at different layers of the software library that Energy Micro provides,
with focus on the examples that are provided for the different development boards.

1.2

Motivation

The motivation for solving this problem is that Energy Micro has more than 200 example
projects, which is available for the users of theirs starter and development kits. Energy
Micro has nine different development boards. Many of the examples support several of
these development boards, in addition they have to support four different IDEs (Integrated
Development Environment). The problem with testing is caused by the development
boards and more than 200 example projects. In total this will be more than thousand
binaries, which is a huge amount of binaries to test if changes are made in the software
libraries. Testing can be supported with unit tests that are run automatically, instead of
1

manual system tests, which would be very expensive.

1.2.1

Purpose

The purpose of the thesis is to find methods that can be used to reduce the number of
errors that are created during refactoring and modification of code. By using unit testing
it is also possible to adopt agile methods like test-driven development. If more of the tests
activities can be automated, time used on testing can be reduced.
Use of mock modules/objects will be essential in thesis, since interaction with hardware
is not possible during running of tests on a host computer.
The objectives of the thesis is to verify that Unity and CMock is suitable as a unit test
framework for embedded systems and to see if unit testing can work as a method for
regression testing of a software library for several different microcontrollers.

1.3

Goals

For the project there are both research goals and goals that Energy Micro has for testing
their library with unit testing.
The research goals for the project are as follow:
RQ1 Is it possible to reduce the number of errors during development of new features,
modification and refactoring of existing functionality by using mock modules to
verify the desired behavior of the system?
RQ2 Show that CMock is a suitable framework to create mock objects, which can simulate the behavior of the hardware, which is not available during testing on a native
system.
The goals for Energy Micro are:
1. Reduce the number of errors in the code, before releasing it to the customers.
2. Take advantage of agile development, with unit testing and test-driven development.
3. Be able to refactor or change the code on a later time, without breaking the code.
The first of the goals is connected with RQ1, since they both look at the reduction
of errors. The second goal is accomplished if Unity and CMock is suitable for testing.
The last goal is fulfilled if the framework can be used to create tests, which can be run
automatic on a host computer, for the code that is changed.

1.4

Approach

The thesis consists of two parts, where the first is to use the test framework in practice
to test functionality in the software library. The second part will be an experiment where
we look at how mock modules can reduce the number of errors during development.
2

To be able to use a unit test framework to perform regression testing on a software library
that supports about a hundred different microcontrollers, we need to look at advantages
and disadvantages of this solution. Testing of peripherals, kit specific code and example
projects will be done to find barriers and challenges with unit testing, in order to find
a solution. When testing a software library for microcontrollers, it is impossible to test
all the code, since it will depend on hardware. The focus will be on the mock modules
created by CMock, since this is an important tool to be able to solve the tasks, when
hardware is not available under testing.
The experiment will be used to study the reduction of errors when the developers use unit
tests based on mock modules to add new features to the system, and under modification
of existing features. The participants in the experiment will implement features in a few
tasks that are based on examples for the Energy Micro Giant Gecko STK3700 Starter
Kit.

1.5

Outline

Background gives background information about hardware, software and techniques

that is used in the thesis.
Embedded Unit Test Framework summarizes testing of the Energy Micro software
libraries, and looks at challenges and solutions by using unit testing to perform
regression testing.
Experiment contains the experiment that was performed during the thesis, which includes the definition, planning, operation, analysis & interpretation and conclusion.
Discussion contains the discussion of the results from the testing and the experiment,
and a comparison of these two parts.
Conclusion contains the conclusion, together with further work.
Appendix contains guidelines for the experiment, and the test case used for measurement in the experiment. It also contains source code, tests and scripts as documentation when needed as a reference for other parts of the report.

CHAPTER

2
BACKGROUND

This chapter describes some of the hardware features that are available on the EFM32
microcontroller. The chapter also includes some unit testing theory, how testing is performed on embedded systems/software and a part on how to write tests for code without
tests.

2.1

EFM32

This section gives a short introduction to the EFM32 microcontroller family developed
by Energy Micro, the ARM Cortex-M processor and a short description of a few peripherals that are included on the microcontroller. It also includes a short description of
the development boards that are produced by Energy Micro, the support for IDEs and
description of the parts in the software library that are provided for development on the
microcontrollers.

2.1.1

Energy Micro

Energy Micro is a Norwegian company that was founded in 2007. Energy Micro develops
the EFR (Energy Friendly Radios) family and EFM (Energy Friendly Microcontrollers)
based on ARM Cortex processors [5]. The focus for the company is to provide microcontrollers with low energy usage. This project will focus on the embedded unit testing of
the software for the microcontrollers.

2.1.2

Hardware

The main part of the hardware in this project is the microcontroller, which is based on an
ARM processor. In addition there exists several peripheral devices, and hardware features
that are available on the development boards.
5

Microcontroller
A microcontroller consists of an on-board memory unit, a microprocessor and I/O devices,
and it is designed for low-cost systems [24]. The EFM32 microcontrollers are delivered
with different amount of RAM and flash memory, different peripherals and different packages. All EFM32 microcontrollers use the ARM Cortex-M3 microprocessor.
ARM Cortex-M3
The Cortex-M processor is the processor that is used in the EFM microcontrollers. The
Cortex-M processor architecture is created by ARM, and is made for tomorrows embedded applications [1]. It is designed to be energy efficient, and uses a processor built on
the RISC architecture.
Peripherals
The EFM32 microcontrollers have several peripheral devices. The microcontrollers use
the peripherals for input, output and storage. The peripherals on the microcontroller are
designed for energy efficiency. Below is a list of some of the peripheral device that is
available on the EFM32 microcontrollers:
ACMP The Analog Comparator (ACMP) is a module that compares analog signals and
gives a digital voltage to indicate which of the analog signal that is highest.
ADC The Analog to Digital Converter (ADC) is used to convert analog signal to a digital
signal. The peripheral can give a digital output with a resolution of 12 bits. The
input can be selected from 6 internal signals and 8 external pins [11].
AES Advanced Encryption Standard (AES) is a peripheral for encryption and decrypting. The module supports 128 and 256-bits encryption [11]. The hardware module
uses considerable less energy, than what a software implementation would use.
UART Universal Asynchronous Receiver/Transmitter (UART) is a serial IO peripheral,
which support full duplex and half duplex communication.
LCD This peripheral is used to drive the segmented Liquid Crystal Display (LCD). The
LCD can for example be used to display text.
Watchdog The purpose of the watchdog timer is to reset the system to a known state if
it encounters a system failure. The watchdog has a timer with a configurable period
that is the timeout threshold. The timer will count continuously with the signals
given from an oscillator. The processor must feed the watchdog with a reset signal
before timeout.
Bit-banding
Bit-banding is a technique that is used to map words to bits. This is done since writing
to a single bit will require reading the word, modify and then write the bit back. By using
bit-banding, it is possible to do this in a single operation [11, p. 16]. Figure 2.1 illustrates
bit-banding.
6

Figure 2.1: Bit-banding

2.1.3

Software libray

The software library that is delivered by Energy Micro consists of several layers as seen
in Figure 2.2. Each of these layers is described below.

Figure 2.2: Software Layers

Cortex abstraction layer CMSIS (Cortex Microcontroller Software Interface Standard)

is a hardware abstraction layer for the ARM Cortex-M processors. This layer is independent of the microcontroller and provides an interface to the ARM processor
architecture.
CMSIS peripheral support library This layer consists mainly of memory registers
definitions and bit field definitions. The layer does not include any functionality
that can be tested. Each microcontroller has different specifications in this layer,
depending on the availability of memory and peripheral devices on the microcontroller. The support layer also includes a system layer for the microcontroller on
the ARM Cortex-M3.
Energy Micro peripheral library This layer is a HAL (Hardware Abstraction Layer)
and an API (Application Programming Interface) for each of the peripherals, and
has little logic. The peripherals will be tested later in this thesis, to look at possibilities and problems with testing on this layer. This layer is used by every microcontroller, where the CMSIS peripheral support library defines the ability for the
given microcontroller. Although this layer has little logic, unit tests will be needed,
7

because these are necessary for regression testing, and for testing with test-driven
development.
Board support package This is a support package for the starter/development kit,
which includes support functions and communication package definitions for the kit.
The board support package also includes drivers and examples that are described
below. Each starter/development kit has its own board support package. It may
be necessary to unit test some of the functionality of the board support functions.
Device drivers The drivers are used to demonstrate functionality on the development
board. The device drivers are created for several starter and development kit, and
configuration is used to specify them for each kit. Examples of drivers are segmented
display driver and NAND flash driver. Unit testing of the device drivers is necessary,
and it can be tested on the development board, or with use of mock objects.
Examples The examples are given as a quick start to programming on the development
boards. The examples demonstrate functionality that the development board has.
Each of the kits has 10 - 50 different examples. Some of the examples exist on
several of the kits, but they are different due to hardware and memory differences.
Most of the examples support four IDEs. Some of the examples will be tested later
in the thesis.

2.1.4

Integrated development environment

Energy Micro support four IDEs (Integrated Development Environment) [12]. The support gives the ability to open example projects in the given IDE. The IDEs that are
supported are:
IAR Embedded Workbench for ARM
Rowley Associates - CrossWorks for ARM
CodeSourcery - Sourcery G++
Keil - MDK-ARM
When compiling and building, each of the IDEs will create a binary, which will result in
four binaries for an example. This raises the complexity with testing due to the amount
of binaries that will need testing. Compiling each example for all the IDEs is necessary,
since the IDEs uses different compilers. The C programming language standard specifies
the language, but it does not specify behavior in all cases, the language has cases of
undefined behavior1 , implementation-defined behavior2 and unspecified behavior3 . It is
thus necessary to compile and test on all IDEs/compilers.
1

behavior, upon use of a nonportable or erroneous program construct or of erroneous data, for which
this International Standard imposes no requirements [17, 3.4.3]
2
Unspecified behavior where each implementation documents how the choice is made [17, 3.4.1]
3
use of an unspecified value, or other behavior where this International Standard provides two or more
possibilities and imposes no further requirements on which is chosen in any instance [17, 3.4.4]

2.1.5

Development boards

Energy Micro has several kits that can be used by developers to test and debug the EFM
microcontrollers. Energy Micro provides two types of kits; development kits and starter
kits. The starter kits are simple kits that can be used to try out some of the features
in the microcontrollers. These kits are mostly an evaluation of the microcontroller. The
developer kits are made for advanced users that want to develop embedded systems. The
developer consists normally of more functionality and more advanced features than the
starter kits, and has better support to build custom circuits. With the advanced energy
monitor, the users are able to see the energy usage on the system, and are able to evaluate
and debug with the developer kit. With tools to monitor energy usage, the developers
are able to optimize their system, to use as little energy as possible. The starter and
developer kits differ for each microcontroller family developed by Energy Micro.

2.2

Unit Testing

This section gives an introduction to unit testing, description of a few test doubles and a
description of test-driven development.

2.2.1

Unit testing

Unit testing is a technique for white box testing, where the testers now the internal
structure of the system. Unit testing is about breaking the code down to small units of
code, a unit could be a single method in the code or a complete class (object-oriented
programming). Test cases are written to verify the correct behavior of a unit, and the
test cases are independent of each other. To be able to do unit testing, the developers
will normally need a test harness, which will include functionality to do assertions, setup
for test and afterwards clean-up after the test. In addition they normally have some
functionality for test report.
When unit testing, some part of the system may not be available, because it needs hardware that is not available during testing, functionality that not are fully developed yet or
using functionality like network or databases that will result in that tests uses too long
time to finish. For these parts it could be necessary to use mock objects, test stubs or
other test doubles to simulate and be able to run the tests. The benefits of using unit tests
are that it is possible to test and find errors early in the process, and it can be used to
check that code is not broken during changes or refactoring of the code. Unit tests itself
is documentation that the code is tested and should behave like the developers intended.
Using unit testing during development on embedded system will reduce the number of
times it is necessary to compile the code for the hardware, upload it and then run the
tests on the hardware. During testing with unit tests it is important that the code is
testable, by structuring the code so only one layer do access the hardware, which will
improve the design of the software. When a problem is identified it is easier to fix errors
that are discovered by the tests, than having to debug on the hardware. This will reduce
the amount of work-hours the developers use. During development of hardware that is not
ready for production, it is possible to start implementation of software and test it using
unit test and mock objects, before the hardware is ready, which mean that development
9

can start early in the process.

2.2.2

Test doubles

During testing of code, several collaborators are working together, like methods, classes,
data and modules. The unit that is tested with a unit test depends on several of these
collaborators. A test double represents and acts like a method, class, data or module [16].
There exist many kinds of test doubles; this part will only look at stubs, fakes and mock
objects.

Stubs
A stub is a function that is created to only return a value, without any functionality.
Similar to a test dummy, that is a function that only is created to be able to compile/link
an application [16].

Fakes
A fake object is an object that is used to break dependencies during testing. A fake
is a collaborator that acts as an object or a module during testing [15]. The fake has
implemented some functionality, but is not fully implemented, since the main purpose is
to break the dependency. The fake can be used to break the dependency to for example
hardware, which will make it possible to run the test without hardware. The fake will
never be able to detect any errors, but will act as a collaborator that may simulate some
form of behavior.

Mock objects
A mock object is an advanced version of a fake module/object. The difference is that
the mock object/module uses internal assertions [15]. When using the mock objects, the
developers sets the expectations and inject return values before the unit tests are run,
afterwards the expectations are verified. The verification in mock object will check that
the method is called and will verify the parameters with assertions. The mock module
will also verify that the method is called the right number of times. It can also verify that
function calls are done in the correct order.
When using unit testing, it is necessary to use objects and modules that can behave
like components that are not ready for testing or is not available during automated unit
testing, for example hardware. The mock object can be used to describe the call between
modules, and the expected parameters and return values [16].
The programming language that is used for the embedded software in this thesis is C. Since
C is not an object-oriented language, mock object will be referred to as mock modules
instead.
10

2.2.3

Test-driven development

TDD (Test-driven development) is a software development process. The purpose is to

ensure that all code that is written is being tested. The main idea is that you write code
if you have a failing test, and a cycle in the development is finished when all the tests are
passing. A cycle in test-driven development is commonly known as red-green-refactor [16].
This means that you first write a test that is failing, the reason why it is failing should
be a compilation error, because the compiler cannot find the definition of the function
the test is calling. The developer should then fix this error by adding a stub function,
run the tests to verify that they are green. Then the tests should be changed so it asserts
values for different input values. When the code is red again the feature can be fully
implemented. When the tests are run again they should be green and the developers are
able to refactor the code, if it is needed. When using this TDD cycle, it will be possible
to discover and eliminate bugs early. Figure 2.3 [16] display a state machine to describe
how test-driven development is used in C programming.

Figure 2.3: TDD State Machine [16]

When using a process like TDD during development on an embedded system, the developers will have to write code that is intended to run on two different systems; the
development system and the hardware that will be used for production. With this ready
the developers can run tests during development, which will be used to verify that the
production code is correct without the need of an embedded device.

2.2.4

TDD for embedded software

It is possible to use test-driven development for embedded systems. Each time new
functionality is created, a test will be created for this function. When functionality is
refactored, modified or functionality added, the unit tests can check that the functionality
that existed before modification is still working afterwards. This can be used as automated
11

regression testing of the software. To ensure that a system with low number of errors are
delivered, system tests will still be necessary, but with the automated regression tests,
the time that will be needed to use for system and acceptance testing will be reduced.
Since one of goals is to take advantage of agile methods, test-driven development is a
good suggestion. TDD can also be used together with other agile methods like Scrum or
Extreme Programming.
Test-driven development is a method that will help to ensure that the software is welltested, which can be used to find errors in the hardware. These errors can be found
since well-tested software often are easier to understand, and confidence with that code
is well-tested, will prevent the developer from starting extensive debugging, because the
developer thinks the error is within the software. In Effective Test Driven Development
for Embedded Software [18] there is suggested a four level strategy, the first level is
automated unit testing and for each level more human interaction is necessary. The levels
described are:
Automated unit testing Unit tests are added during development, and mock objects
generated when needed. The mock can be generated in a MCH design pattern
to be able to decouple the logic from hardware. The automated unit tests are
run by the developer. The tests are run on development boards, emulators or as
cross compiled tests on a computer. The tests can also be automatic tested by a
continuous integration tool.
Hardware level testing A combination of unit testing and interaction by the tester is
used to test user interaction and hardware. The most important part of this testing
is the hardware functionality [18]. These tests will be run infrequently compared to
automated unit testing.
Communication channel testing Use developer tools to capture test results from the
embedded device. Both hardware and software will be tested, and it will require
interaction from the testers to create communication events during testing.
End to end system testing This layer executes the scenario tests, which is to test the
user scenarios that are defined for the system. Important things to test are timing
issues, user interface and the responsiveness of the system.

2.3

Testing of Embedded Systems

This section will briefly look at unit testing on embedded systems, and how system testing
can be done for software and hardware.

2.3.1

Hardware testing

Testing of embedded software is in many ways different from regular software for personal
computers. Some of the differences are the architecture, platform, processing speed, memory available and storage capacity. Debugging is also different, on software for computers,
debugging can be done through the IDE. For embedded systems there will be a need for
a development board with a debugging interface, for example JTAG, which can be used
12

for example for single stepping and break pointing when debugging on the embedded
device. JTAG was developed as a test interface, but is also widely used as an interface
for embedded debugging [22]. When testing embedded software, testing must cover both
hardware and software, since errors can exists in both or in a combination of them.
In Effective test driven development for embedded software [18], three methods of testing is described: ad-hoc testing, debugging and final testing. The article describes the
current state of these and the shortcomings with the three methods, when testing on
embedded systems.
Ad-hoc testing and experimentation is often used during development to discover infrequent cases in the system, usually to test system boundaries. The knowledge that is
gained from such testing and experimentation during implementation of the functional
code will usually be discarded or shelved. So later in the development process this knowledge is lost, and work-hours will be used later during development, due to the fact that
tests are not kept updated [18].
Since errors can come from both hardware and software, embedded software uses many
debugging and inspection tools during development. There exist many advanced debugging tools for embedded systems. The effect of these tools has however, some bad side
effects. The use of debugging tools can result in that the software is designed for debugging and not for testability. Finding errors with debugging is difficult, since it is unknown
where the error is. Debugging to find a specific error is something that is done once, it
is not necessary to design the code for debugging when the errors is found and corrected.
When using debugging widely to test and develop, it will often result in bad coding practices, because the code is not designed to be testable, so hardware and software is not
decoupled [18]. A positive effect with debugging is that it can be used to find errors, when
developers not able to understand the results from a given input. Another weakness is
that test equipment for microcontroller can be expensive, so it will be impossible for low
budget projects [14].
Development processes like waterfall are often used during development of embedded
software systems. The process starts with a design of the system, then development of
the system, testing of the system and finally production. Testing is a costly stage of the
development and it is difficult to find the correct point for when the testing is finished
and the system is ready for production. Since testing is one of the last stages, testing can
be reduced for projects that are delayed and need to meet a deadline. The result will be
that the system is not tested well enough and can contain critical errors. When testing is
done late in development, there is a chance that the code is not designed and developed
for easy testing [18].

