Tools and Methods for Continuously Expanding Software Applications

Tools andMethods forContinuouslyExpandingSoftwareApplications—

Andrzej Wasowski

PROCESS AND SYSTEM MODELS GROUP

Variability Is Everywhere

c© Andrzej Wasowski, IT University of Copenhagen 2

Variability Adds Complexity


Domain vs Application ProcessesPohl et. al. Software Product Line Engineering


Handling Variability


Which variabilitytechniques are

effective in practice?

Under what conditions?


AGENDA

Variability Management vs EncouragementArchitectural Patterns for Variability Management

Clone-and-Own with version controlInterpreted configuration filesFeature Modeling and compile-time variabilityDomain Specific Languages (code generation,interpretation)


Two Opposing Approaches to Variability

Variability

Management

Variability

Encouragement

Controlling scopeEliminating variantsIf no significant businessvalue added

Unleashing community innovationEncouraging competition

Letting community decide scope


T. Berger. H. Pfeiffer. R. Tartler. S. Dienst. K. Czarnecki. A. Wasowski. S. She: Variability mechanisms insoftware ecosystems. Information & Software Technology 2014

Some Correlations


Let’s summarize the broad spectrum

Feature modeling works for static variability in engineeringdomains

Feature models scale well in the embedded domain

Open and dynamic ecosystems grow fast!They rely on dynamic binding, runtime-service lookup, and easydownload and installation


AGENDA




Clone-and-Own with Version Control

1.1 2.01.2 1.3

only maintenance (bug fixes) on 1.x branch

tiger

1.1 1.2

uses

tigeruses

eagle

uses

0.1

only maintenance (bug fixes)

lion

Version control records changes to files overtime so that you can recall specific versions

later. It can do it for any software source code,but also for any type of file on a computer.


Scott Chacon. Ben Straub. Pro Git. Everything You Need to Know about Git. 2nd Edition

When is Clone-and-Own beneficial?

Read-onlyaccess toreleasessuffices

Very rarechanges to oldreleases

You do notmaintain norpropagatechangesbetweenvariants


S. Stanciulescu. S. Schulze. A. Wasowski. Forked and Integrated Variants in an Open-Source FirmwareProject. ICSME 2015

Drowning in Clone-And-Own

blue.cc

blue.cc

CHANGE

blue.cc

green.cc green.cc

blue.cc

green.cc

BUG FOUNDINDEPENDENTLY

INDEPENDENTBUG FIXES

?


SUCCESSFUL REUSE IS

PROACTIVEPLANNEDMANAGED


Spectrum of Variability ArchitecturesStay as close to the left as possible

property & configuration files + build system

feature models + build system

feature models + product specific code

domain specific languages + code generation

frameworks + framework completion code

only product specific code (no reuse)


AGENDA




A Configuration FileInterpreted at Boot-Time

A config file controls differences between customersOne SW for all customers, different config fileManage configurations with branching/forking

ProsSimple, easy to maintain, conceptually one productEasier to certify for functional safetySeen in many embedded projects (including automotive,see AUTOSAR, and avionics)

ConsHW resources may be spilledHard to control IP spill (rare)Hard to manage configurations if large


AGENDA




Feature ModelingExample from Czarnecki’02

mandatory

pulls trailerenginebody

car

electric gasoline automatic manual

optional

exclusive choice

inclusive choice

transmission

Hierarchy constraints, for example:manual requires transmission (each child node requires its parent node)

Groups constraints: engine is electric or gas driven or bothNot all constraints in hierarchy & groups, cross-tree constraints in text:

electric requires automatic

Attributes are added like to classes (eg. engine volume)


Feature ModelingConfiguration

mandatory


car


optional

exclusive choice

inclusive choice

transmission


+ ++ +

+ + +c© Andrzej Wasowski, IT University of Copenhagen 26

Applications of Feature Models

Design & Management

domain modeling

product line scopingproduct line mngmt

code generationdriving build system

drivingtesting

Development & Test


Static Variability with Feature ModelsAn Example from The Linux Kernel

1 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES2 if (inode->i_sb->s_magic == ) . . .3 #endif

void pts_sb_from_inode(struct inode * inode)

4 #ifdef CONFIG_UNIX98_PTYS5 pts_sb_from_inode (tty->driver_data);6 #endif

void pty_close(struct tty_struct * tty)

7 tty = kzalloc(sizeof (*tty), GFP_KERNEL);8 pty_close (tty)· · ·

Important to vary onfeatures not customersA model of featuresneeded if manyConnect to build systemsAs always: less is betterPros: simple, youprecisely control resourcesCons: testing and QAtooling cannot be used onall variants simultanously