2.3.2

Unit testing

There exist several unit test frameworks created for testing of embedded software. There
are different ways in which unit tests can be used to test embedded software. Functionality and Design of the CMock framework [20] lists three common soultions:
A development board or a development kit as described in the preliminary study
chapter can be used for automated testing of the embedded software. The test
binary is deployed on the development board and then executed, the results from
13

the tests are written to a memory location or transferred back to the computer
through a debugging interface. Testing on a development board is a slow method
since it requires deployment to the development board, and the embedded system
usually has a processor that is much slower than a normal computer. The big
advantage is that the test is run with access to the peripheral devices.
An emulator that can be used to simulate the behavior of the hardware. The
advantage from the development board is that the emulator is much faster, but like
the development board it will need a debugging interface to transfer the results from
the tests. A debugging interface is not necessary a feature that is provided by the
emulator and the support for peripheral is not always available, or is only partially
implemented.
The last solution is to run the tests on the local computer. This is the fastest
solution. But the tests have no access to peripheral devices and real hardware. To be
able to test with usage of a peripheral device, there is a need to develop mock objects
and stub functions to simulate the behavior, but the results may not be the same as
on the embedded device. There might be problems with the compatibility between
the embedded device and the local computer, since the platforms are different and
uses for example different sizes on data types.

2.3.3

System testing

System testing looks at testing at all the stages in the development of an embedded
system.
Hiearchy of testing
Testing Complex and Embedded Systems [19] explains a bottom to the top hierarchy
of testing software for embedded software. The different levels of testing are described
below:
Unit Testing This is testing of the smallest units of software. This level includes build
tests and regressions tests, where the build tests is responsible for testing the functionality that was changed, and regression test is run to verify that no other software
components was affected by the changes in the code.
Component Testing Is testing on a higher level, where tests are focused on the components instead of the units.
Integration Testing Test the integration of software and hardware, to test interaction
and that the systems behavior is as intended.
System Testing This is integration of the subsystems, which includes multiple hardware and software components. Some suggested system level options are: FAT
(Factory acceptance testing), OT (Operational testing) and MQT (Manufacturing
qualification testing).
Field Testing Testing of software in real environments, the system is complete and is
tested as it would be used in real-life.
14

Simulation and emulation

The goals of simulation are to derive requirements and find unknown interactions and
details in the system design [19]. The simulation is often a model of a function or process.
The requirements that are derived are used to create the tests that are needed in the
development process.
Emulation is different from the simulation, since an emulator is a replica of software/hardware that is used for debugging and testing [19].
Model to final product
Development of embedded software needs many test activities, during the different stages
of the process. In Testing embedded software [13] a process is described for building
embedded systems in stages. These stages are simulation, prototyping, pre-production
and post-development.
When the requirements for a system are finished, a simulation model is created to verify requirements and to optimize the system design. The simulation is usually run in
dedicated software tools to simulate the model [13]. The simulation is executed in three
stages.
One-way simulation: Testing of the system in isolation, where only one input is
tested at a time. Tools are used to generate input to the model.
Feedback simulation: Testing the interaction between different parts in the system.
Initially the system should be simulated only with needed parts, environments is
not part of this simulation.
Rapid prototyping: Testing of the system with a real environment. This is done
by connecting the model to the real environment by connecting the computer to
actuators and sensors.
The second stage in the process is to release a unit for pre-production, and validate the
model used in simulation. The model is incrementally replaced with real hardware and
software. The prototype is run on either the real hardware or with an emulator on a
host computer. This stage consists of several test levels; software unit testing, software
integration testing, hardware integration testing and environmental testing [13]. For each
of the levels the software, processor, hardware and the environment is gradually replaced
from simulation to experimental prototype to the target.
The goals of the pre-production stage are to verify that all requirements are implemented
and the system is working as expected. The system is tested with real hardware and
software in a real environment. Examples of test techniques that can be used are real-life
testing, random testing and fault injection [13]. Examples of test types in this stage are
acceptance tests and safety execution tests [13].
The previous stages have made the system ready for production. The post-production
testing ensures that production facilities have high enough quality. The stage is based on
development, testing and inspection of production facilities.
15

2.4

Existing Code

Adding unit tests to existing code that is without tests and not designed for testing can
be difficult. For Energy Micros software there exist only a few tests, which mean that
starting to use agile methods like test-driven development may be difficult, and since only
a few tests exist there is also a chance that the code is not designed for testability. The
ability for testing is important to able to use unit tests and by having unit tests, these
can be used as regression testing when software is changed.

2.4.1

Why change software?

There is four reasons for changing software; adding a new feature, fixing an error, improvement of the design and optimize the performance [15].
Adding a new feature is one of the most common reasons to change software, and when
adding new features these should always be tested by unit testing. A new feature can be
both changing and adding of behavior, which would result in a need to test dependencies.
Testing is important when fixing an error. If there already is a test for this code unit, the
test should be changed so it tests the wanted behavior of the function. If not, we should
write a test to be able to test the corrections that are done in the code, and to be able to
execute regression tests later.
Improvement of the design can result in that features are lost and result in an error. When
design is improved without losing behavior it is called refactoring [15]. To be able to do
structural changes there is a need to have unit test to perform regression testing. When
the structure is changed in the code, there may be need to change existing tests.
Improving the performance of code is like improving design, but the goals is not the
same. When algorithm is changed or the structure is changed to optimize the code, it is
important to have tests to be able to do regression testing, to ensure that behavior of the
code is not changed.

2.4.2

Dependencies

When working with code, dependencies to other libraries and functionality is one of the
biggest barriers for testing. When unit testing, you want to use units that are as small
as possible, the units dependencies can results in many interactions or interaction with
functionality that is not available under tests, like hardware. To be able to test a unit
you do not want to make any changes to the code, so it is necessary to look for seams in
the code. A seam is to change the behavior of the program, without doing any changes
in the code you want to test [15]. In C programming there is two good ways to do this.
One of the solutions is to do the modifications of the programs in the preprocessor. The
preprocessor can be used to define variables and add addition behavior. It is also possible
to use macros to change existing parts of the code. In chapter 3, a macro is used to change
the Delay() method, so delays do not occur in the loop during testing.
By using static linking it is possible to link other libraries or functions during the linking
part of compilation. This can be done with the mock modules that are generated with
CMock. Instead of using the actual module, the mock module is linked to create the test
16

binary.

2.4.3

TDD with existing code

Starting to use test-driven development with existing code without tests, can be a problem
to find where to start. In Test-driven development for embedded C [16], a policy for
adapting test-driven development is suggested for a team that has code without tests:
For new code, use test-driven development.
Before modifying the code, add tests for the existing code.
Test the code after modification.
The first point states that when new functionality is developed, the development will
follow the default cycle of test-driven development, where the cycle starts with a failing
test. The second point suggests that tests for the existing code must be written before
any modification is done to the code. When tests have been added, the code can be
modified to support new functionality. After modification the tests can be run. After
this modification, more of the source code is covered by unit tested and the developers
have tested both new and modified code with unit tests. The goal when using TDD on
existing code should be to get the module that is changed under test.

17

18

CHAPTER

3
EMBEDDED UNIT TEST FRAMEWORK

This chapter discusses possible solutions for testing embedded software. Energy Micros
software library consists of different layers, but the most important part is testing of
examples projects that are delivered to the development boards.

3.1

Introduction

The purpose of testing the software library is to look at problems with unit testing of
a software library for more than hundred microcontrollers, and to find solution to these
problems. This chapter will also look at how suitable Unity and CMock is for testing a
library like this. If a sufficient part of the software library can be unit tested, it is possible
to use these tests for regression testing, since regression testing of the software library is
impossible with more than a hundred microcontrollers after a modification in the library.
The section will mostly focus on software unit testing, but in some parts also look at
software integration testing and integration testing. In addition it will be necessary to
see if the examples for each of the kit can be tested with unit tests.
To be able to run tests that supports different microcontrollers, it will be necessary to
look at an automatic build system to make testing easier and reduce the number of tests
that are run after each modification, Ceedling is a tool that will be looked into here.

3.1.1

Testing of the software library

The layers that will be looked at during testing are the examples, device drivers, board
support package and the Energy Micro peripheral library. Testing of the examples and
drivers is the most important, because most of logic is placed in these layers.
There will be not testing of the Cortex Abstraction Layer, since this layer should be tested
with access to hardware. The CMSIS Peripheral Support Library will not be tested since
it mainly consists of definitions of memory registers and bit fields.
19

3.1.2

Literature

To be able to solve problems that occur during testing, two books have been used to find
ideas for how to solve the problems. During testing of software, different problems will
be discovered, and it might be possible that the solution of the problem can be found in
these books. The two books are listed below:
Test Driven Development for Embedded C by James W. Greening [16].
Embedded Testing with Unity and CMock by Mark VanderVoord [21].

3.2

Dependencies

Dependencies to other modules and libraries are one of the barriers to unit testing. During unit testing the tester wants to test the functionality for the given unit, and is not
interested in the effects of the dependencies. Testing with all the dependencies will results in slow tests because they test much more functionality than the actual unit, and
the functionality will be tested over and over again. Dependency injection is one way to
solve this, but we will need a structure of the code for these, and the method is mostly
used in object-oriented languages. Stub functions, fakes and mock modules can be used
for this during testing, where these is modules is set by static linking or use of macros.
Below follows an example over the dependencies for the blink example. Figure 3.1 shows
the dependencies for the simplest example, which is blink. Blink is actually only a template to start developing on the starter kit. The figure shows the dependencies to other
modules, but it does not include inline methods that are defined in other header-files,
which also are dependencies. If the inline functions were added there would be at least
dependencies to the bit-banding peripheral and the device peripheral.
Figure 3.1 shows that the blink example needs the board support package for the LEDs
and Trace. In addition it need CMU (Clock Management Unit) for to set up a clock
for the delay functionality in the loop. The board support package for the LEDs needs
GPIO to set the pins to be able to turn on and off LEDs on the development board
and it also needs to enable clocks in the clock management unit. Both CMU and GPIO
have dependencies for assert methods. None of these dependencies are needed when
doing software unit testing, since they test much more that the actual unit under test.
Because of this it should be possible to use stubs, fakes or mocks to remove all of these
dependencies. During software integration testing there will be a need to test all these
dependencies, since it will be necessary to test the interactions between the modules.
Figure 3.2 shows the dependencies for an example under test that uses mock modules.
They are actually not dependencies but a module the capture the calls and compare them
with the data and calls that the test expects. The mock module will in addition inject the
response to the unit under test. By doing so it is possible to unit test the software without
dependencies. When there is a need to do software integration testing, the mock modules
will prevent testing the interactions between the software subsystems. When testing the
blink example there will be need for three mock modules; the LEDs on the development
board, the clock management unit to verify that the clocks is set as expected, and the
trace functionality in the board support package.
20

Figure 3.1: Blink Dependecies

Figure 3.2: Blink Dependecies with Mock Modules

21

3.3

Software

This section explains the tools, test frameworks, compilers and IDEs that have been used
for testing of the software library. Some parts only explains which tools that have been
used, and other give a wider description of how these tools can be used.

3.3.1

Unity

This unit test framework is written in the C language [7]. It is a small framework that
only consists of a source file and two header files. The framework has features that are
special for projects in the embedded space. Unity also includes helper tools. An example
of a helper tools is the test runner generator, which generates all necessary code after the
tests are written.
Unity has a lot of different assertions to create test for booleans, integer, floats, bitwise,
integer ranges, strings, pointers and memory assertions. Integer variables can be tested
at given bit-size, hex-values and of signed/unsigned type. Examples of some of these are
given below:
TEST_ASSERT_TRUE Assert that a variable is true.
TEST_FAIL Will always return a test failure.
TEST_ASSERT_EQUAL_INT64 Compares two 64-bit signed integers.
TEST_ASSERT_INT_WITHIN Checks that an integer is in a given range.
TEST_ASSERT_BITS_HIGH Checks that the correct bitmask is set to high.
TEST_ASSERT_FLOAT_WITHIN Compares two floating values, with a given
fault tolerance.
TEST_ASSERT_EQUAL_STRING Compare that two strings are equal.
TEST_ASSERT_EQUAL_PTR Checks that two pointers are the same.
TEST_ASSERT_EQUAL_MEMORY Checks two blocks of memory, to compare
packed structs and buffers.
There are extensions to Unity, which make it possible to use test fixtures. Other assertions
and options can be found in the Unity summary [8].

3.3.2

CMock

CMock is a software library used to generate mock modules from C header-files. CMock
uses Ruby to generate mock modules from the function declarations that are specified in
the header-files [3].
When CMock has generated mock modules, the next step will be to compile and link
the mock modules instead of the real modules. Header-files are still used for declaration
22

of the methods. In the tests, the expectations of methods with parameters are set and
return values set for non-void methods. The method under test is then called and the
unit test is ready for compilation and testing.
When the tests has run it will give feedback if the function was called as many times as
expected, if not it will return that it was called less or more times than expected. Return
values will be injected when this is specified. All parameters will be asserted against the
expectations that were set. It is also possible to configure CMock to verify that methods
are called in the correct order.
Parsing
CMock is limited in parsing C. The way CMock do parsing is to use regular expression
to parse the header-files. This limits the support to what it actually needs, but anyway
it supports most of the necessary functionality. One point where CMock lacks support is
parsing of custom data types, for examples structures, which becomes difficult to compare.
Instead of parsing header-files with regular expressions, CMock could have used a C parser
written in ruby, but currently there is none that supports most of C99 language. The
developers have earlier announced that support for structure is on the way [9].
Mock types
CMock has five mock module types that can be generated. These are expect, array,
callback, Cexception and ignore [4]. Expect is the basic function, that is used in most
of the tests, it defines that the function shall be called, and all the parameters to the
method are asserted. The array type takes an array as parameter, together with the size
of the array and asserts all of these elements. The callback function will take a callback
function as a parameter, and call this stub function when the mock is called. Cexception
is a function that can be used together with Unity and CMock to create exceptions,
Cexception is not used in this thesis. The last function is ignore, which ignore all calls to
the mock function. An example of each of these functions is showed in Listing 3.1.
1

/* Expect */
void print_Expect(params);
void print_ExpectAndReturn(params, return_val);
/* Aray */
void print_ExpectWithArray(*ptr params, size);
void print_ExpectWithArrayAndReturn(*ptr params, size, return_val);
/* Callback */
void print_StubWithCallback(CMOCK_print_CALLBACK callback);

11

13

15

17

/* CException */
void print_ExpectAndThrow(params, throw_value);
/* Ignore */
void print_Ignore();
void print_IgnoreAndReturn(return_val);

Listing 3.1: Mock Module Types

23

Configuration
The configuration of CMock is done with YAML , where the configuration file is used
as input to CMock. Some of the features that can be configured is that attributes can
be ignored, configuration of which plugins to use for mock types, how types and custom
types shall be threated and configuration of how strict the mock module should be with
verifying call in the correct order. For a full overview of configurations is CMock, see the
documentation [4].
Structures
The biggest problem with CMock is how it handles comparisons of structures. CMock
do comparisons of structures, by comparing the memory byte by byte. This will result
in problems when the structures are not packed. On embedded systems, structures will
often be used as a memory register, and packing memory register can result in incorrect
testing since packing a structure will result in a register that is different from on the
embedded system. This can result in testing on a model that does not represent the
actual model on the hardware. An unpacked structure will contain garbage values where
data alignment/padding is necessary, which will make a test fail. There are some ways
this can be solved:
If it is possible to pack the structure, add -fpack-struct as an option to GCC, which
will force that the structure is packed.
Create a helper method for Unity that do the assertions, and add this to the configuration file for CMock.
One possible way is to initialize the structure with calloc(size_t, size_t),
which will initialize the memory block as zero.
If it is not important to test the feature, ignore the mock module.
Inline functions
Inline functions are often used when programming in the C language, for functions that
are often used. Since function call are expensive, inline functions can be used instead so
the function are included in the code, instead of using a function call. static inline
can be used to be make the definition private to one translation unit1 .
In Energy Micros software library static inline is often used, which will make the
code faster og by that use less energy. When generating mock modules with CMock this
is a problem.
CMock ignores inline functions by default, because if a mock function of the function
was created, it is not defined which of the functions that will be used by the compiler
[10]. The way to solve this is to wrap the definition into a macro and define a function
that will be used during test, as shown in Listing 3.2.
1

A Translation unit is the output from preprocessing, that will be used by a compiler to create an
object file.

24

#ifdef TEST
void func_write(uint32_t *addr, uint32_t value);
#else
__STATIC_INLINE void func_write(uint32_t *addr, uint32_t value)
{
*((volatile uint32_t *)addr) = (uint32_t)value;
}
#endif

Listing 3.2: Redefinition of Inline Method when Using Mock Modules

3.3.3

Ceedling

Ceedling is an extension to RAKE [6], so it works as a build system for C projects.

Ceedling is created to support test-driven development in C and is designed to connect
Unity and CMock [2]. The functionality that Ceedling can support when running tests,
is to list all test tasks, run all tests, run a specified module or run tests that depends on
changes since last time the tests was run.

3.3.4

Compilers & IDEs

In addition to software used for testing there have been used several IDEs and compilers.
GCC have been used for compiling and linking the tests that are run on the host computer.
Compiling and upload in Windows was done with IAR, in Linux have CodeSourcery been
used to compile the examples.
The ruby language has been used to create necessary script and to run the scripts that
are used by Unity, Ceedling and CMock.

3.4

Testing of Peripherals

This section covers testing of a few peripheral APIs. The section will look at features that
are difficult or impossible to test, and how to solve some of these problems. The APIs
that was tested was the watchdog, the peripheral reflex system and the AES (Advanced
Encryption Standard) peripheral. All the source code in this chapter is taken from the
Energy Micros software library, and modified to enable testing of the functionality.
The peripheral APIs contains little logic, and is mainly a hardware abstraction layer. The
value of testing this functionality outside the hardware has little value. But it will still
be necessary to test this on a local computer to support test-driven development, where
testing is performed during development.

3.4.1

Watchdog

The watchdog is a peripheral that will reset the system in case of a system error. The
main functionality for the watchdog is that it uses a clock to count towards a timeout,
without resetting the timer, the watchdog will trigger a timeout. The watchdog can be
25

initialized with different settings for energy modes, extern clocks, enabling, locking and
period for timeout.
Unit testing
Most of the assertions that were done by the unit tests in to check that correct bits are set,
since the main functionality for the watchdog code makes changes to the control register
for the watchdog. Reset of the counter is done by writing a bit to the command register.
The watchdog do write operations to the low frequency domain, and need to wait to all
previous write operations are finished. Waiting is done by a while-loop that checks if the
SYNCBUSY register is set. There is no good solution to test this, one possible solution
is to test this is to use threads. One thread will be used to go into the waiting-loop, while
the other thread is used to clear the SYNCBUSY register and verify that other thread
waits until the register is cleared. Another solution is to change the code structure to
make it more testable, which can be done by adding methods that read and write to the
SYNCBUSY register, by having a method for accessing the register, it is possible to create
a mock object, where the return values can be injected. In this way several expectations
can be set and verified. The best solution to check that this is done is to test the feature
on hardware and by using code review, but unit test to be run for regression testing may
also be important.
A mock module was generated with CMock and used for the bit-banding, because the
mapping between the peripheral registers and the bit-banding is not available without
hardware. Use of a mock module work, but the generation had some problems, which is
described below.
Problems with testing
In context of the whole software library, the watchdog is a small and simple peripheral
device. During unit testing of the software functionality for the watchdog, we found a few
problems. Which are listed below.
Bitbanding Bit-banding is a technique that is used to modify single bits in memory
registers or memory. With bit-banding it is possible to write a word to a memory location, the hardware is then responsible for setting the correct bit, by using this technique
updating a bit can be done in one operation.
The bit-banding module consists of two static inline functions that take the address of
a register, the bit in the register and the value as parameters. The function will then
calculate the memory address in the bit band and write the value as a 32-bit unsigned
integer. The implementation can be seen in Listing 3.3.

26

__STATIC_INLINE void BITBAND_Peripheral(volatile uint32_t *addr, uint32_t

bit, uint32_t val)
{
uint32_t tmp = BITBAND_PER_BASE + (((uint32_t)addr - PER_MEM_BASE) * 32)
+ (bit * 4);

*((volatile uint32_t *)tmp) = (uint32_t)val;

6

Listing 3.3: Implementation of BITBAND_Peripheral

Since writing from the bit band is a hardware feature, this will not happen during unit
testing on a computer. To verify that the software actually modifies the bit, there are
three possible solutions:
Create a mock module/object that sets the expected behavior, to check that functions were called with the correct parameters.
Allocate memory for the bit band, and use unit test assertions to check that the
values are set in the memory address.
Create a function to be used under testing, that takes the same parameters and
then set the correct bit in the watchdog memory registers, and run the unit test
assertions. An example is shown in Listing 3.4. Creating a function for this can
be dangerous since you also add functionality that only are used during testing,
which means that errors could occur in the test function and we get more code to
maintain. It is also possible that the implemented code is not correct according to
how the hardware works.
1

#ifdef TEST
void BITBAND_Peripheral(volatile uint32_t *addr, uint32_t bit, uint32_t val
)
{
uint32_t tmp = (uint32_t)*addr;
if (val)
tmp |= (1 << bit);
else
tmp &= ~(1 << bit);

11

13

15

17

19

21

*addr = tmp;
}
#else
__STATIC_INLINE void BITBAND_Peripheral(volatile uint32_t *addr,
uint32_t bit,
uint32_t val)
{
uint32_t tmp = BITBAND_PER_BASE + (((uint32_t)addr - PER_MEM_BASE) * 32)
+ (bit * 4);
*((volatile uint32_t *)tmp) = (uint32_t)val;
}
#endif

Listing 3.4: Redefinition of BITBAND_Peripheral

27

Mock module for bit-band Automatic generation of mock module with CMock for
the bit-band header-file is not possible without modification of the header-file before
generation. CMock takes header-files with function prototypes as input.
The bit-band functions are defined as static inline to improve the performance
for this functions, since function calls are expensive. The compiler will try to inline
the function in the code when inline is used. In the C language, static inlinefunctions are often implemented in the header-file, so it is possible to include them.
CMock do not accept implemented functions as input, which is a problem. Another
problem is that CMock also ignore functions defined as static inline. It is therefore
necessary to modify the header-file before generation to only be function-prototypes, and
replace the static inline with void. After this is done it is possible to automatic
generate mock modules.
When the input file for generation is modified, it is necessary to do other modifications,
to actually use the mock module. The function prototypes used must be included in the
header-file for the mock module. The functions that actually use bit-band, must include
the header-file for the mock module, instead of the original file, therefore the include must
be modified, an example of this is shown in Listing 3.5.
1

#ifdef UNITTEST
#include "Mockem_bitband.h"
#else
#include "em_bitband.h"
#endif

Listing 3.5: Macros Used to Set Mock Module Under Test

Syncronation to low frequency domain A problem is that SYNCBUSY is defined

as read-only (violatile const), this can be modified by wrapping the definition into
a macro where it will not be defined as read-only. Another solution is to do an implicit
cast to void *, which is not a safe method. This will overwrite the data and may result
in the application crash, depending on where the data is stored. Listing 3.6 shows how
the access restrictions are changed during test, to make it possible to write to values that
are defined as read-only.
1

#ifdef __cplusplus
#define
__I
permissions
#elif TEST
#define
__I
C only)
#else
#define
__I
permissions
#endif
#define
__O
permissions
#define
__IO
permissions

volatile

/*!< Defines read only

*/

volatile
*/

/*!< Set read/write under testing(

volatile const
*/

/*!< Defines read only

volatile

/*!< Defines write only

*/

volatile

/*!< Defines read / write

*/

Listing 3.6: Modification of Access Restrictions

28

Memory registers The memory registers for the watchdog is given the address it has on
the bus on the microcontroller. To be able to run test, it is necessary to give it a valid address on the computer that it is executed on. This is done by declaring a pointer, see example in Listing 3.7. For the setup(void) for the test it is now possible to allocate memory
to the memory registers, which can be done by using calloc(size_t,size_t). This
will initialize the memory to zero. When using calloc(size_t,size_t) it will be
necessary to call free(void*) deallocate the memory block, this is important to avoid
memory leakage.
1

#ifdef TEST
WDOG_TypeDef *WDOG;
#else
#define WDOG
pointer */
#endif

((WDOG_TypeDef *) WDOG_BASE)

/**< WDOG base

Listing 3.7: Modify Address of Memory Register

Hardware specific functionality All watchdog functionality cannot be tested, since

it depends on hardware. The reset signal that is given when the watchdog times out is
not possible to test, since it only triggers an interrupt which results in a reset. This is
functionality that is done by the hardware, and cannot be tested by using unit tests.
One of the registers in the watchdog is SYNCBUSY, this is a register where data is set
by the hardware during writing to the low frequency domain. When hardware is not
available, this functionality cannot be tested without modification.
The mapping between memory registers and the bit band for peripherals is hardware
dependent, so this mapping will not work under testing.

3.4.2

Peripheral reflex system

PRS (Peripheral Reflex System) is a peripheral API that makes it possible for peripheral
devices to communicate without use of the CPU. The PRS has 12 channels that can be
configured for sending pulse or level signals [11]. The API consists of methods that are
used set signal sources, level and pulse triggers.
The first part that was necessary to fix before testing, was to define a pointer for the
type declaration for the PRS registers, instead of using a base pointer that would result
in segmentation faults. This was done in the same way as in Listing 3.7. The next step
is to create a test file with set up and tear down, that first initialize the PRS registers
using calloc(size_t, size_t) to have zero-initialized register, in the tear down it
is necessary to free this memory block with free(*ptr). The next step was to create
all the tests and generate a test runner and then compile and run tests.
Since the methods in the API only writes to registers that are read- and writeable, testing
of this peripheral was easy. Methods are called with given parameters, and assertions are
done on the registers afterwards to verify that they were set correctly. Testing the channel
arrays in the registers worked also without any problems. The purpose of having unit tests
for this API is to be able to do testing during development, to verify that no functionality
was broken.
29

3.4.3

Advanced encryption standard

AES (Advanced Encryption Standard) is an encryption standard that is implemented in

Energy Micros microcontrollers. AES is added as en peripheral device to reduce the
need of CPU usage during encryption and decryption and thereby reduce energy usage.
The microcontroller support 128-bit and 256-bit encryptions [11]. The API consists of
methods for configuration of interrupts, and the methods that are necessary for encryption
and decryption.
Under testing of the AES peripheral API two problems was encountered. The first problem
was that the tests would not compiled since a reverse instruction was used, that was
not available in the x86 instructions set, which is used when running tests on the host
computer. The second problem is that data is read and written to registers that are not
memory mapped. When data is written here, the hardware does the shifting of the data
internally, in this case barrel shifting. To be able to test that the correct data is read or
written, hardware is necessary, or it must be possible to compare values in some way.
Instructions
The rev instruction that results in compilation errors is an instruction that is used to
reverse the byte order of a word in the ARM instructions set. Since these tests are run
on the x86 processor and compiled with GCC this instruction will not exist. Since the
behavior of the instruction is important for the behavior of the unit, it must be replaced
with code that does the same operation. Listing 3.8 shows how this is fixed, where the
method is chosen depending on if TEST is set in the preprocessor or not. It should also
be possible to solve this problem using a macro instead of having to wrap the CMSIS
method for test instrumentation. Adding code for test purpose can be a pitfall, because
errors can occur in this code too. Using a mock module to manually inject the return
value is also possible. When mock module is created for one function in CMSIS, the
function will also need to be moved out of the header-files, which may result in code that
is difficult to understand and maintain. A possible solution is to create a library of stub
functions for testing, which can be used as input to CMock.
1

11

#ifdef TEST
__STATIC_INLINE uint32_t __REV(uint32_t value)
{
return ((value>>24)&0xff) |
((value<<8)&0xff0000) |
((value>>8)&0xff00) |
((value<<24)&0xff000000);
}
#else
__attribute__( ( always_inline ) ) __STATIC_INLINE uint32_t __REV(uint32_t
value)
{
uint32_t result;

13

15

17

__ASM volatile ("rev %0, %1" : "=r" (result) : "r" (value) );

return(result);
}
#endif

Listing 3.8: Reverse Instruction for Test

30

Data registers
The methods in AES API write directly to a structure member that is not memory
mapped. The data-register is 128-bit, but only 32-bit is accessible in the structure, and
the hardware is responsible for the barrel shifting. By the current structure of the code it
is impossible to test that the correct values is passed to the register under test, only the
last 32-bit can be asserted. A possible way to solve this is to use set/get or read/write
methods to access the registers. With these methods it is possible to generate mock
modules that can be used to verify that all assignments to the data register is correct. By
using inline methods, the only difference should be that there are a few extra methods
and that the performance only is different during compilation.
Listing 3.9 shows an example of how methods can be used to access data registers in a
structure. The structure AES_TypeDef includes a member DATA, which is a register
on an embedded device. When data is written to this register, the hardware will do
the processing of the data. Reading data with read_write(uint32_t) during the
unit tests will only be able to assert the last value that was set. By adding read and
write methods to the DATA member with the methods AES_DataRead(void) and
DataWrite(uint32_t), it is possible to create mock modules that can be used to
verify expected calls and inject the return values that are necessary for further processing
in a method.
1

typedef struct
{
__IO uint32_t DATA;
} AES_TypeDef;

11

13

__STATIC_INLINE uint32_t AES_DataRead(void)

{
return AES->DATA;
}
__STATIC_INLINE void AES_DataWrite(uint32_t data)
{
AES->DATA = __REV(data);
}

15

17

19

void read_write(uint32_t *in)

{
int i;
uint32_t *out;
for (i = 3; i >= 0; i--)
{
AES_DataWrite(_iv[i]);
}

21

23

25

for (i = 3; i >= 0; i--)

{
out[i] = __REV(AES_DataRead());
}

27

29

Listing 3.9: Access to Data Registers

31

3.5

Use of Mock Modules

Mock modules/objects are essential when testing is performed without access to hardware.
To be able to have automated testing of all the binaries that are created for different IDEs,
examples and development kits, it is necessary to use these mock modules.
Mock modules will be used to verify that calls are done to the hardware, and it is also
used to inject values that are to be returned on specified function calls, where the function
returns data. By using these mock modules, the hardware will not be needed, since no
values are written to memory registers on the hardware, but are stored into the mock
module for verification.
It is possible to manually write mock modules, but since emlib consist of many hundred
functions, this will take too long time, and it will be too difficult to manage modification
of the library over time. CMock will be used to do the generation automatic, which is
done in under a minute. Another advantage is that this can be run each time, so the
mock modules are updated with new functionality that may have been added.

3.5.1

Energy Micro peripheral library

Energy Micro Peripheral Library (emlib) is the HAL (Hardware Abstraction Layer) in
the software library. The HALs task is to be a programming interface and support all
the peripheral devices. Since emlib do all the calls to the hardware, mock modules can be
generated for emlib and be used instead of using the emlib. With the mock modules, the
example project can be tested automatically and without use of hardware. In this way
developers will not be able to test the functionality in the modules that were mocked, but
the functionality that uses the mock modules can be tested without dependencies.
After an attempt to use the mock modules it was discovered that mock modules for some
parts of the Cortex abstraction layer and some of the peripheral support library also was
necessary. Later it was also discovered that we needed to create mock modules for board
support package and the driver to remove the dependencies in the examples. The biggest
problem with the dependencies in the examples was when using the segment LCD driver.
Only for initialization, the segment LCD has over twenty function calls to LCD API,
and writing to the text segments will results in even more calls to the LCD peripheral
API. This is unnecessary and will result in too much time used on write tests for these
functions.

3.5.2

Generation of mock modules

Since many of the functions that exist in emlib are inline_static, it is impossible to
generate mock modules from emlib without modification, since these functions must be
included in the mock modules. To solve this, a Ruby script was written. The main task
for this script is to wrap all inline static functions into preprocessor macros that use void
functions instead. The header-file is then preprocessed with the GCC preprocessor, and
after this the header-file can be used to generate mock objects.
Include directives for header-files is important when the file is used in compilation, but
during preprocessing to generate mock objects, these includes will only generate conflicts
32

and functions that are not needed for generation of mock modules. Therefore it is also
necessary to wrap the include directives into #if directives so that they not are included
during preprocessing of mock objects. But some of the include files are necessary since
they include definitions that are used to define methods. These includes are accepted by
the script.
The script is run by specifying a YAML configuration as input. This configuration file
specifies paths to all necessary include files, paths where the files shall be saved after
generation and paths to CMock. The source can be found in appendix C. The script only
works on the emlib, and some manual work is still necessary to be able to generate mock
modules. The script was created to automate some of the process during generation of
mock modules.
The output from running the script will be the header and source file for the mock modules,
which will be used instead of the emlib. The script will also generate new header files
that must be used instead of the header-files in the emlib. This is important because
inline_static functions will create conflicts during compilation, so these header files
must be used.
Generation of mock modules for other parts than the peripheral library (emlib) was done
manually using CMock. Scripting of all parts will be necessary during automatic building
and testing. The reason for this is that changes can have been done to the library, and
this will require new mock modules. We should therefore generate new mock modules
each time all binaries and tests are compiled.

3.5.3

Testing with mock modules example

Listing 3.10 and Listing 3.11 shows some example code and tests using mock module. The
behavior of click_left(void) is that a variable is incremented, an LED is toggle and
a text containing the number of clicks is written to the segment display. The test function
for click_left(void) will first set an expectations for that a LED is toggled, the first
parameter tells which LED to toggle, and then the return value for method, since the
method returns an integer for a status code. The next line sets the expectation for that
SegmentLCD_Write(char*) is called with the expected parameter "1 click". The
next line is the function call, and afterwards the mock modules are verified, to check
that all the expected functions are called with the correct parameters. The test runner
which is generated will have the methods for running each of these methods and do the
initialization and verifying of the mock modules. The last assertion is a unit test that
verifies that the left click variable has been incremented.
2

static uint32_t leftClick = 0;

char output[8];
void click_left(void) {
leftClick++;
BSP_LedToggle(1);
snprintf(output, 8, "%lu CLICK", leftClick);
SegmentLCD_Write(output);
}

Listing 3.10: Code Under Test with Mock Modules

33

void setUp(void)
{
leftClick = 0;
}

void tearDown(void)
{
}

11

void test_click_left(void) {
BSP_LedToggle_ExpectAndReturn(1, 0);
SegmentLCD_Write_Expect("1 CLICK");

13

click_left();
15

TEST_ASSERT_EQUAL_UINT32(1, leftClick);
17

Listing 3.11: Test Using Mock Modules

3.6

Testing of Examples

This section will look at possible ways to perform testing of the examples that are created
for the starter and development kits. Testing will be performed in two different ways:
Mock modules By using this method, tests will be run on the host computer, and the
tests will not have access to hardware. Instead of using emlib, the test will use mock
modules for emlib. These mock modules are generated by using CMock.
Development board The tests will be run on the development board, which means
that the software have access to the hardware, and the unit tests will be able to
compare values in the memory registers. While running tests on the development
board, the test results will be written back to the host computer using SWO (Serial
Wire Output).
Both of the methods will have advantages and disadvantages, which will be covered later
in this section. The blink example in Listing 3.12 is taken from Energy Micros software
library.

3.6.1

STK3700

All the examples that are tested are created for the EFM32GG STK3700 starter kit.
Currently there exist 24 different examples for this starter kit, and most of them have
support for all the IDEs that was mentioned in the background chapter. Compilation of
these examples it will result in nearly a hundred different binaries that need to be tested.
Figure 3.3 shows an overview of the STK3700 starter kit, with some of the features that
are available on the board, the figure is taken from the STK3700 user manual.
34

Figure 3.3: STK3700 Hardware Layout [12]

The starter kit is based on the Giant Gecko microcontroller, and is suitable to test many
of the features that this microcontroller supports. The example tests will use some of the
examples that are shipped with the microcontroller.

3.6.2

Testing with mock modules

When mock modules are generated instead of using the CMSIS and emlib, it will no longer
be possible to detect errors in these layers during testing. But the mock modules will be
able to verify that the correct function calls are run in the examples. The advantage of
testing without the use of hardware is to make testing automated, which will save time.
Testing of emlib and CMSIS can be done by other tests, so we can presume that these
layers are without errors, and focus on the other layers that are used by the examples.
By testing with mock most of the dependency is removed during software unit testing.
Software integration testing is needed when the software is tested layer by layer. Below
three of the examples are tested using mock modules.

Testing of blink example

The first test of the example projects uses the blink example, which is a simple template
to start developing for the starter kit. The only additional functionality in addition to
initialization is the LEDs that are blinking.
Listing 3.12 shows the source code of the example, the first issue that needs to be fixed
before testing is to rename the main function, since the entry point is needed for the tests.
This is solved by defining the main function as app_main(void) if TEST is defined.
The next changes that was done was to move the infinity loop to a new method, that was
named blink_loop and call this function from the main method. To prevent the loop
from never stopping, the while loop was change to a do-while loop, to ensure that it run
35

at least run once. Above blink_loop(void) some test instrumentation is added. The
first thing is to define INFINITY to zero during testing and one else, the last define is a
macro that set the delay to zero under test. Setting delay to zero is necessary during test,
since there is no hardware that can create interrupts and increment the variable during a
system tick. The source code can now be tested with unit tests and mock modules.
1

volatile uint32_t msTicks;

void SysTick_Handler(void)
{
msTicks++;
}

void Delay(uint32_t dlyTicks)

{
uint32_t curTicks;

11

curTicks = msTicks;
while ((msTicks - curTicks) < dlyTicks) ;

13

}
15

17

19

21

23

25

27

29

31

33

35

37

39

#ifdef TEST
#define Delay(x) Delay(0)
#define INFINITY 0
#else
#define INFINTY 1
#endif
void blink_loop(void)
{
do
{
BSP_LedToggle(0);
BSP_LedToggle(1);
Delay(1000);
} while (INFINITY);
}
#ifdef TEST
void app_main(void)
#else
int main(void)
#endif
{
CHIP_Init();

41

BSP_TraceProfilerSetup();

43

if (SysTick_Config(CMU_ClockFreqGet(cmuClock_CORE) / 1000)) while (1) ;

45

BSP_LedsInit();
BSP_LedSet(0);

47

blink_loop();
49

Listing 3.12: Blink Source Code

36

Listing 3.13 shows the tests for the blink example. The file shows three different tests
where the first function is testing of the main method, where also the loop is run once.
Most of the code lines in the test are expectations for which functions that shall be called
by the main method. When all expectations and return values are set, the main method
is run. All verifying of the mock modules is done by the test runner, which is not visible
here. The second test verifies the blink loop and the expectations for these methods are
that both LEDs are toggled. The last test asserted that msTicks is incremented by one,
when an interrupt occur.
1

void test_app_main(void)
{
CHIP_Init_Expect();
BSP_TraceProfilerSetup_ExpectAndReturn(0);
SysTick_Config_ExpectAndReturn(0, 0);
CMU_ClockFreqGet_ExpectAndReturn(0x040020,10);
BSP_LedsInit_ExpectAndReturn(0);
BSP_LedSet_ExpectAndReturn(0,0);

BSP_LedToggle_ExpectAndReturn(0,0);
BSP_LedToggle_ExpectAndReturn(1,0);

11

app_main();

13

}
15

17

19

21

23

25

27

void test_loop(void)
{
BSP_LedToggle_ExpectAndReturn(0,0);
BSP_LedToggle_ExpectAndReturn(1,0);
blink_loop();
}
void test_systick(void)
{
uint32_t temp = msTicks;
SysTick_Handler();
TEST_ASSERT_EQUAL_UINT32(temp + 1, msTicks);
}