Integrated Variability and Version Controlengine

gasoline

engine

electric gasoline

1.1 2.01.2

1.3


tiger

1.1 1.2

uses

tigeruses

eagle

uses

0.1



AGENDA




Are all variability models trees?A Fire Alarm System

detection zonesalarm zoneswiringthree differentstructuresModeled asconstrained classdiagrams!Sometimes calledtopologicalvariability

Berger. Stanciulescu. Øgård. Haugen. Larsen. Wasowski. To connect or not to connect: experiencesfrom modeling topological variability. SPLC 2014Fantechi. Topologically configurable systems as product families. SPLC 2013


Modeled using class diagrams in PapyrusA model view showing two model hierarchies

PowerLoopDriverAL_ComInterface

ExternalComs

IO_Module

OperationPanel AutroFieldBus

OperationZoneDomain Group

AutroSafePanel

AFB_PowerControl

Installation

AFB_Unit

AlarmZone DetectionZone

[1]

[*]

[1]

[*]

[1]

[*]

[1]

[1]

[1]

[1..*]

[1]

[1]

[1]

[1..32]

[1]

[*]

[1]

[1..*]

[1]

[0..7] [1]

[1..32]

physicalstructure

logical structure

A home grownconfigurator usedto instantiate mod-els


Feature Models vs DSLs

Feature models are ready and simple (no design effort, deep insight)DSL requires design effort, but rewards with more expressivenessEffort also translates to maintenanceFM effort is offset by existing feature modeling toolsDSL development effort is offset by language workbenches

Feat

ure

M

od

el

## CPUFreq processor drivers#CONFIG_X86_PCC_CPUFREQ=mCONFIG_X86_ACPI_CPUFREQ=yCONFIG_X86_POWERNOW_K6=yCONFIG_X86_POWERNOW_K7=yCONFIG_X86_POWERNOW_K7_ACPI=yCONFIG_X86_POWERNOW_K8=yCONFIG_X86_GX_SUSPMOD=yCONFIG_X86_SPEEDSTEP_CENTRINO=yCONFIG_X86_SPEEDSTEP_CENTRINO_TABLE=yCONFIG_X86_SPEEDSTEP_ICH=yCONFIG_X86_SPEEDSTEP_SMI=yCONFIG_X86_P4_CLOCKMOD=mCONFIG_X86_CPUFREQ_NFORCE2=yCONFIG_X86_LONGRUN=yCONFIG_X86_LONGHAUL=yCONFIG_X86_E_POWERSAVER=mC

on

figu

rati

on

DSL

Me

ta-

Mo

de

lD

SL M

od

el

Debug Level :int Compress Data

Journalling Flash File System

Misc. Filesystems

Support ZLIB Default Compression

None SizePriority

conforms to conforms to


ConclusionClone-and-Own with Version Control

1.1 2.01.2 1.3


tiger

1.1 1.2

uses

tigeruses

eagle

uses

0.1


lion

Version control records changes to files overtime so that you can recall specific versions

later. It can do it for any software source code,but also for any type of file on a computer.


Scott Chacon. Ben Straub. Pro Git. Everything You Need to Know about Git. 2nd Edition

Feature Modeling (II)Example from Czarnecki’02

mandatory


car


optional

exclusive choice

inclusive choice

transmission

Hierarchy constraints, for example:manual requires transmission (each child node requires its parent node)

Groups constraints: engine is electric or gas driven or bothNot all constraints in hierarchy & groups, cross-tree constraints in text:


Attributes are added like to classes (eg. engine volume)


Static Variability with Feature ModelsAn Example from The Linux Kernel

1 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES2 if (inode->i_sb->s_magic == ) . . .3 #endif

void pts_sb_from_inode(struct inode * inode)

4 #ifdef CONFIG_UNIX98_PTYS5 pts_sb_from_inode (tty->driver_data);6 #endif

void pty_close(struct tty_struct * tty)

7 tty = kzalloc(sizeof (*tty), GFP_KERNEL);8 pty_close (tty)· · ·


Modeled using class diagrams in PapyrusA model view showing two model hierarchies

PowerLoopDriverAL_ComInterface

ExternalComs

IO_Module

OperationPanel AutroFieldBus

OperationZoneDomain Group

AutroSafePanel

AFB_PowerControl

Installation

AFB_Unit

AlarmZone DetectionZone

[1]

[*]

[1]

[*]

[1]

[*]

[1]

[1]

[1]

[1..*]

[1]

[1]

[1]

[1..32]

[1]

[*]

[1]

[1..*]

[1]

[0..7] [1]

[1..32]

physicalstructure

logical structure