Listing 3.13: Blink Test Code

The changes and tests above make it possible to do unit testing of the example, where
almost all code was tested. The only method that not was tested was the delay function,
which is impossible to test without hardware or use of threads.
Testing of clock example
The clock example is an application that uses the segment LCD to display the time,
buttons to set the time, and a real time counter that updates the clock each minute. For
each iteration the clock enters an energy efficient mode, which pause the loop, and the
system wakes up by either user input from the button or the real time clock. Interrupts
and entering different energy mode do not happen during test, since hardware is not
available, so calls are verified by mock modules.
Some changes need to be done to the source to make it possible to run the tests. The
main method must be renamed and the while loop must be changed to a do-while that
37

do not run infinitely during test. This example already has a loop method, so this is not
needed here. To be able to run methods and do assertions from the test file, a header-file
is needed, this file is shown in Listing 3.14. In the header-file, variables are defined as
extern and methods are declared so the unit tests can access these.
2

#ifndef CLOCK_H
#define CLOCK_H

#include <stdbool.h>

extern uint32_t hours;

extern uint32_t minutes;
extern bool oldBoost;

10

12

void
void
void
void

GPIO_ODD_IRQHandler(void);
GPIO_EVEN_IRQHandler(void);
RTC_IRQHandler(void);
clockLoop(void);

14

16

18

#ifdef TEST
void app_main(void);
#else
int main(void);
#endif

20

#endif /* CLOCK_H */

Listing 3.14: Header-file for Example Test

The mock modules are used to verify that correct calls are done, and to inject necessary
return values. Mock modules are used for the peripheral APIs and the VDDcheck driver.
The test covers all of the code in this example, no dependencies in other drivers and
peripheral APIs are tested since mock modules are used instead. The clock example has
also some logic that is tested by unit tests. This include that time is set correct. Being
able to test all units and methods in the example should verify that the example works
as intended in the tests. That it works together with other software components and
hardware must be tested with software integration tests and integration testing.
Testing of NAND flash example
The last example that was tested using mock modules was the NAND flash example, which
is used to perform commands for reading and writing to the NAND flash memory on the
microcontroller. The commands are given to the development board and microcontroller
by writing commands over a serial interface. The examples demonstrate features in the
NAND flash driver. The example was difficult to test because of many definitions that
need to be changed for testing.
To be able to run this example on the host computer it was necessary to generate mock
modules for the NAND flash driver and the driver for the serial interface. It uses some
of the mock modules of API in emlib. In addition we could have created mock modules
for features that are used in the standard libraries of C. If verification of parameters
to printf(...) and putchar(char) is not important, stub functions can be use
instead.
38

Like in the other examples there was need to change some of the structure, to be able to
do the testing. Function declarations were moved to a header-file, test instrumentation
to rename main methods was added and the loop was moved out of the main method.
With these changes it became possible to compile and start writing tests for the system.
In this example we used calls to methods in the standard libraries of C. printf(...),
putchar(char) and getchar(void) are used to write and read text over the serial
interface. The functions used for text output can easily be ignored using a macro to remove
output to the test output. Example of this is shown in Listing 3.15. But getchar(void)
is needed to read input, which is used by the loop to execute the command. Since the
commands need to vary during the test, this cannot be fixed by a macro. There are
two solutions to this problem, either create a mock module, which can be used to inject
values to getchar(void) or ignore the getCommand(void) which is used to read
commands and put them in a command queue, make the command queue available in the
tests and manipulate them before test. This way it is possible to test all the functionality
expect for getCommand(void), which can be tested if mock modules are created. The
functions printf(...) and putchar(char) can also be mocked to verify that the
correct values are printed.
1

#define printf(...) printf("")

#define putchar(x) putchar(\0)

Listing 3.15: Redefinition of Functions with Macros

Segmentation fault was found during testing of the example. The problem is the use of
the DWT (Data Watch point and Trace) unit. The first issue is that DWT is turned on
in the core debug register in the CPU. This register is defined as a pointer to the register.
To be able to test it, the register need to be defined as CoreDebug_Type pointer and
memory allocated for the definition. The definition is changed in CMSIS and memory is
allocated in the setup method in the tests. In addition we need to change two variables
that are defined as pointers to a memory address; DWT_CTRL and DWT_CYCCNT. These
were changed to 32-bits unsigned integers during test and the variables was added as
extern in the header-file to be accessible for assertions by the tests.
It was possible to cover most of the code with tests using mock modules and unit tests.
Mock modules were used for expectations and return value injection for all method calls
to the drivers and the peripheral API. Unit test was used to assert that values were set
correctly by the code. Most of the variables were not visible outside the source file, so
extern was added for these values in the header-file during test, for static variables,
the static keyword was removed to make it accessible for other source files. All changes
to variables was put into preprocessor directives that checked if TEST was defined.

3.6.3

Testing on development board

By running the unit tests on the development board, it will be possible to find errors in
all levels of the software library, and it might be possible to detect errors in the hardware
too. IAR where used as an IDE during testing on the development board. By testing the
example, all the layers are used and software integration testing is done, in some degree
integration testing too, since hardware is used. To be able to see the results of testing,
SWO (Serial Wire Output) was used to write the test results back to the IDE.
39

To be able to do assertions on the development board, Unity was used. To be able to run
the tests, some modifications were needed in the example code. The main method was
renamed, and the loop was moved out of the main method and changed to a do-while
loop, so the loop would not run infinitely. Unity compiled with IAR, and run as expected
on the hardware. Assertions were done on registers that were readable, to verify that the
example had modified the registers as required.
Mock modules were tested on the development board. The mock module for the segment
LCD was tested to verify that it worked in the same way as when testing on a host
computer. The segment LCD was tested and worked as expected. Using mock objects on
the development board may be necessary to test some functionality and to make it possible
to automate some testing, without having to depend on user interactions or additional
hardware.
Automation and scripting of testing several examples on hardware might be possible, and
it is possible to use other test environments than an IDE and SWO to get test results
back to the test host. This project did not look into how to automate this. Testing on
the development board was done to see if it was possible to run automated unit tests on
the development board, and to verify that the test frameworks Unity and CMock worked
as expected on the board.
The advantage of running the test on the development board is that the code runs on
the intended environment with hardware. When hardware is in place, most functionality
can be tested, and there is no need to remove functionality like the delay method, since
the interrupts is not working on the host computer during unit testing. But it could
be necessary to change the delay during unit testing, since a second delay in a unit test
is bad practice, because the tester will have to wait longer for the tests to finish. The
disadvantage is that there still is not any user interaction during automated testing, and
that not all register can be read and asserted.
One important issue when using assertions on hardware is that not all registers can be
read, since they are only writeable. This means that other actions must be taken do
test the feature, since a write operation to a register will result in some behavior by the
hardware. It will work to verify that the effects are as expected, which can be asserted
either manually or automatic, which means that it can be asserted by a test harness that
is reading other registers or by the developer. Cases where an operation is done and the
effects of the operation done by the hardware is verified, are integration testing, where
both the software and hardware is tested.
Some registers are also mapped to other register using bit operations. When a LED is
toggled, a bit is written to the register DOUTTGL, and the register output can be read
from DOUT. Example is shown in Listing 3.16.

GPIO->P[4].DOUTTGL = 1 << 3;
2

TEST_ASSERT_EQUAL_UINT32(8, GPIO->P[4].DOUT);

Listing 3.16: Assertion on Development Board

40

3.7

Testing of Drivers

The board support package is a driver to support the functionality of the board controller
and the devices on the development board. The drivers are collections of methods that
are often used by the examples to use functionality in emlib. Two drivers was tested, the
first was VDD voltage check, which is used to check the voltage to the microcontroller.
The other was the LEDs driver for the starter kit.

3.7.1

VDD voltage check

This driver is used to check if the voltage in the microcontroller is below a given value.
This is used for example in the clock example to check if it is necessary to boost the
voltage to the segment LCD.
It is straight forward to write tests for this driver. Unit tests can be used to assert the
return values from the function, and expectations is set for the function calls to the voltage
comparator mock module. This is done for each of the methods in the driver.
The function VCMP_Init(VCMP_Init_TypeDef) takes a structure as parameter. When
a mock module is created for this, a memory comparison is done of the data in the actual
and the expected structure. Due to data alignment/padding, a boolean member that
takes one byte in memory, will use 4 bytes in the structure. When a boolean is set, only
one byte will be written, and the other three byte will be garbage. This results in a
failed test. There is three ways to solve this. The first is to tell the compiler to pack the
structures. The second option is to write zero values to the structure after initialization.
The last option is to write a helper method and include this in the CMock configuration
before generation of mock modules. It is also possible to ignore the method, if it is not
important to test it.
When we have a mock module of the voltage comparator, it is possible to test that the
loop waits for the measurement of voltage to finish, by setting exception and inject return
values that makes the loop continue. If the loop is not waiting, the result will be that the
function was called less times and the test fails.

3.7.2

Board LEDs

The board support package for the LEDs is used for the LEDs that are available on the
development. The driver includes functionality for initialization, get LED status and to
turn on and off LEDs.
Tests for these drivers were written with unit tests and mock objects. A mock module for
GPIO and CMU was used, where expectations for all of the function call from the drivers
was set for the tests. Unit tests were used to assert the return from the functions. Tests
for all conditions for each of the methods were written. No problems were found when
testing this driver.
41

3.8

Automated Build System

This section will look at automated building and testing of the system, both during
development and testing, and as periodic testing of a complete repository. To be able
to run automatic testing, we need to prepare the existing code for this. It will also be
necessary to have a suitable test framework and the code need to be structured in a decent
way.

3.8.1

Test preparation

To be able to automate the test activities, it will be necessary to do some modification to

the code. Some of these might be permanent, while some changes are necessary to be able
to run the tests on a host computer. A permanent change to improve the testability of
the code is a manual task that can be done gradually when code is changed and tests are
added. For functionality that is added we should focus on writing code that is testable.
Many of the changes that are needed to make the code testable, can be automated by
scripting. The script can be chosen to run when necessary or be run each time tests are
run. To be able to generate mock modules automatic with CMock it is necessary to do
modifications of the header files for the peripheral APIs. This was covered in section 3.5.
All the memory register to the peripheral device are set to specific memory addresses that
are mapped to the device on the microcontroller. These addresses cannot be used while
testing on the host computer. Since there are files that define the memory registers for each
microcontroller, there will be several hundred lines to changes, which make it necessary
to automate this task. Since the definitions of memory registers may be changed, it will
be necessary to run this script each time the tests are run.
To be able to test the examples, several changes are needed. The first is to rename the
main method under test. This change can easily be done by a script for all of the examples.
Moving the loop of the function out of the main method is done in some of the examples,
but not all. This is something that should be done to make it easier do to the testing,
and make it possible to unit test the functionality that is done in the loop. This is a code
change that must be done manually. Changing loops from a while to a do-while can be
done by scripts before testing, it is an easy change that can be done by a simple parser,
and the changes is only necessary for the test. Some example access memory registers
directly on a given memory address. This will result in segmentation faults, and must
be changed to either variables or use of a structure, when the tests are written for the
example. Removing for example the delay function in the blink example should be done
manual when creating tests, since this a feature that not occur in all examples, and other
example can have equivalent features that need to be fixed during testing.

3.8.2

Ceedling

Ceedling is an automated building tool can be used with Unity and CMock [2]. Ceedling
is a ruby gem, which uses Rake (Ruby Make) to perform the building [2]. Ceedling is
just an extension of Rake to focus on testing and compilation of C projects. Ceedling is a
good tool for supporting test-driven development since it has many options for automatic
compilation and testing. By using Ceedling it is possible to write and run unit tests
42

together with Unity and CMock, without having to create Makefiles, or create test runners
and generate the mock modules with CMock.
The main functionality of Ceedling is the ability to compile and run tests, without having
to specify which files that needs to be compiled and linked. A Ceedling project consists
primarily of directories for source code and tests, and a configuration file. In the configuration file it is possible to specify options for the compiler and preprocessor that shall be
used. In addition, the configuration holds the Unity and CMock configuration.
When Rake is used to test, it will look up the tests that are needed and create test runners
for these tests. After this, compilation of all source file will be done. If a mock module
is needed, this will be generated with CMock and compiled. The last steps are to link all
the object files, run the tests and write the test results to the developer.
For testing, Ceedling can run all tests, run the tests in a given directory, test a specific file
or only run tests for files that have changed since last time Ceedling was run. By having
the ability to specify which test to run, it is good for use in test-driven development,
where tests are run for each change. Testing all tests each time could result in a hour
break between each change when using TDD.

3.8.3

Test structure

A good file structure is important to separate tests and source files. The source should be
separated from test files, since it do not depend on functionality in the tests. The tests
will consist of the test framework, the actual unit tests, mock modules and a test runner.
The unit tests, mock objects and the framework should be stored separately.
A Ceedling project uses a simple structure, where all source- and header-files are stored in
a source directory (src/). In addition, there is a test directory where all tests are stored.
Ceedling will look through all subdirectories in both source and test folder and look for
header, source and test files. This means that the structure in these directories does not
matter. In the test directory the tests should be divided into unit tests and integration
tests. The unit tests should also be divided according to the part of the software library
that it tests.
Energy Micro already has their own system for compiling and testing their software.
Changing this to use Ceedling would result in some work and there is also a risk that it
could results in errors. A simple test, moving the source directory for the Energy Micros
software into a Ceedling project, was done. One example test and one peripheral API test
was added to the project. The next step was to configure Ceedling, where configuration
for CMock, Unity, the preprocessor and the compilator were done. There were only a two
errors that were needed to fix before the tests was run:
1. Ceedling will generate the mock modules during compilation, the header-files that
are used need to be prepared so they do not have any inline methods.
2. Some files had equal name, so the wrong file was chosen. The options for configuration of files with the same name were not looked into.
The biggest complexity in Energy Micros software is that it has a peripheral API that
shall support all of the microcontrollers. Many of the examples are also created so that
43

they can be run on different development boards. The CMSIS peripheral support library
takes care of definition of memory registers, availability of hardware and other definitions
that are need for the specific microcontroller. The configurations that make the drivers
work for the examples are defined in the board support package. The main problem with
testing is how to be able to test examples for several development boards, without having
to write duplicate tests for each of the development boards. Using the same tests to test
the functionality in emlib for several microcontrollers is also a problem.
When compiling the peripheral support library, a definition of microcontroller is done in
the microcontroller. To test this for all microcontrollers, the tests can be compiled and
run several times, one run for each microcontroller. This can easily be scripted and done
automatic, but there will be a problem with this solution: The microcontrollers do not
have the same features and all tests are not necessary, and some of the asserted values
will be different for each microcontroller. These problems have not been looked at in this
thesis. But using the preprocessor to define which tests to run is possible. The same
goes for the values that are asserted. Bit-field and masks are defined in the peripheral
support library, and could also be used as values that are expected by the tests. Using
definitions in the tests will make it harder for the developer to understand the values that
are expected.
The definition of the microcontroller is done in the preprocessor for the examples projects.
The same problems as with the peripheral APIs will also exist for the examples with
software integration testing, but this will not be a problem when mock modules are used.

3.8.4

Automated test runner

With test-driven development, a developer will have to run the tests each time a change
is done. By running the tests that often, it is important that tests can be run fast, at a
maximum of one minute for all the tests that are run. It is therefore necessary to minimize
the test that need to be run each time. The developer should at least run all the test
for the module that is under development. It could also be necessary to test some the
functionality that depends on the code.
By only testing a limited part of the software library under development, there is a need
to do the rest of the testing elsewhere. A continuous integration server can be used for
this, which can run tests, when code is committed to the software repository. If there are
many commits to this repository, there might only be possible to run some integration
tests here. To be able to test all, nightly jobs can be run, where all software is build and
tested.
When testing the software library, a lot of test instrumentation code is added to the
source code. This is code that only is used for testing purposes, and the code that shall
not be shipped with the code to the customers of Energy Micro. Before release of code it
is necessary to have scripts that go through all the source code to remove the test instrumentation. To ensure that the script is able to remove all of the instrumentation code, it
is necessary to specify what kind of defines that is allowed for the test instrumentation.
44

3.9

Summary

This section has discussed testing of embedded systems, using Unity as a test harness and
CMock as a tool for generation of mock modules from function prototypes in header-files.
CMock generates mock modules that use Unity for assertions. Unity has the necessary
functionality for doing assertions, and there exists plugins that can make Unity support
fixtures. Since CMock does not accept header-files that consist of functions defined as
static inline or by macros, it is necessary to modify these header-files before generation of
the mock modules. Ceedling is an automated building tool, which was created to make
Unity and CMock work together in a better way.
For software unit testing, the different parts of the software library have been viewed as
different layers, as shown in Figure 2.2. When a part of the software package has been
tested, all dependencies to other modules have been used as mock modules. By doing so,
it is possible to test each of the unit in isolation, without concern about the functionality
in the dependencies. To make it possible to create the mock modules automatic, a script
was created to automatically remove inline functions in the header-file and change them
to normal function definitions.
To check whether Unity and CMock could be used to test the functionality in Energy
Micros software, different peripheral APIs, drivers and examples was tested. All the
parts were tested with use of mock modules and the peripherals were also tested by doing
assertions with unit tests on the memory registers that was defined as structures. Testing
an example on the development board was also done.
Making changes to the software to be able to test software can be time consuming. It may
be necessary to create scripts that can remove test instrumentation before releasing code,
and to change the definition of memory registers from fixed memory mapped addresses to
pointer definition of the type structure. Scripts to modify header-files that will be used
as input to CMock must also be used, due to the amount of static inline definition in
the header-files. Automation of compilation and testing is necessary to be able to choose
which tests to be run and to support agile development like test-driven development.
Most of the problems that occur during testing can be solved by adding test instrumentation that makes the preprocessor modifies the code when the code is under test. Macros
can also be used to redefine behavior under test. In some cases it is also necessary to do
some changes to the existing code. When writing to memory registers, it can be necessary
to add read and write methods, to be able to create mock modules that can capture the
data. In the example it is necessary to do some structural changes, so it is possible to run
tests without going in an infinity loop.

45

46

CHAPTER

4
EXPERIMENT

This chapter contains the experiment that was performed to looks at use of unit testing
and mock objects/modules to perform regression testing in an existing embedded system
where functionality is added or modified. The unit testing will be done on the host computer, where mock modules are used as a hardware abstraction layer. The structure of the
experiment follows the structure suggested in Experimentation in Software Engineering:
An Introduction [23].

4.1

Definition

This section forms the idea of using unit testing to perform regression testing when an
existing system is refactored, modified or when functionality is added. In the experiment,
the unit tests are created in advanced, so that the participant can verify that the implemented functionality behaves as expected. All the unit tests that are used, is integrated
in mock modules.

4.1.1

Goal definition

The experiment is motivated by the Energy Micros case, where they have over hundred
example project, where almost every project supports four different IDEs and the example
may be used in different development boards. That an example runs on different development boards does not mean that they are equal, since the board layouts are different.
This will result in over thousand binaries, which is almost impossible to test manual by
deploying to development board and then run regression or system tests. Important to
look at efficiency by using unit tests to do regression testing and see if the number of
errors is reduced when using unit testing to perform regression testing after modification
or refactoring of code without tests.
When unit test is used without access to hardware, there are several approaches that can
be used; some of them are listed below:
Use mock objects/modules as a hardware abstraction layer.
47

 Setup all memory registers as variables, and perform assertions to these.

Run the tests in a hardware simulator.
This experiment use mock modules as a hardware abstraction layer, where expectations
is set, assertions are done inside the mock module, and return values is injected. When
using these methods, the tester does not have full control over functionality that is done
by the hardware, but it can verify correct calls to the hardware abstraction layer, and
assert that the calls are done with correct arguments.
The objects of the study is existing source code of an embedded system. The participants will write code to complete the requirements that are specified for the task. Since
refactoring is difficult to measure, the participants will add functionality to the system
or modify the code, to change the behavior of the system. Some of the tasks will also be
tasks that are not working, and the participants should make them work by adding the
missing parts. For the experiment there will be five tasks, described briefly below:
1. This task is a simple application that counts down from a given value to zero. It
uses the segment LCD to display the count value, and uses the LEDs to indicate
that the counter has reached zero. The participants will need to implement the
initialization of the segment LCD and the LEDs, and then implement the behavior
when the counter reaches zero.
2. The application uses the push buttons on the development board to turn on and off
the LEDs. In addition the total number of clicks is written to the segment LCD.
The participants will implement the interrupt handlers for the buttons, so that the
LEDs are set or cleared and the total numbers of clicks are written to the display,
when a push button is clicked.
3. This task uses the push buttons and the segmented LCD. The segment LCD will
display a binary string, and the decimal value of this string. The buttons will be
used to increment and invert the value of the binary string. The participant will
implement the behavior of the interrupt handler and write the binary string to the
segment LCD.
4. This is a simple application that displays a string on the segment LCD. The task for
the participants is to modify the functionality so that the user can read the whole
string on the display. To be able to do so, the string will need to rotate from the
start to the end on the screen.
5. This application uses the capacitive touch slider on the development board. It
reads the value from the capacitive sense interface and displays the value of the
touch slider on the segment LCD. In this task the experiment participants will do
the initialization and setup of the capacitive sense system. In addition they will
modify the functionality so that the segment LCD is turned off when no value is
read from the touch slider.
Object of study The object of study is the source code of existing embedded systems,
and how this system can be tested in an efficient way during refactoring, modification
of the source code or when functionality is added, to reduce the amount of errors
before the software is deployed to hardware.
48

Purpose The purpose of the experiment is to see if use of unit-testing during development
of embedded system can reduce the number of errors and see if the developers will be
more efficient when using unit testing instead of system and regression testing after
deployment of code to the embedded system. If the reduction of errors is sufficient
it may not be necessary to test all binaries on hardware.
Quality focus The main effects that are studied in the experiment are the reduction
of errors under development on embedded system, by using unit tests and mock
modules/objects. If numbers of errors are reduced it may also reduce the time used
on testing.
Perspective The perspective is the developers view, since they are doing the testing and
therefore want an efficient way of doing testing, which will reduce the number of
errors in the software. Outside the experiment both project managers and customers
may be interested in the reduction of errors before the source code is delivered, since
they want software without errors.
Context The environment of the experiment is a computer lab that consists of a computer and an Energy Micro EFM32GG STK3700 development board. The participants in the experiment will modify an existing source code for five simple examples,
which will be provided. The participants will be students with basic knowledge of
programming in C.

4.1.2

Summary of definition

Analyze using mock modules as a method for testing an embedded system for the purpose
of regression testing with unit tests with respect to reduce number of errors and efficiency
from the point of view of the developer in the context of a computer lab.

4.2

Planning

This section provides the planning part of the experiment, which uses the definition from
the previous section. The planning consists of context selection, hypothesis formulation,
variable selection, the selection of subjects, experiment design, instrumentation and validity evaluation.

4.2.1

Context selection

The experiment is an off-line context since it uses tasks that are created in advance for
this experiment. Experiment is not done in a professional organization, but in a lab.
The participants in the experiment will be students that know the basics of programming
in C.
The code used in the experiment is not created for any real project, so the experiment
will be a toy problem.
The problem is specific because the focus is to use unit testing with use of mock modules
for regression testing of embedded software.
49

4.2.2

Hypothesis formulation

The subjects in the experiment will be divided in to groups, one that uses unit test, the
number of errors for this group is expressed as x . The number of errors for the other
group, is expressed as y , this group not will have access to unit tests. In this experiment
measures the number of errors found during testing on development board. The number
of errors found during testing can be compared by a ratio scale. The number of errors for
each group is the number of errors found with system tests after the code is deployed to
the development board.
The null hypothesis is that there is no difference in number of errors found during testing
between the group that uses unit test and the group that not uses unit tests.
H0 : x = y

(4.1)

The alternative hypothesis H1 defines that the testing on the development board resulted
in fewer errors for the group that used unit tests.
H1 : x < y

4.2.3

(4.2)

Variables selection

Independent variables The independent variables are the students that will perform
the experiment, and the difficulty for each of the tasks the participants will do in
the experiment. The independent variable that will be studied is the development
process. In this case this will be use of unit testing or without use of unit-tests.
Dependent variables The dependent variables in the experiment will be number of
errors that were discovered when testing on the embedded system. The number of
errors are found when system tests are executed after the code is uploaded to the
embedded system.

4.2.4

Selection of subjects

The subjects that are used in the experiment will be students. The students have some
experience with the C programming language, and most of the students should have
some experience with programming for embedded systems. The sampling used in the
experiment is convenience sampling.
The tasks in the experiment do not require deep understanding of the C programming
language, but they should know the basic syntax of the language and the use of pointers.
The experience with programming on embedded system is needed to be able to finish the
experiment in a reasonable time, since the participants will need to use documentation for
APIs, reference guides, and be able to download the compiled binary to the development
board.
50

4.2.5

Experiment design

The factor for this experiment will be the development process. The first will be a group
that uses unit testing during development on the tasks. The second group will have to
implement the functionality without being able to use unit tests.
The experiment will thus have one factor with two treatments, and use a completely
randomized design. The method of analysis will be the t-test.
Randomization The subjects for the experiment are not chosen randomly, since they
need experience with the C programming language and experience with programming on embedded systems. The subjects will be assigned to one of the treatments
randomly.
Blocking Due to the different amount of experience with programming on embedded systems, it will be necessary to put the subjects into blocks and then divided them into
different treatments. This is necessary to eliminate the effect of different experience
with programming on embedded systems.
Balancing Each of the treatment in the experiment will have an equal number of subjects.
Standard design types The design of the experiment is one factor with two treatments,
where the factor is the development process with use of unit testing, and the treatment is using unit tests or not. The number of error is the important dependent
variable, which can be measured on an interval scale. In this case a t-test will be
used for analysis of the data.

4.2.6

Instrumentation

The participants will be given five tasks that are the objects of the study. They will add
functionality and modify the source code for these tasks. The group that has access to
tests can run these tests and modify the source code until tests pass or they give up.
When the groups have downloaded the code to the development board they will not have
access to modify the source code.
The participants will be given guidelines that describe the tasks and the steps in each of the
tasks. The documentation will consist of instructions to upload code to microcontroller,
and for each task there will be requirements, instructions and test cases. In addition we
will hand out necessary documentation of the software library.
When the system is running on the development board, the participants will perform the
tests cases that are given in the guidelines and documentation to the experiment. The
total number of errors, which is the dependent variable, will be the number of test cases
which failed. A failing test case that fails because of an error in one of the other test cases
will not be counted as an error.

4.2.7

Validity evaluation

The threats to the validity follow below:

51

 The participants get bored if the experiment lasts too long. We do not know how
long the experiment will last. We have tried to reduce the amount of work that is
necessary for each task, but each task is quite repeatable since each task will require
deployment of code to embedded system and it needs to execute test cases.
The task is too easy so it will not create any difference between the two groups.
The tasks are created easy so it can be possible to accomplish the experiment in
1-2 hours, and there should be some tricky details that may produce errors for the
developers.
Not enough participants that participate in the experiment will result in weak statistics result and make it difficult to make any conclusion of the results from the
statistically analysis.
One of the groups will use mock modules with unit tests. By looking at the tests code
the participants can use this as a guideline for how to implement the functionality
for the task. The only way to prevent this is to ask the participant to not look at
the test code, but use the tests to find errors in the software.
Due to limited space on the computer lab, not all participants can do the experiment
at the same time. This may be a problem since the last participant can be informed
about tasks in the experiment. It is probably only a few or none that have knowledge
of the Energy Micro microcontrollers, and students will probably not try to find
solution before they meet up on the experiment, this should not be a problem.
If the documentation and guidelines for the experiment are not good enough, it will
result in problems for the participants, which can influence the results, since they
not are able to carry out the experiment. This may be prevented if we test the
experiment before the experiment is carried out by the subjects.

4.3

Operation

The operation of the experiment uses the design and definition to carry out the experiment. This section consists of the preparation, execution and data validation.

4.3.1

Preperation

The participants were aware of that the experiment would look at unit testing on embedded systems. They did not know how the result was measured.
Finding participants that had experience with C and microcontrollers was difficult. The
experiment ended up with 16 participants from XCOM14, where only half of the participants had experience with C/C++. The experiment was made simpler, to minimize the
use of pointer/data structures that differ from Java programming. Each of the participants was paid 500 NOK.
Each of the participants got the guidelines and documentation at the start of the experiment. The documentation included documentation for the drivers and APIs they needed
to use to implement the necessary functionality. The guidelines consisted of an introduction, instructions of how to download the necessary source code, tools and compiler.
52

In addition it included instructions for how to upload code to the hardware. The most
important part of the guidelines is the instructions of how to perform the five tasks. The
tasks consist of requirements, instructions and description of how to implement this. The
experiment guidelines is found in Appendix A.
Each of the participants needs to download the necessary package that included tests,
source code, compiler and tools to upload binaries to the development board. The source
package and the guidelines were created in versions for the group with tests and the group
without tests. The guidelines in Appendix A are for the participants with tests.
Before the experiment was executed with the 16 participants, one participant tested the
experiment to check that it was easy to understand, how long time it took to finish the
experiment for a participant and that everything worked. The participant had about the
same experience as the participants, the experiment took over 3 hours and we decided to
skip test cases and upload to hardware. The result of this participant was included with
the results.

4.3.2

Execution

The experiment was performed on 2-3 May by four groups that consisted of four participants. Since half of the group had no experience with the C programming language, they
got help with syntax and a few hints on how they could solve the tasks. The participants
used in average 2.5 hours to finish the experiment, where the fastest was finished in less
than 2 hours, and the slowest used almost three and a half hour. The participants were
observed during the experiment to look at how they solved the tasks and what kind of
problem they experienced with the unit tests.
Before the execution of the experiment, all necessary source code, tools and compiler
was download and prepared. This was done so that they could use most of the time on
implementing the functionality given by the tasks. The participants did not use hardware
and execution of test cases. This was necessary to reduce the time used by the participants.
After the participants had finished the experiment, all the tasks were compiled and upload
to the development board, before the test cases was executed. This was not done by the
participants. The test cases for the tasks is found in Appendix B.

4.3.3

Data validation

Data was collected from 17 participants, where 8 executed the experiment without unit
test and 9 finished it with unit tests.
Three of the participants had little experience with programming and had problems with
many of the tasks. Two of these participants executed the experiment with unit tests, and
ended with 8 errors each, which is twice the mean value. The participant without tests
ended up with the same number of errors as the mean value for the participants without
tests.
The results from the experiment are found in the next section.
53

4.4

Analysis & Interpretation

This section uses the experimental data from the execution of the experiment to present
these data and make conclusion that is made from the results.

4.4.1

Descriptive statistics

This part presents the results from tests that were executed after the participants have
finished the experiment. The presentation consists of a boxplot of the data and a graph
that shows the outliers.
Boxplot
The boxplot in Figure 4.1 shows the number of errors for each of the participants seen in
Table 4.1. Figure 4.2 shows the distribution of errors for participants with and without
tests. In the last figure the x-axis shows the number of errors, and y-axis shows the number
of participants that had the given amount of errors. This graph is used to identify the
outliers.
Table 4.1: Errors per Participant
With tests

Without tests

8
7
8
4
9
7
8
8

2
2
4
8
3
2
4
8
3

Results
The total results for each of the test cases can be seen in Table 4.2.

54

Figure 4.1: Boxplot of Test Results

Figure 4.2: Distribution of Errors

55

Table 4.2: Errors per Test Case

Errors
Test case

With tests

Without tests

Total

Existing functionality is working

Segment LCD initalized
BSP LEDs initialized
LEDs blink alternating
RESET Centralized
Task 1

0
0
0
2
0
2

0
0
1
3
7
11

0
0
1
5
7
13

Existing functionality is working

Buttons is working
LEDs updated on click
Falling edge
Interrupts is cleared
Display updated on click
Task 2

0
1
1
1
1
1
5

0
1
2
0
4
1
8

0
2
3
1
5
2
13

Existing functionality is working

Value incremented on right click
Value inverted on left click
Number is correctly updated on click
Text is correctly updated on click
Task 3

1
1
1
2
7
12

0
0
2
0
7
9

1
1
3
2
14
21

Existing functionality is working

Value is written to segment LCD
Pattern in requirement is correct
Rotation restarts when reach end
Task 4

0
1
4
5
10

0
3
8
4
15

0
4
12
9
25

Existing functionality is working

Caplesense is initialized
Segment Ring correct
Segment Number correct
Display empty on no input
Task 5

0
2
4
0
1
7

0
2
6
5
3
16

0
4
10
5
4
23

Total

36

59

95

56

4.4.2

Observations & results

Observations and results description for the participants is found below; they are divided
into sections for the group without tests and the group with tests.
Participants without tests
This first task was easy for the participants, since they were told what to do in the
instructions. With the test cases, 11 errors were found. The most frequent errors were
that text was not centered as defined in the last requirement, and that the LEDs did not
behave as required. Results for each of the requirements are found below:
TC1-1: No errors were found by any of the participants, since there was no need to
modify any code in this section.
TC1-2: No errors were found during testing.
TC1-3: One error, the participant forgot to initialize the LEDs.
TC1-4: A total of three errors. The main failure is that they did not use the LED toggle
to switch state on the LEDs. Some of the participants tried to add some logic to
first check if the LED was on then cleared or set it.
TC1-5: Seven failures were found. Almost all had written "RESET" left aligned.
The test cases in task 2 resulted in eight errors for the participants without tests. Only a
few errors in this task. It was easy for the participants to do, since they are told exactly
what they needed to do. Description of the errors follows below:
TC2-1: No errors found.
TC2-2: One error found during testing, due to wrong parameters set in the pin setup.
TC2-3: Two errors found since the functions to update LEDs never were called.
TC2-4: No errors found.
TC2-5: A total of four errors, the main reason were that the participants did not know
how to calculate the correct pin.
TC2-6: One error was found, the reasons for errors was that the functions was not called
in the interrupt handler.
Nine errors were found in the third task for participants without tests. Just a few errors
before the binary string should have been written to the segment LCD. There were many
failures on the last test case, since many gave up this task. Some errors were due to
writing 0 and 1 to the char array instead of 0 and 1.
TC3-1: No errors found.
57

TC3-2: No errors found.

TC3-3: Found two errors here, one task was not done and the other set the value incorrectly.
TC3-4: No errors found.
TC3-5: Very difficult task for the participants, only one made this test pass. There were
many different approaches to solve the problem.
In the fourth task, 15 errors were found. Most of the errors are that the string is not
displayed correct. The errors occur when the string is displayed wrongly at the start of
the rotation, or at the end when the rotations shall start from the beginning. Without
tests, it is difficult to verify that the string is written correctly.
TC4-1: None broke the existing functionality.
TC4-2: Three participants forgot to write the result to the segment LCD.
TC4-3: A total of eight errors here. There was many different errors, most of them
displayed the string wrong in the end.
TC4-4: Four errors in this task. The main error is that the string never is set to begin
at the start again, which may result in a segmentation fault.
There were 16 errors in the last task. The main failures are that not all requirements was
implemented, and problems turning the segment rings on and off correctly.
TC5-1: No errors found.
TC5-2: Two errors found in initialization and set-up. The documentation is not good
on describing what each of the functions does.
TC5-3: A total of six errors. Most of the errors are that the indicators are not turned
off when the value becomes lower and that wrong indicators are turned on/off.
TC5-4: Five errors for this test case. Either they forgot to write number or they modified
the value before it was written to the display.
TC5-5: For these tests there were three errors, most of them because they forgot to turn
off either the ring or the number segment.
Participants with tests
The first task was easy and the description told the participants what to do. Two errors
were found in the implementation by the participants with unit tests. Using the tests as
in TDD is possible, since there are only a few errors the tests will give feedback on. The
difference from TDD is that the participants did not need to write the tests. Starting
running the tests from the beginning will result in two failed test cases for methods that
were not called. In addition, there are three tests that are passing, which covers the
existing functionality. The unit tests worked good for the participants for this task, since
the feedback that was given from the test result was easy to understand. Observations
for each of the test cases follow below:
58

TC1-1: No errors were found by any of the participants. Tests for the existing functionality were included.
TC1-2: No errors were found during testing. Mock modules verified that this was done.
TC1-3: No errors, the test checked that this was done.
TC1-4: A total of two errors. The main failure is that they did not use the LED toggle
to switch state on the LEDs. Some of the participants tried to add some logic to
first check if the LED was on then cleared or set. It was difficult to understand why
the test cases were failing with a message that a function was called more times
than expected. It is possible that the solution could have worked, but it could not
be checked by the tests since the test expected that the problem was solved in a
different way.
TC1-5: The participants had no errors. Many correct this error by running unit tests.
It was easy to understand since the tests will report with what string it expects and
what it actually was.
Five errors for the users with test were found in the second task. One of the participants
had 4 errors because of little understanding of how to solve the task. It was easy for the
participants to do, since they are told exactly what they need to do. Users with unit
test were able to correct errors with wrong parameters in pin setup and correct pin to
clear for an interrupt. The output from the tests can be difficult to understand, since
many function calls are done in each test. This will result in that almost all test say
that a function was not called as expected, and the error message for this error is printed
first since the mock modules is verified first. The tests worked well for verifying that the
correct parameters were given to the function that was called. The observations follow
below:
TC2-1: No errors found.
TC2-2: One error found during testing, due to wrong parameters set in the pin setup.
TC2-3: One error found since the functions to update LEDs never were called.
TC2-4: One error found. This error occurred since interrupt configuration was not done.
TC2-5: One error, the main reason was that the participants did not know how to
calculate the correct pin. Participants with unit tests could find out and verify that
correct value was set.
TC2-6: One error was found, the reasons for errors was that the functions was not called
in the interrupt handler.
In the third task 12 errors there were found, which are three errors more than the group
without unit tests. A few errors were found before the binary string should have been
written to the segment LCD. There were many failures on the last test case, since many
gave up this task, and some errors were due to writing 0 and 1 to the char array instead
of 0 and 1. All of the tests run the "loop", which results in calls to the segment LCD.
This make all tests fail if the number or text is not written correctly to the display. The
59

best way to solve this is to declare variables as external in the header-file and use unit
tests to assert the value, in addition to mock modules. One of the participants had four
errors on this task.
TC3-1: One error found where the participant used a break statement in the infinity
loop, the test cases was not able to detect this. This is also difficult to test with a
unit test, since the tests cannot run to an infinitely long time.
TC3-2: One error since the increment was put inside an if expression that never was
true, the error is discovered by the unit test.
TC3-3: One error found here. The functionality was not implemented.
TC3-4: Two errors found. The value was modified inside the main loop in an approach
to write the binary string to the text segment. The error was discovered by test,
but never fixed.
TC3-5: Difficult task for the participants, only two participants made this test pass.
There were many approaches to solve the problem, for example greatest common
divisor and comparing each bit. Unit test do not help when participants do not
know how to solve the task.
In the fourth task, ten errors were found. Two things are tested with mocks here: 1. The
initialization is done correctly. 2. The correct value is written to the display for each
iteration in the loop. It can be difficult to distinguish between the start and the end of
the pattern in the unit tests. Some of the participants took advantage of using unit tests
to guide them to the correct result. Many of these errors with unit tests could have been
corrected by trying and failing. The experiment also shows that it is difficult to test logic
by using mock modules, unit test assertions should be used instead.
TC4-1: None broke the existing functionality.
TC4-2: One participant forgot to write the result to the segment LCD.
TC4-3: Four errors were found. Many different errors, most of them displayed the string
wrong in the end of the rotation.
TC4-4: Five errors in this task, this is more errors than for the group without tests. The
reason for this is unknown, it could either be random or that the participants had
problems to solve the previous task, and did not get this feature tested. The main
error is that the string never is set to begin at the start again, which may result
in a segmentation fault. With unit test it is difficult to understand if the error is
detected at the beginning or the end of the string, since the same string is expected.
Another failure was that the whole string was rotated inside the loop without delay,
the participant was sure that the result was correct, and that his implementation
was different from what the test expected of calls to the modules.
There were seven failures for the last task. There are many ways to solve this problem,
even if the unit tests fails, since there are many approaches and functions that can be used
to meet the requirements. The unit tests uses one approach, and has expectations for
60

which functions that need to be called. This can be confusing for the participants. The
tests depend on the loop, since the data are retrieved by polling. It was possible to get the
test passing and the test cases failing, since functions were called in the wrong order. It
is possible to configure CMock to generate mock modules that verify that functions calls
are done in correct order, but it will probably be difficult to use in practice. Updating
the segment ring needs eight method calls to the driver. When so many calls are done in
a test, it is difficult to use the output the failing test gives.
TC5-1: No errors found.
TC5-2: Two errors found on initialization and setup. Documentation is difficult to understand. One of the unit tests passed, but failed anyway, since the functions were
called in the wrong order.
TC5-3: Four errors found. The output from unit testing is difficult to understand, since
there are many functions call to the ring indicators. Most of the errors are that
the indicators are not turned off when the value becomes lower and that wrong
indicators are turned on/off.
TC5-4: No errors found.
TC5-5: One error found, forget to turn off the ring indicators. This participant had also
an error on test case 3, and could therefore not see the error.

4.4.3

Data reduction

When looking on Figure 4.2, we see three outliers. The first is the participant without
tests that only had a total of four errors on the tasks. The two other outliers performed
the experiment with tests and resulted in 8 errors, which is twice as much as the other
with tests.
The participant with only four errors without using tests was the most experience programmer, with some C++ knowledge. We do however, not see any reasons to remove this
participant from the analysis.
The two outliers with tests did not have much experience with programming. Half of
the errors for these participants were by failing completely on one of the tasks. The first
had four errors on task 2, where there were totally five errors for all the nine participants
without tests. The second outlier had four errors on task 3, by neglecting test case 5,
this is more than half the errors for all participants with tests. Since they have about
an average numbers of errors of the other tasks, these outliers are not removed from the
analysis, but the errors on task 2 and 3 should be noted.

4.4.4

Hypothesis testing

The hypothesis that developers using unit test based on mock modules results in fewer
errors than developers without any tests is calculated with t-test. Below follows the
calculation of the t-value.
nx = 8
61

(4.3)

ny = 9
Pn

i=1

x =

xi

= 7.375

(4.5)

= 4.000

(4.6)

x)2
= 1.506
n1

(4.7)

n
Pn

i=1

y =

yi

sP

(4.4)

n
i=1 (xi

sx =

sP

n
i=1 (yi

y)2
= 2.398
n1

(4.8)

x y
t= q
= 3.514
s2x /nx + s2y /ny

(4.9)

sy =

Since variance is not equal for the two samples, degrees of freedom are calculated with:
s2y
u = 2 = 2.535
sx
df =

u 2
)
ny
2
1
+ n2 (nuy 1)
n2x (nx 1)
y

( n1x +

(4.10)

= 13.61 = 13

(4.11)

Degrees of freedom is rounded down to 13. By look up the t-value and degrees of freedom,
the p-value is 0.004. The statistical results are shown in Table 4.3.
Table 4.3: T-test Results
Without tests
With tests

t-test

Mean

StDev

8
9

7,375
4,000

1,506
2,398

df

t-value

p-value

13

3.514

0.004

The hypothesis H0 is rejected since the difference with and without unit tests is significant,
with a p-value as low as 0.4%. The reasons for this is that developers with unit test are
able to actually test and correct errors before testing software on hardware. The feedback
given for correct functions calls is also valuable.

4.4.5

Threats

There are three things that threaten the results from being applied to the real world.
That is the experience of the participants in the experiment, the complexity of the tasks
and that it was simulation.
62

The experience with programming among the participants was low, since half of the group
had experience with programming in C or C++. Using a group with higher experience
and good knowledge would reduce the number of errors to near zero, and the group with
tests should have zero errors, but they could also have errors. In the tasks there were
found two types of errors that could not be found by the test. The first of these errors
was that function calls where done in wrong order. The second error was that a break
statement was used inside the infinity loop, which could not be discovered by the tests,
since they only run the loop on time.
The experiment was simulated; it was not performed on a real problem, but for a few
tasks that was created for the experiment. Testing this on a real problem would look
on other aspects of testing with unit tests and mock modules since the problems become
more complex.
The complexity of the tasks is low, since the participants has little experience with programming. This makes it difficult to test more advanced functionality and applications
with much more code. How well the testing of an application with over thousand lines of
code would work is difficult to say after this experiment, but it is reasonable to believe
that errors also will be found in a complex system. In a more complex system, it is natural
to use more unit tests to test the logic, and the complexity would also need more test
instrumentation.
Since the experiment is a simulation of a real problem and it was aimed to last two hours
for each participants, it may be difficult to compare the experiment to a real situation.
But even in a real situation, the mock modules will work in the same way, and assert the
number of calls and the parameters of these calls. In a real problem there will also be
more experienced developers that will be able to solve all tasks, so the significant result
may be lower.

4.5

Summary

The experiment with comparing not using tests and testing with mock modules was done
by 17 participants. Using mock modules to unit test the software was significantly better
than not using tests. It was difficult to find participants to experiment, and second grade
student was used, which have limited experience with programming and programming in
C.
There is a difference in where the participants with tests have errors. In the first two
tasks, where the users had clear requirements to implement and good guidelines on how
to solve it, there were only a few errors. On the three last tasks there is more errors, the
participants need to implement some logic, and the participants with little experience has
problems implementing the logic. It is also difficult to understands the results from the
mock objects here. It will probably be better to use unit tests. The structure from the
examples provided by Energy Micro was used for the tasks, but was changed as little as
possible to be able to write tests. Structuring the code even better would make it easier
to also add unit tests. It is also possible that using a stub function also would help, if
functions with logic are called many times.
The mock modules verify that a function is called, and that the correct parameters are
passed to the method. There were found errors that not could be discovered by tests,
63

since the methods was called in the wrong order. This could be solved by also verifying
the order of calls, but might results in that there will be more difficult to change code
and write tests. Understanding the results from the mock modules is difficult when we
have many method calls to the same method in a function under test. When many mock
modules is used in the same test, the test will fail and return the error of the first mock
module that fail, which may give unexpected messages as a test results, which is difficult
to understand. This can in many cases be solved by having a better structure where a
function has fewer dependencies to other modules.
The result from the experiment is significant at the 1% level, and the mock modules will
find errors and verify the correct behavior also in real problems. Testing of logic will be
tested better with a better structure of the code and when using unit test to assert this
part. Using mock modules to verify behavior, will work as regression testing to check that
correct calls still are done to the dependencies.
Further work will be to use mock modules in a real project that is developed by experienced
developers. It can also be important to look at how well use of mock modules will work
when changes are done in the modules that depend on hardware, where there are only
used mock modules.

64

CHAPTER

5
DISCUSSION

The section discusses the results from the previous chapters and does an evaluation of
this. The chapter does also answer the goals that were given in the first chapter of the
thesis.

5.1

Experiment

The result from the statistical analysis in the experiment showed that use of mock modules
was significant better than not performing any tests of the software. The result was not
a surprise since testing will reduce the amount of errors in software. The average number
of errors for participants with tests was 4 and the average number of errors was over
7 for participants without tests. There were a few outliers that were included in the
analysis, but without these the difference would have been bigger. The outliers come
from the difference in experience by the participants, and the average participant had low
experience in programming in general and with programming in C. The low experience
among the participants made it necessary to simplify the experiment. The low experience
and the simple task are probably the biggest threats to validity of the experiment.
Another threat to validity is that the experiment is only a simulation, and is not based on
a real problem. If the experiment was run with experienced developers with knowledge
about the Energy Micro microcontrollers, the differences between in errors would probably
be smaller and the average of errors would be much lower. It would also be better to try
the experiment in development of a real example. Because of the significant results in
the experiment, it is likely that use of mock modules would reduce the number of errors
even with experienced developers in a real software project. Since the problems and
tasks are only simulation, they are still similar to the examples that are created for the
STK3700 start kit from Energy Micro. The main difference is that the examples used in
the experiment are simpler.
The first research question in thesis is Is it possible to reduce the number of errors
during development of new features, modification and refactoring of existing functionality
by using mock modules to verify the desired behavior of the system? From the statistical
analysis from the experiment, the results of this question is that it is possible to reduce
65

the number of errors when developing new features or modifying existing functionality.
The experiment did not look at refactoring, since this is complex and require code review
to see that it actually was changed and run the tests to see if it is working. By injection
an error into an example that has good unit test coverage and mock modules, it is possible
to detect this error by running the tests, so using mock modules would probably reduce
the number of errors in refactoring too.
In addition to the statistical analysis we did some observations during the experiment.
The main problem with test results, is that it can be difficult to understand the feedback
given, especially when many calls are done to the same methods, which will results in an
message saying that a method was called to few or to many times. The same problem
occur when many of the tests involves the same modules, where the tests will verify the
same mock modules first, and all tests will fail because of these, without possibility to
see if other features are working. Testing logic in the application with mock modules
also makes it difficult to understand results, so testing logic should by unit test assertions
instead of mock modules, where mock modules is used to solve some of the dependencies.
From the test results it is possible to see that the number of errors is much higher for
participants with tests on task 3 and 4, which are the tasks with most logic. Most of the
problems above can be solved by making the code more testable, by using more functions,
where each function has fewer dependencies.
Two errors was also found that not was discovered by the unit tests, one of these could
have been found by changing the configuration of CMock, the other could not have been
found by the unit tests. But software unit testing is only one necessary test activity, and
some errors will be found in other test activities. It is impossible to be able to detect all
kind of errors in software unit testing. Developers will also create errors when writing
errors while writing unit tests.
Advice 1: Use of mock modules reduces the number of errors in development and
refactoring of embedded software.
Advice 2: It is difficult to test logic with mock modules, so variables in the examples
should be made accessible for the tests, so that the variables can be checked with unit
tests assertions instead.
Advice 3: A good code structure will make it easier to write tests for the software, the
developers should therefore focus on testability.
Advice 4: The units that are tested should be as small as possible, because dependencies to many mock modules may result in feedback from the tests that are difficult to
understand.

5.2

Mock Modules

If Energy Micro wants to do software unit testing of their software library, mock modules
will be an important part of the solution. Mock modules will be important for three
66

reasons. The first is that it does internal assertions of parameters, and assertions against
the expectations that are set for the Mock modules can also be used to inject return
values, when needed. By doing this, we can test that a unit calls the correct methods
and that the dependencies are removed from the unit during test. The second reason
is that hardware is not available during tests, and by using mock modules these can be
used instead of the real modules that need to access hardware, and return values from
hardware can be injected into the mock modules when setting expectations. The third
reason is what we will be able to capture data that are written to registers that cannot
be read and asserted, but the expectations can be set to verify all calls.
Energy Micro will need to create mock modules for all the peripheral APIs, the board
support packages and the device drivers. This is needed to be able to perform software
unit testing, where the units are as small as possible. The mock modules will be important
in testing of the examples, where the experiment showed that the number of errors was
reduced with use of mock modules during testing. During software integration testing
and testing of peripheral APIs there is also need for some mock modules for methods that
read or write to memory registers, to be able to capture the data and do assertions on
data that are read or written to hardware.
CMock is a great tool for generating mock modules, and automates much of the work
with creation of test doubles. A few problems have been found with CMock: the most
important one is probably the support for inline methods and functions defined in macros.
This can easily be solved by writing a script that adds non-inline methods during test.
CMock also lacks support for doing assertions on structures and custom types. Currently
these assertions are done by comparing the memory of the actual and expected structure.
CMock has announced that CMock will support structures and custom types in later
version, but currently there is no development in work. CMock is also open-source project,
so participation in development is possible if features are needed.
Observations from the experiment showed that too many dependencies also will results in
many mock expectations, which can be difficult for the developer to interpreted. A typical
structure of an example is a main method, which consists of set up of the microcontroller,
and a loop that contains most of the functionality. In addition there may be a few methods.
To improve the testability in the examples, it is necessary to split the functionality into
methods to reduce the responsibility for each of the existing methods. Many of the
memory registers is accessed directly from a function. When these read or write to
memory registers that cannot be read by the test harness it is possible to verify that the
correct values was read or written to the registers. To be able to test these it might be
necessary to create inline methods that access the memory registers, with this method
it is possible to create a mock module that can capture the data that are written, or to
inject return values for data that is read.

Advice 5: To be able test the examples, it is necessary to create mock modules of emlib,
drivers and board support package.

Advice 6: It is necessary to create script that modifies the inline functions in the
header-files, so it will be possible to generate mock modules with CMock.
67

Advice 7: To improve testability, it is necessary to split the functionality into methods

to reduce the responsibility for each of the existing methods.
Advice 8: It may be necessary to create inline methods to read and write to registers, so
mock modules can be create the capture the calls and inject return values to the register.

5.3

Embedded Testing

Unity and CMock works well together when testing embedded software. Where CMock is
used for generation of mock modules that are used for testing, all assertions in CMock are
with Unity. Unity has a wide range of different assertions that can be used, and many of
these are created for the needs of an embedded systems. The main problem with CMock
is that it does not accept inline methods and functions defined in macros, but this can
be fixed by preparing the header-files with a script before generation of mock modules.
Ceedling is also a good tool. This is an automated building system that uses CMock and
Unity. Both Unity and CMock are important tools when it comes to testing of embedded
systems. Ceedling is also a good tool for automation of building, but it is unknown if this
can be integrated with Energy Micros current system for building and testing.
Most of the problems that occurs when writing a test can be solved by using mock modules
to remove dependencies, creating macros to redefine or change functionality that is not
wanted or available during testing. The last thing that is used to change behavior during
tests is to use conditional preprocessor expression. By using this we can for example use
a pointer to a structure instead of a pointer to a fixed memory address. Not all can be
solved by having good testing framework, it is also necessary to structure the code so that
it is testable.
One complexity in testing for Energy Micro is that examples shall work with four different
IDEs, examples shall work on different development boards and the peripheral API shall
work for all EFM32 microcontrollers. The tests for the examples and peripheral APIs can
be reused, by compiling and running the tests for each microcontroller, which is defined
for the preprocessor. Due to for example different peripheral devices and sizes of memory
for each microcontroller, it may be necessary to modify the tests in some cases. This part
of testing has not been done in this study, and there is still work left, to find out how to
solve this in an efficiently way.
The second research question for the thesis is Show that CMock is a suitable framework
for creating mock objects, which can simulate the behavior of the hardware, which is
not available during testing on a native system. CMock is easy to configure and creates
mock modules that works well during tests. There are some problems with structures
and custom types and that inline methods are not accepted as input, but these problems
are manageable. By having a framework that can generate mock modules and a unit test
framework to do assertions, unit tests can be written to test software on a native system.
When having tests that can cover most of the source code, errors in the system will be
reduced. One of Energy Micros goals is reduce the number of errors in the code before
release. CMock and Unity should solve this, since adding tests will reduce the number of
errors and testing with CMock and Unity will work on Energy Micros code.
After writing tests for the software library, the library will consist of a lot of test in68

strumentation code, which is only used to test the system. Before release of the code to
customers, there will be necessary to remove all the test code instrumentation from the
source code. Remove of test instrumentation can be done by a script, and the reason for
removing the test instrumentation is that the customers have no need for this, and will
it make the code harder to read and understand.
Advice 9: Most of the problems that occur during testing can be solved by using mock
modules, macros and redefined functionality during test by using a preprocessor.
Advice 10: CMock works well in generation of mock modules that are needed for
testing. By using a tool that automatically generates mock modules, developers can
reduce time used for testing since they do not need to write the mock modules manually.
Advice 11: Before release of code, it will be necessary to remove test instrumentation
code, to make the code more readable and easier to understand.

5.4

Test Strategy

This thesis has only looked at software testing and mainly at software unit. Section
2.3 gives a brief overview of both hardware and software testing. Since much of the code
involves interaction and calls to lower layer in the software library, it is necessary to remove
the dependencies during software unit testing. To be able to remove these dependencies,
mock modules is an important part of the solution. By using these modules in tests,
the unit can be tested in isolation, and the behavior can be tested automatically. Unit
testing is important to remove errors in the unit that is tested. These tests can be used to
verify changes during modification, and that refactoring do not change or break existing
functionality. The unit tests will find and reduce the number of errors in the units, but
will not test the interaction between software components. Unit testing is most important
for the examples and drivers. It is also important for the peripheral API, even though
this layer mainly consists of an abstraction layer that write configurations to memory
registers, the layer consist of some logic that needs to be tested.
Software integration testing is testing of the interactions between software subsystems.
When doing software integration testing, use of mock modules will be less important,
since most assertions will be done by verifying that the correct registers are set by the
software. Mock modules are needed when hardware interaction is necessary to be able to
capture data that is written or inject data response when data is read.
Integration testing is testing to verify correct behavior when software interacts with hardware. Integration testing will normally require more manually work by the testers, since
it often will require user interaction during testing. This is especially true when testing
the examples, but testing of the peripheral APIs this can be done more or less automatic
by unit tests. Testing on hardware is also the most correct place to test the hardware
abstraction layer, but testing on a host computer can also reveal errors, and it will support
test-driven development.
Energy Micro wants to reduce the number of errors before release of software. In addition
they have many examples that need to be tested, so an additional goal is to be able
69

to minimize the testing of examples on hardware, since this is time consuming. This
means that Energy Micro needs to reduce integration testing of the examples. By having
code that is covered by unit tests, where software is tested both with software unit tests
and software integration tests, the number of errors may be reduced, because unexpected
behavior will be found by the unit tests. When these errors are found early, the need for
testing on hardware is reduced. Testing on hardware will still be necessary, but the testing
can be focused on examples that are directly affected by the changes in the software.
Advice 12: All errors cannot be found by using unit test. Use of integration testing
will still be necessary to verify that all parts of the system works as expected.
Advice 13: It is important to create a test strategy, so that the company has a plan
for how testing of the software library shall be done.

5.5

Agile Development and Test-driven Development

One of the goals for Energy Micro when it comes to unit testing is to take advantage
of agile development, with unit testing and test-driven development. The main changes
needed to take advantage of agile development are related to how the project is structured
and that development is done in iterations. By developing in iterations, it is possible to
make decision late and adapt changes that occur during the development. Automated
testing with unit tests is often used in agile development, and test-driven development
is a technique that guarantees that tests are written for code that is developed. To use
TDD, it will be necessary to have test frameworks that will support the test activities.
Unity and CMock will work with test-driven development.
Automated building and testing tools is also important to support test-driven development, since tests are run often by the developers. Ceedling is a tool that is developed
to support test-driven development with Unity and CMock. Ceedling was tested and has
good features to perform testing during TDD. It can be difficult for Energy Micro to take
advantage of using Ceedling since they already have an environment for compiling and
testing their software. Another issue is that Energy Micro does not develop against a
standalone product for a microcontroller, but creates many different examples for many
different microcontrollers. How this support is in Ceedling is unknown.
Much of the code and examples that are written by Energy Micro is without tests, which
makes it hard to do automatic regression testing and detect errors that are created during development and modification of existing functionality. Test-driven development is
addressed to writing test before creating any code, but when changing code without tests,
there will be a need to write tests for the unit that are going to be changed first. When
tests are created it is possible to do modifications or do refactoring, since the test now
will fail if some of the existing behavior is broken.
Section 2.2.4 refers to an article of how TDD can be done effectively for development on
embedded systems. In section 2.4.3 we show how to apply test-driven development on
existing code without tests.
In the experiment, half the participants had access to tests that was created before they
started on a task. This is test-driven development, where tests are written before code
70

is implemented, without the refactoring step. A problem that was observed is that it
is difficult to understand the feedback that is given by tests that are failing. In the
experiment all the tests was created first, and many of these tests covered almost all the
functionality. By having smaller units the feedback that is given during the tests will be
easier to understand.
Advice 14: Keep the size of the units under development in each of the iterations in
TDD as small as possible.
Advice 15: Use of building tools that supports test-driven development will help the
developers. Ceedling is a tool that can support test-driven development with Unity and
CMock.
Advice 16: Before existing code without tests is changed, unit tests should be written.

5.6

Regression Testing

The last goal for Energy Micro is to be able to refactor or change the code on a later
time, without breaking the code. Automatic regression testing with unit tests will be able
to detect errors that are created, both in the software unit (by running the software unit
tests), and errors that may occur in the in the interaction between software components.
Not all errors can be found by the unit tests, so there will still be need for testing on
hardware, to ensure that it behaves as expected. But by having good code coverage, the
need for integration tests will be lower.
Good code coverage with tests is the most important part of being able to test changes
when refactoring, since most of the functionality will be tested and asserted to see if it
behave as before the changes was done. Even for a software library that do not have
many tests, much can be solved by adding tests to the code before changes are done, so
at least these changes are tested. But these changes can still break the functionality in
components that depends on the unit that was changed. Errors can also be created if the
model that the developer creates tests for do not match the real model. To detect errors
like these, integration tests are needed.
Another way to reduce the need for testing all examples, is to be able to trace the changes
that are done to the code. Doing a change does not necessarily mean that all examples
will need to be tested. There is no need to test examples that do not use the component
that was changed. By having a table or matrix that shows which examples uses which
modules. It will be possible to see which examples that actually needs to be tested.
When testing during development or with test-driven development it is not possible to
carry out all the tests every time. A repository that has a good coverage would consist
of several thousand tests and even more assertions. To be able to verify that all tests are
passing, it will be necessary to run the tests incrementally. There are two suggestions for
this, one is to run night builds and run all tests that are available for the software, or run
them on commits to the repository. The last can be difficult if there are a large amount
of tests or commits to the repository, since the build servers not will be able to keep up.
71

Advice: 17 To reduce the need for integration testing, it will be useful to have a table
or matrix showing the dependencies between examples and modules.
Advice: 18 A developer cannot run all tests every time. Therefore it will be necessary
to run nightly builds that run all tests in the repository.

72

CHAPTER

6
CONCLUSION

Below follows the conclusion for the thesis and further work for testing of embedded
software.

6.1

Conclusion

The first part of the thesis looks at testing of Energy Micros software library, where
examples, drivers and peripheral APIs are tested. The second part is an experiment to
check that testing with mock modules will reduce the number of errors, compared to not
using any techniques for automated testing. In the beginning of the project, two research
questions were defined.
RQ1: Is it possible to reduce the number of errors during development of new
features, modification and refactoring of existing functionality by using mock
modules to verify the desired behavior of the system?
The experiment showed that testing with mock modules would reduce the number of
errors that are created during implementation and modification of existing functionality.
RQ2: Show that CMock is a suitable framework to create mock objects, which
can simulate the behavior of the hardware, which is not available during testing on a native system.
CMock is a suitable tool for automatic generation of mock modules from header-files.
CMock do not accept inline methods or macro functions as input, but this can easily be
fixed by creating a script that modify the header files. In addition, CMock do not have
any real support for custom types and structure, but there are several ways that this can
be solved.
By using Unity and CMock as frameworks for testing, Energy Micro will be able to reduce
the number of errors in their software and make it possible to verify that functionality
is not broken during modification and refactoring of software. By having the ability
to automate testing of the software, Energy Micro can take advantage of test-driven
development. Use of CMock to generate mock modules and use the mock modules to
solve dependencies and perform testing without hardware, will reduce the amount of
73

errors in embedded software.

6.2

Further Work

The most important further work will be to check the validity of the experiment by trying
the experiment in a real situation, where professional developers use mock modules in the
project and trying out test-driven development as a methodology. By using mock modules
in a real environment, it can be tested whether the use of mock modules really works,
and that the result in the experiment is significant.
Ceedling is a tool that makes testing easier with Unity and CMock, and which can support
test-driven development. This thesis has looked only briefly on Ceedling and just made
a simple test with it on Energy Micro software. It is necessary to make a more detailed
study of this tool. In addition it will be necessary to look at how tests can be efficiently
reused, when tests are run for several different microcontrollers that have different features
and other differences.
This study has mainly looked at testing of software on a native computer, and both testing
on emulators and development board is possible. It would be interesting to see if testing
on an emulator or simulator could be more efficient than testing the software with mock
modules. Testing the software on hardware is important, so looking for solution that can
test the software automatic on hardware is an important issue. Looking for solution for
how to reduce the necessary user interaction during testing on hardware is also an issue.

74

BIBLIOGRAPHY

[1] ARM
Cortex-M.
http://www.arm.com/products/processors/
cortex-m/. [Online; accessed 21. February 2013].
[2] Ceedling
Intro.
http://throwtheswitch.org/white-papers/
ceedling-intro.html. [Online; accessed 21. April 2013].
[3] CMock
Intro.
http://throwtheswitch.org/white-papers/
cmock-intro.html. [Online; accessed 08. March 2013].
[4] CMock Summary. https://github.com/ThrowTheSwitch/CMock/blob/
master/docs/CMock_Summary.md. [Online; accessed 21. May 2013].
[5] Energy Micro. http://www.energymicro.com/. [Online; accessed 21. February
2013].
[6] RAKE Ruby Make. http://rake.rubyforge.org/. [Online; accessed 1. June
2013].
[7] Unity
Intro.
http://throwtheswitch.org/white-papers/
unity-intro.html. [Online; accessed 08. March 2013].
[8] Unity Summary.
https://github.com/ThrowTheSwitch/Unity/blob/
master/docs/Unity%20Summary.txt. [Online; accessed 21. May 2013].
[9] White Papers - CMock, pointers, and structs. http://throwtheswitch.org/
white-papers/cmock-pointers-and-structs.html. [Online; accessed 21.
May 2013].
[10] White Papers - Mocking inline functions or macros. http://throwtheswitch.
org/white-papers/mocking-inline-functions-or-macros.html.
[Online; accessed 01. March 2013].
[11] Energy Micro AS. EFM32GG Reference Manual, 0.96 edition, 24 April 2012.
[12] Energy Micro AS. User Manual Starter Kit EFM32GG-STK3700, 0.11 edition, 31
May 2012.
75

[13] Bart Broekman and Edwin Notenboom. Testing embedded software. Addison-Wesley,
2003.
[14] M. Dowty. Test driven development of embedded systems using existing software
test infrastructure. University of Colorado at Boulder, Tech. Rep., Mar, 2004.
[15] Michael Feathers. Working effectively with legacy code. Prentice Hall, 2004.
[16] J.W. Grenning. Test Driven Development for Embedded C. Pragmatic Bookshelf
Series. Pragmatic Bookshelf, 2011.
[17] ISO/IEC JTC1 SC22 WG14. ISO/IEC 9899:TC2 Programming Languages - C.
Technical report, May 2005.
[18] M.J. Karlesky, W.I. Bereza, and C.B. Erickson. Effective test driven development
for embedded software. In Electro/information Technology, 2006 IEEE International
Conference on, pages 382387. IEEE, 2006.
[19] Kim H Pries and Jon M Quigley. Testing complex and embedded systems. CRC Press,
2011.
[20] S. Raffeiner. Functionality and design of the CMock framework.
[21] Mark VanderVoord. Embedded Testing with Unity and CMock. Throw-The-Switch
Productions, 2010.
[22] M. Williams. Low pin-count debug interfaces for multi-device systems. 2009.
[23] C. Wohlin. Experimentation in Software Engineering: An Introduction. The Kluwer
International Series in Software Engineering. Kluwer Academic, 2000.
[24] W. Wolf. Computers as Components: Principles of Embedded Computing System
Design. The Morgan Kaufmann Series in Computer Architecture and Design. Elsevier
Science, 2008.

76

LIST OF ABBREVIATIONS

ACMP Analog Comparator

ADC Analog to Digital Converter
AES Advanced Encryption Standard
API

Application Programming Interface

CMSIS Cortex Microcontroller Software Interface Standard

CMU Clock Management Unit
EFM Energy Friendly Microcontrollers
EFR Energy Friendly Radios
emlib Energy Micro Peripheral Library
GCC The GNU Compiler Collection
GPIO General Purpose Input/Output
HAL Hardware Abstraction Layer
IDE

Integrated Development Environment

JTAG Joint Test Action Group

LCD Liquid Crystal Display
LED Light-emitting Diode
PRS Peripheral Reflex System
RISC Reduced Instruction Set Computing
SWO Serial Wire Output
TDD Test-driven Development
77

UART Universal Asynchronous Receiver/Transmitter

YAML YAML Aint Markup Language

78

APPENDIX

A
EXPERIMENT GUIDELINES

A.1

Introduction

In this experiment you will implement five small examples that use some of the features
that are available on EFM32 Giant Gecko STK3700 starter kit. The code you implement
will be tested later, and the results will be compared between the groups that have access
to unit tests and the participants without tests during implementations.
The goal for the experiment is to see if regression testing with unit tests and mock
object/modules can reduce the number of errors before software is uploaded to hardware.
In the experiment you will finish five simple tasks, which will demonstrate some of features
on the development board. The examples that you implement will use the segment LCD,
LEDs, buttons and the capacitive sensor on the development board. Each of the tasks
have a set of requirements that must be implemented and instructions of what to do in
each task.
After the implementation is done and the binary file is uploaded to the development
board, you are not allowed to modify the code for the current task.
The participants will be divided in two groups. One on of the groups will have access to
unit tests during the experiment, and can run these whenever they want to. The unit
tests will use mock modules to simulate the behavior of the hardware, since hardware
is not available during unit testing. The mock modules are used to verify that the correct
function calls are performed, with correct parameters. Return values are injected from
the mock modules. The expectations and injection of return values is set in the tests.
When tests are run, the correct number of calls and parameters to the function calls are
compared.

A.2

Instruction

This section consists of instructions for downloading the necessary code, overview of the
documentation, instructions for upload of code to microcontroller over J-Link and a short
introduction to the developpment board.
79

A.2.1

Code

Before you begin on the experiment you will need to download and unpack the necessary
files. The tools and source code are packed. The archive includes the CodeSourcery compilers, energyAware Commander that is used for uploading binary to the microcontroller
and the source code.
Download the source code and tools with:
# wget http://folk.ntnu.no/eribe/eutf/zebkrcdu.tar.gz

Unpack the files:

# tar zxvf zebkrcdu.tar.gz

All the necessary examples, the compiler and tools will now be available in eutf/ folder.

A.2.2

Documentation

You should have received the necessary documentation for the experiment. The documentation includes the Doxygen documentation for the following modules:
The GPIO (General Purpose Input/Output) peripheral API.
Segment LCD driver.
Capacitive sense driver.
The Board Support Package for the LEDs on the starter kit.
The documentation can also be opened on the computer with:
# evince eutf/docs/refman.pdf

A.2.3

Upload to development board

Upload to development board will be done with energyAware Commander, this tool can
be found in:
# cd

/eutf/tools/energymicro

The tool is started with:

# ./eACommander.sh

Figure A.1 shows the startup screen, to connect to the development chose the correct
J-Link device and click connect. To upload a binary to the board, click flash on the left
side.
After clicking "flash", browse the file you want to upload, and click Flash EFM32. The
binary should now be uploaded to the development board. An example is shown in
Figure A.2

80

Figure A.1: Connect with energyAware Commander

Figure A.2: Upload with energyAware Commander

81

A.2.4

Development board

An overview of the development board can be seen in Figure A.3, the figure is taken from
the STK3700 user manual. In this experiment LEDs and the segment LCD will be used
as output, and push-buttons and touch slider as input.

Figure A.3: STK3700 Hardware Layout [12]

Segment LCD
The driver documentation for the segment LCD is found on page 38 of the documentation
that is handed out. The segments that will be used on the display are:
Text: It is possible to write 7 characters to the text segment on the display.
Numbers: 4 digits can be written to the number part of the segment display.
Ring indicators: The ring indicator consists of eight indicators that form a ring.
User LEDs
On the development board there is two LEDs, these will be used in most of the task.
Documentation is found on page 27 of the Doxygen documentation. LED1 is the left
LED on the development board, and LED0 is the right LED.
User push-buttons
There is two push-buttons on the board that will be used for input. The necessary
documentation for this is found in the documentation that is handed out. The buttons
will use the GPIO peripheral API.
82

 Button 0 (left button) is connected on pin 9, port B.

Button 1 (right button) is connected on pin 10, port B.
Touch slider
The touch slider uses the capacitive sensor. It will be used for input in the last task, and
documentation for this is found in the capacitive sense driver on page 30, of the Doxygen
documentation provided.

A.3

Tasks

The experiment consists of five tasks, which follow below.

A.3.1

Task 1

This task will finish an application, which is a counter that starts on a given value and
counts downwards. Your task will be to initialize the LEDs and the segment LCD. In
addition you will have to implement the behavior when the counter reaches zero.
Requirements
Below are the requirements you will need to implement.
REQ1-1 The system shall initialize the segment LCD without voltage boost.
REQ1-2 The system shall initialize the LEDs on the development board.
REQ1-3 The system shall write "RESET" centered on the displayed, when the counter
reach zero.
REQ1-4 The LEDs shall blink alternating, when the counter reach zero.
Instructions
The project folder for this task is:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task1/

The source file for the task is found in:

# eutf/src/kits/EFM32GG_STK3700/examples/exp-task1/exp-task1.c

The first part of the task is to initialize the segment LCD and the LEDs on the board,
this must be implemented in the main() function. The documentation for the LCD is
found on page 38, and the board support package LEDs is on page 27 in the Doxygen
documentation. The segmented LCD shall be initialized without voltage boost.
The second part is to implement the behavior when the counter reaches zero, this will
be implemented in loop(). The wanted behavior is that the LEDs on the board blink
alternately and that "RESET" is displayed centered on the display.
83

After implementation you can run the tests. The tests are found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task1/tests

The tests are compiled and run with:

# make -f Makefile.exp-task1.test

The results of the unit tests are written to the screen and you can correct your code
and then re-run the tests until you are finished and want to upload your code to the
development board.
To compile the tests go to the folder below:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task1/codesourcery

and run the following command:

# make -f Makefile.blink

If you have a problem with compiling your code, ask for help.
Upload the binary file to the development board by using the upload instructions. The
binary file is found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task1/codesourcery/exe/blink.bin

Results
If you have any comment, please write them in the box below:

A.3.2

Task 2

The task is to implement the functionality needed to be able to turn on and off LEDs by
using the push buttons on the development board. In addition the segmented LCD will
be used to display the total count of clicks on each push button.
Requirements
The requirements for this task follow below.
REQ2-1 Buttons shall be initialized and set up for interrupts.
REQ2-2 The LEDs shall toggle on click, the left LED is toggled on left clicks and the
right LED is toggled on right clicks.
84

REQ2-3 The button shall give an interrupt on falling-edge.

REQ2-4 An interrupt shall be cleared after an interrupt.
REQ2-5 The segmented LCD shall display "[left-clicks]/[right-clicks]".

Description
The project folder for this task is:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task2/

The source file for the task is found in:

# eutf/src/kits/EFM32GG_STK3700/examples/exp-task2/exp-task2.c

The first part is to configure the buttons as input in the GPIO (General Purpose Input/Output), which will be done in buttons_setup(). Button 0 is connected to pin
9 on port B and button 1 is connected to pin 10 on port B, the function to setup this is
GPIO_PinModeSet() and can be found on page 16 in the documentation. Description
of GPIO_Mode_TypeDef and GPIO_Port_TypeDef is found on page 10 and 11. It will
also be necessary to configure GPIO interrupts, which is done with GPIO_IntConfig(),
and the interrupt should occur on falling edge, see documentation on page 14.
The second part is to clear a pending interrupt, this is done with GPIO_IntClear(),
and documentation is available on page 13. Clearing of interrupts is done in the interrupt
handlers GPIO_ODD_IRQHandler() and GPIO_EVEN_IRQHandler(). GPIO_IntClear()
takes flags as input, which means you will have to set the correct pin to clear. The bit
string 000...01000 is the flag to clear the interrupt for pin 3. Use either an integer value
or use the left shift operator.
The last part is to make the LEDs turn on/off on clicks and write the total number of
clicks to the segment LCD on an interrupt. Implement this where you want to make it
work, but do not modify loop().
After implementation you can now run the tests. The tests are found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task2/tests

The tests are compiled and run with:

# make -f Makefile.exp-task2.test

The results of the unit tests are written to the screen, and can correct your code and then
re-run the tests until you are finished and want to upload your code to the development
board.
To compile the tests go to the folder below:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task2/codesourcery

And run the following command:

# make -f Makefile.blink

If you have a problem with compiling your code, ask for help.
Upload the binary file to the development board by using the upload instructions. The
binary file is found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task2/codesourcery/exe/blink.bin

85

Results
If you have any comment, please write them in the box below:

A.3.3

Task 3

This task uses the buttons to perform simple binary operations on a binary bit string.
Your task is to implement the behavior when each button is clicked and output the values
of the bit string as specified in the requirements.
Requirements
Below are the requirements you will need to implement in this task.
REQ3-1 The binary bit string shall be inverted when the left push button is clicked.
REQ3-2 The binary bit string shall be incremented by one when the right push button
is clicked.
REQ3-3 The number segment on the segment LCD shall show the decimal value of the
binary bit string.
REQ3-4 The text segment shall display the seven last bits in the binary bit string.
Description
The project folder for this task is:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task3/

The source file for the task is found in:

# eutf/src/kits/EFM32GG_STK3700/examples/exp-task3/exp-task3.c

The first part is to implement the logic when the buttons is clicked. On left click, the bit
string shall be inverted, on right click the bit string shall be incremented by one. The bit
string is stored in an 8 bit unsigned integer, defined as value. The maximum value of this
integer is 255. This functionality must be implemented in the interrupts handlers.
In the second part you should write the value of the integer to the number segment on
the LCD. Documentation of the segment LCD is found on page 38. Implement this in
loop().
86

The last part is to write the 7 last bits of the value to the text segment on the LCD.
Implement this in loop().
After implementation you can now run the tests. The tests are found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task3/tests

The tests are compiled and run with:

# make -f Makefile.exp-task3.test

The results of the unit tests are written to the screen and you can correct your code
and then re-run the tests until you are finished and want to upload your code to the
development board.
To compile the tests go to the folder below:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task3/codesourcery

And run the following command:

# make -f Makefile.blink

If you have a problem with compiling your code, ask for help.
Upload the binary file to the development board using the upload instructions. The binary file is found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task3/codesourcery/exe/blink.bin

Results
If you have any comment, please write them in the box below:

A.3.4

Task 4

This task is to modify the behavior when displaying a long string, since the segmented
LCD only have 7 segments for text, and the string will have to loop over the display.
Requirements
Below are the requirements you will need to implement in this task.
REQ4-1 The string shall be written to the segment LCD.
REQ4-2 The display shall loop through the string as shown in Figure A.4.
87

REQ4-3 The loop shall continue from A, when the end is reached.

Figure A.4: Text Pattern

Description
The project folder for this task is:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task4/

The source file for the task is found in:

# eutf/src/kits/EFM32GG_STK3700/examples/exp-task4/exp-task4.c

The task is to make a long string rotate on the segment LCD. The string is defined as
a char array defined as alphabet. The pattern that shall be displayed can be seen in
Figure A.4. Implement this in loop(). You can use snprintf(char, size_t,
const char, ...) to temporary write the output to the string defined as value.
snprintf(char *str, size_t size, const char *format, ...)
*str The character string to be written.
size The number of bytes to write to *str
*format The format string
... Variable arguments, defined by the format string
snprintf(output, 8, "%s", &alphabet[i]) will write a 7 byte string from the
ith byte of the char array (character string) to the array output, the last char in output
will be \0.
After implementation you can run the tests. The tests are found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task4/tests

The tests are compiled and run with:

# make -f Makefile.exp-task4.test

The results of the unit tests are written to the screen and you can correct your code
and then re-run the tests until you are finished and want to upload your code to the
development board.
88

To compile the tests go to the folder below:

# eutf/src/kits/EFM32GG_STK3700/examples/exp-task4/codesourcery

And run the following command:

# make -f Makefile.blink

If you have a problem with compiling your code, ask for help.
Upload the binary file to the development board by using the upload instructions. The
binary file is found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task4/codesourcery/exe/blink.bin

Results
If you have any comment, please write them in the box below:

A.3.5

Task 5

In this task the segment LCD is used as output and capacitive touch slider is used as
input.
Requirements
The requirements for the last task in the experiment:
REQ5-1 The system shall initialize and setup the touch slider.
REQ5-2 The indicator ring on the segment display shall show the position of the touch
slider.
REQ5-3 The number segment on the segment LCD shall display the decimal value of
the touch slider position.
REQ5-4 The segment LCD shall be blank when there is no input on the touch slider.
Description
The project folder for this task is:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task5/

89

The source file for the task is found in:

# eutf/src/kits/EFM32GG_STK3700/examples/exp-task5/exp-task5.c

The first part is to initialize and set up the capacitive sense system. Do not set the
system in sleep. This is important so it is possible to poll for values from the touch
slider. The documentation for capacitive sense driver is found on page 30. Implement the
initialization and setup in main().
The next part is to finish update_segment_ring(int). The segment ring consists of
8 segments, that need to be updated corresponding to the integer value parameter. The
maximum value of this parameter is 48.
The last part is to update the segment number and segment ring on the display for each
iteration. No number shall be displayed when no position is given from the slider. Do
this in loop().
After implementation you can now run the tests. The tests are found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task5/tests

The tests are compiled and run with:

# make -f Makefile.exp-task5.test

The results of the unit tests are written to the screen and you can correct your code
and then re-run the tests until you are finished and want to upload your code to the
development board.
To compile the tests go to the folder below:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task5/codesourcery

And run the following command:

# make -f Makefile.blink

If you have a problem with compiling your code, ask for help.
Upload the binary file to the development board by using the upload instructions. The
binary file is found in:
# eutf/src/kits/EFM32GG_STK3700/examples/exp-task5/codesourcery/exe/blink.bin

Results
If you have any comment, please write them in the box below:

90

A.4

When Finished

When you are finished with all the tasks please deliver your code and this document.
Thanks for participating in this experiment!
If you have any comments please write them in the box on the next page.

91

92

APPENDIX

B
EXPERIMENT TEST CASES

B.1

Task 1

Table B.1 - Table B.5 shows the test cases that was used to test the expected behavior
for the code that was implemented by the experiment participants.
Table B.1: Test case TC1-1
Header

Description

Description
Tester
Prerequisites

Test that system behave as expected during countdown.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Counter is decremented and the "ring" on the segmented LCD is showing the
progress.

Feature

Execution
Expected result

1. Reset the microcontroller by clicking the reset button.

2. Look at the progress on the segmented LCD.
2. Verify that number in the upper-left corner is counting down.
2. Verify that the "ring" on the segmented LCD grows for each eighth that is counted
down.

Table B.2: Test case TC1-2

Header

Description

Description
Tester
Prerequisites

Segment LCD initalized

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The segment LCD is initialized correctly.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Check that the segmented LCD is working.

Expected result

2. Verify that number in the upper-left corner is counting down.

93

Table B.3: Test case TC1-3

Header

Description

Description
Tester
Prerequisites

BSP LEDs initialized.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Initialized LEDs is done.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Wait for the counter to reach zero.

Expected result

2. When counter reach zero, the LEDs shall blink.

Header

Description

Description
Tester
Prerequisites

LEDs blink alternating.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
LEDs shall blink alternating, when countdown is finished.

Table B.4: Test case TC1-4

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Wait for the counter to reach zero.

Expected result

2. LEDs shall be blinking alternative.

Header

Description

Description
Tester
Prerequisites

RESET Centralized
Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Segment LCD shall display "RESET" when countdown is finished.

Table B.5: Test case TC1-5

Feature
Execution
Expected result

1. Reset the microcontroller by clicking the reset button.

2. Wait for the counter to reach zero.
2. Segmented LCD should display "RESET". First and last text segment shall be
empty.

94

B.2

Task 2

To verify correct behvior of the implemented code in task 2, test cases Table B.6 - Table B.11was executed.
Table B.6: Test case TC2-1
Header

Description

Description
Tester
Prerequisites

Existing functionality is working.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Check that CHIP is initalized and loop is running infinitely.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Click the buttons a few times.

Expected result

2. Verify that left LED is turned on or LCD is updated on each click.

Table B.7: Test case TC2-2

Header

Description

Description
Tester
Prerequisites

Buttons is working.
Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Buttons shall be initialzed and interrupts configured correctly.

Feature

Execution
Expected result

1. Reset the microcontroller by clicking the reset button.

2. Click the left button.
3. Click the right button.
2. Verify the segmented LCD shows "1/0".
3. Verify that LEDs or LCD is updated.

95

Table B.8: Test case TC2-3

Header

Description

Description
Tester
Prerequisites

LEDs updated on click.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
LEDs shall be toggle when a button is clicked.

Feature

Execution

1.
2.
3.
4.

Reset the microcontroller by clicking the reset button.

Click the left button.
Click the right button.
Click the left button.

Expected result

2. Verify that left LED is turned on.

3. Verify that right LED is turned on.
4. Verify that left LED is turned off.

Table B.9: Test case TC2-4

Header

Description

Description
Tester
Prerequisites

Interrupts is cleared.
Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Interrupt shall be cleared when a interrupt occur.

Feature
Execution

Expected result

1. Reset the microcontroller by clicking the reset button.

2. Click the left button.
3. Click the right button.
2. Left LED shall only change state once and segment LCD is only incremented by
one.
3. Right LED shall only change state once and segment LCD is only incremented
by one.

Table B.10: Test case TC2-5

Header

Description

Description
Tester
Prerequisites

Interrupt on falling edge.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The button shall create interrupt on falling edge.

Feature

Execution

1.
2.
3.
4.
5.

Reset the microcontroller by clicking the reset button.

Click and hold the left button.
Releas left button.
Click and hold the right button.
Releas right button

Expected result

2.
3.
4.
5.

Verify that left LED is turned on and left counter is incremented to 1.

Left LEDs and segment LCD shall not change.
Verify that right LED is turned on and right counter is incremented to 1.
Right LEDs and segment LCD shall not change.

96

Table B.11: Test case TC2-6

Header

Description

Description
Tester
Prerequisites

Display updated on click.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The number of clicks shall be displayed on the segment LCD.

Feature

Execution

1.
2.
3.
4.

Expected result

2. The text segment on the LCD shal display 1-0.

3. The text segment on the LCD shal display 1-1.
4. The text segment on the LCD shal display 2-1.

B.3

Reset the microcontroller by clicking the reset button.

Click the left button.
Click the right button.
Click the left button.

Task 3

Test cases for task 3 is found in Table B.12 - Table B.12.

Table B.12: Test case TC3-1
Header

Description

Description
Tester
Prerequisites

Existing functionality is working.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Check that CHIP is initalized and loop is running infinitely.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Click the buttons a few times.

Expected result

2. Verify that LCD is updated on each click.

97

Table B.13: Test case TC3-2

Header

Description

Description
Tester
Prerequisites

Value incremented on right click.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The binary bit string shall be incremented by one when the right push button is
clicked.

Feature

Execution

1.
2.
3.
4.

Reset the microcontroller by clicking the reset button.

Click the right button.
Click the right button.
Click the right button.

Expected result

2. The number segment display: 1

3. The number segment display: 2
4. The number segment display: 3

Table B.14: Test case TC3-3

Header

Description

Description
Tester
Prerequisites

Value inverted on left click.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The binary bit string shall be inverted when the left push button is clicked.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Click the left button.
3. Click the left button.

Expected result

2. The number segment display: 255

3. The text segment display: 0

Header

Description

Description
Tester
Prerequisites

Number is correctly updated on click.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The number segment on the segment LCD shall show the decimal value of the binary
bit string.

Table B.15: Test case TC3-4

Feature

Execution

1.
2.
3.
4.
5.

Reset the microcontroller by clicking the reset button.

Click the right button.
Click the right button.
Click the left button.
Click the right button.

Expected result

2.
3.
4.
5.

The
The
The
The

number
number
number
number

segment
segment
segment
segment

display:
display:
display:
display:

1
2
253
254

98

Table B.16: Test case TC3-5

Header

Description

Description

Test that the functionality is implemented correcyly, and the display shows correct
values.
Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The text segment shall display the seven last bits in the binary bit string.

Tester
Prerequisites
Feature

Execution

1.
2.
3.
4.
5.

Reset the microcontroller by clicking the reset button.

Click the right button.
Click the left button.
Click the right button.
Click the left button.

Expected result

2.
3.
4.
5.

The
The
The
The

B.4

text
text
text
text

segment
segment
segment
segment

displays
display:
display:
display:

"0000001".
"1111110"
"1111111"
"0000000"

Task 4

Test cases to verify expected behavior is found in Table B.17 - Table B.20
Table B.17: Test case TC4-1
Header

Description

Description
Tester
Prerequisites

Existing functionality is working.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Check that CHIP is initalized and loop is running infinitely.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Wait a few seconds.

Expected result

2. Verify that LCD is updated on each click.

99

Table B.18: Test case TC4-2

Header

Description

Description
Tester
Prerequisites

Value is written to segment LCD.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The string shall be written to the segment LCD.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Wait for the program to start.

Expected result

2. Verify that text is written to the display.

Table B.19: Test case TC4-3

Header

Description

Description
Tester
Prerequisites

The pattern given in the requirements is correct.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The display shall loop through the string.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Wait for the program to loop through the string.

Expected result

2. Verify that string follows the correct pattern.

Table B.20: Test case TC4-4

Header

Description

Description
Tester
Prerequisites

Rotation restarts when reach end.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The loop shall continue from A, when the end is reached.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Wait for the string to reach the end.

Expected result

2. When the end is reach, verify at the string starts from the beginning again.

100

B.5

Task 5

The test cases that was used for task 5 in the experiment is Table B.21 to Table B.25.
Table B.21: Test case TC5-1
Header

Description

Description
Tester
Prerequisites

Existing functionality is working.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Check that CHIP is initalized and loop is running infinitely.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Use the touch slider.

Expected result

2. Segment LCD shall change when finger is moved on the touch slider.

Table B.22: Test case TC5-2

Header

Description

Description
Tester
Prerequisites

Caplesense is initialized.
Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The system shall initialize and setup the touch slider.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Use the touch slider.

Expected result

2. Data should be read from the touch slider and be displayed on the segment LCD.

101

Table B.23: Test case TC5-3

Header

Description

Description
Tester
Prerequisites

Segment ring correct.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The ring segment shall display the value from the touch slider.

Feature

Execution

1.
2.
3.
4.

Reset the microcontroller by clicking the reset button.

Set finger at the left side of the touch slider.
Move finger slowly to the right side of the touch slider.
Move finger slowly back to the left side of the touch slider.

Expected result

2. None of the segment rings is disabled.

3. The segment rings slowly increase, until a ring is formed.
4. The segment rings slowly decrease, until none ring indicators is visible.

Table B.24: Test case TC5-4

Header

Description

Description
Tester
Prerequisites

Segment number correct.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
Number segment shall display the value from the touch slider.

Feature

Execution

1.
2.
3.
4.

Reset the microcontroller by clicking the reset button.

Set finger at the left side of the touch slider.
Move finger slowly to the right side of the touch slider.
Move finger slowly back to the left side of the touch slider.

Expected result

2. The number segment shall display about 1.

3. The number slowly increases to 48.
4. The number slowly decreases to 1.

Table B.25: Test case TC5-5

Header

Description

Description
Tester
Prerequisites

Display empty on no input.

Experiment participant
Implemented the functionality, compiled the source code and then uploaded the
binary to the development board.
The segment LCD shall be blank when there is no input on the touch slider.

Feature
Execution

1. Reset the microcontroller by clicking the reset button.

2. Set a finger on the touch slider.
3. Remove finger from the touch slider.

Expected result

2. Some values shall be displayed at the segment LCD.

3. No text, numbers or ring indicators is visible on the segment LCD.

102

APPENDIX

C
MOCK PREPERATION SCRIPT

This appendix includes the script that is used to prepare the header-files for mock generation with CMock. The script adds test instrumentation to inline static functions and
includes directives in the header-file, preprocess the file, and start generation of the mock
objects.

C.1

Script

The script to prepare header-files for mock generation is shown in Listing C.1.
1

require yaml

@errors = 0

@config = nil

def parse_braces(line)
chars = line.each_char.to_a
count = Hash.new(0)
chars.each { |ch| count[ch] += 1 }

11

13

15

17

19

21

return (count["{"] - count["}"])

end
def include_accepted?(line)
@config["acceptincludes"].each do |i|
if line == "#include \"#{i}\""
return true
end
end
return false
end

23

25

def process_file(infile, outfile)

curinfile = File.open(infile)
curoutfile = File.new(outfile, "w")

27

103

29

static_inline_line = 0
static_parsing = false
num_brace = 0

31

33

35

37

39

41

43

45

47

49

51

53

55

57

59

61

63

65

67

69

71

73

curinfile.each_line do |line|
line = line.rstrip
curline = line.encode(UTF-8, binary, :invalid => :replace, :undef
=> :replace).split( )
firstword = curline.shift
if static_parsing
num_brace += parse_braces(line)
if num_brace != 0
curoutfile.puts(line)
else
curoutfile.puts(line)
curoutfile.puts("#endif")
static_parsing = false
end
elsif include_accepted?(line)
curoutfile.puts(line)
elsif "#include" == firstword
curoutfile.puts("#ifndef IGNOREINCLUDE")
curoutfile.puts(line)
curoutfile.puts("#endif")
elsif "__STATIC_INLINE" == firstword
static_inline_line = curinfile.lineno
curline
curoutfile.puts("#ifdef TEST")
curoutfile.puts("#{curline.join( )};")
curoutfile.puts("#else")
if line.each_char.to_a.last == ";"
curoutfile.puts(line)
curoutfile.puts("#endif")
else
static_parsing = true
curoutfile.puts(line)
end
else
curoutfile.puts(line)
end
end
if static_parsing
puts "Err: STATIC INLINE was never finished, line: #{static_inline_line
}"
@errors = @errors + 1
end

75

77

curoutfile.close
puts "Finished with #{infile}"
end

79

81

83

def directory_exist(dir)
unless Dir.exist?(dir)
puts "Directory #{dir} do not exist, create it and try again."
exit

104

85

87

89

91

end
end
def file_exist(file)
unless File.exist?(file)
puts "File #{file} do not exist, create it and try again."
exit
end
end

93

95

97

99

101

def get_include_files()
out = ""
@config["includepath"].each do |inc|
out += "-I#{inc} "
end
return out
end

103

105

107

109

111

def get_defined_identifiers()
out = ""
@config["defineidentifiers"].each do |id|
out += "-D#{id} "
end
return out
end

113

115

117

119

121

123

125

127

129

131

133

135

def main
if ARGV.length != 1
puts "Yaml config must be specified\nruby mockfix.rb <yaml-file>\n"
puts "Includefiles must be put in inc/\nAll files that need to be
preprocessed should go in input folder\n"
exit
end
@config = YAML.load_file("#{ARGV[0]}")
directory_exist(@config["inputpath"])
directory_exist(@config["headeroutputpath"])
directory_exist(@config["mockincludepath"])
directory_exist(@config["mocksourcepath"])
directory_exist(@config["tmppath"])
directory_exist("#{@config["cmockpath"]}/mocks")
file_exist(@config["cmockyaml"])
file_exist(@config["cmockpath"])
Dir.foreach(@config["inputpath"]) do |x|
unless x == "." or x == ".."
puts "Starting on #{@config["inputpath"]}/#{x}"
process_file("#{@config["inputpath"]}/#{x}", "#{@config["
headeroutputpath"]}/#{x}")
end
end

137

139

if @errors > 0
puts "Mock fixing finished with #{@errors} errors"

105

141

else
puts "Mock fixing finished without errors\nStarting preprocessing of
headers and genereation of mock objects"
end

143

Dir.foreach(@config["headeroutputpath"]) do |x|
145

147

149

151

153

unless x == "." or x == ".."

unless system("gcc -E -P #{get_defined_identifiers()} #{
get_include_files()} #{@config["headeroutputpath"]}/#{x} > #{
@config["tmppath"]}/#{x}")
puts "Error during preprocessing of #{x}"
end
unless system("ruby #{@config["cmockpath"]}cmock.rb -o#{@config["
cmockyaml"]} #{@config["tmppath"]}/#{x}")
puts "Error during generation of mocks"
end
end
end

155

157

unless system("cp #{@config["cmockpath"]}mocks/*.c #{@config["

mocksourcepath"]}")
puts "Error copying source-files to mock source path"
end

159

163

unless system("cp #{@config["cmockpath"]}mocks/*.h #{@config["

mockincludepath"]}")
puts "Error copying header-files to mock include path"
end
end

165

main

161

Listing C.1: Preperation of Header-files Script

C.2

Config-file

Listing C.2 shows a sample YAML config file for the script to prepare header files.
2

10

12

14

16

#Path to yaml-config file for CMock

cmockyaml: MyYaml.yml
#Path to CMock library
cmockpath: ./
#Path to store temporary files
tmppath: ./tmp
#Include paths
includepath:
- inc
#Path to header-files that are input to mock generation
inputpath: ./input
#Path to where prepared header-files shold be saved
headeroutputpath: ../../../src/emlib/incmock
#Path to output for mock module header- and source-files
mockincludepath: ../../../mocks/inc
mocksourcepath: ../../../mocks/src
#Accepted include directives, will not be removed during preprocessing.

106

18

20

22

24

26

28

acceptincludes:
- efm32.h
- em_device.h
- em_part.h
#Files that need preprocessing before they are included. Functionality not
implemented.
preprocessincludes:
- efm32gg990f1024.h
#Identifiers for preprocessing
defineidentifiers:
- TEST
- IGNOREINCLUDE
- EFM32GG990F1024

Listing C.2: Config-file for Mock Generation

107