COMP61511 (Fall 2017) Software Engineering Concepts In ...syllabus.cs.manchester.ac.uk/pgt/2017/COMP61511/... · We allow mitigation Not properly fixing the bug ... Some routines

1

COMP61511(Fall2017)SoftwareEngineeringConcepts

InPracticeWeek4

BijanParsia&ChristosKotselidis< , @manchester.ac.uk>

(bugreportswelcome!)bijan.parsia christos.kotselidis

2

WhateverWorksPrimitiveTechnology:TiledRoofHut

3.1

Preliminaries

3.2

WhatIsConstruction?Adefinition:

Software construction is the creation, assembly, ormodification of executable programs typically viamodificationofthesourcecode.

3.3

AbstractionHierarchyOfASystem

Nottheonlyformulationofsuchahierarchy!

3.4

ArchitectureVs.Construction

3.5

CodingAsProblemSolvingSoftwareengineeringisproblemsolving

Hence,thefoundationalnatureofproblemdefinitionWritingormodifyingcode

IsalsoaformofproblemsolvingWehopesmallerproblems.

Protip:Alwaysknowtheproblemyou'resolving!

3.6

TheBigFour(PlusTwo)Fourprimaryactivities1. Creating

Weneedfunctionality2. Debugging

Weneedcorrectness3. Refactoring(lastweek!)

Weneedcomprehensibility4. Optimising

Weneedefficiency(wrttosomeresource)Plustwo

Testing&Reading

3.7

TestingIsEverywhereAllprimaryactivitiesinvolvetesting

WhetherformalorinformalE.g.,Creation(whethertestfirstornot)

3.8

ReadingIsEverywhereReadingcodeisakeyskill

Otherpeople'scodethatyouareusingthatyouaremodifying

Yourowncode!whetherusingormodifying

"Reading"(understanding)systemsisakeyskillGraspingtheproblem,requirements,architectureRelatingcodetothose

3.9

10 lines of code = 10 issues.

500 lines of code = "looks fine."

Code reviews.10:58 AM - Nov 5, 2013

106 8,033 4,409

I Am Devloper @iamdevloper

Follow

Reviews

4.1

CreationMindset

4.2

WhatIs(Code)Creation?Code creation (or coding) is the addition of newfunctionality by the generation of new code and units ofcode

Keyactivity!Oftendirectlymeasured

ProductivityasLOC/day(Though,deletingcodemightbebetter!)

DoesnothavetobeexnihiloCut-paste-modifyreusecountsReusecounts!

4.3

PrerequisitesRememberthe !

What'syouroverallproblemdefinitionWhatpartareyoutackling

WhatarethepertinentrequirementsUnderstandthearchitecture

AndhowyourcurrentcodefitsinKnowthelocalstandards

E.g.,codeformattingstyle

prerequisites

4.4

ArchitectureAgoodarchitectureshould:1. helpyoudeterminewhereyourcodeshouldgo2. constrainhowfunctionalityisdivvyedup3. determineyourcommunicationchannels4. giveyouasenseofthingsfittingtogether

thatissharedCode-Architectureconflictsindicate

AproblemwithoneortheotherAlimit

4.5

TechnologyChoicesDifferenttechnologychoicesaffectcode

LanguageMono-vs.poly-glotTypingregime"Batteries"Mindshareetc.

Frameworks,libraries,OSs,etc.

4.6

ToolsDevelopmentenvironments

EditorsIDEsSourcecontrol

DynamictoolsCompilersDebuggers&TestersProfilers

StatictoolsLint,syntax,stylecheckersetc.

4.7

AwarenessesSituationalAwareness

YourperceptionofthecurrentpertinentfactorsfordecisionmakingGoodsituationalawareness

Tracksallpertinentfactorstotherightdegreeinamannertodriveappropraitereactionsatlowcost

DrivestacticsandthusactionUnderstanding

YoursystematicgraspofallfactorsrelatedtodecisionmakingResultsfromsensemakingMorecognitive(indirectlydrivesaction)

4.8

GettingInTheZoneGivenaproblem,oursolvingcanbe

focusedwehavetightsituationalawarenessthe"situation"istheproblemandsolutionspacewereactratherthanact

unfocusedourawarenessisscattered

distracted/multitaskingdisengagedconfused

The"zone"isamuchhigherproductivitystate

4.9

AdminRecord-keepingisextremelyhelpful

Andsometimesrequired,e.g.,billablehoursTrackinghelps!(alotcanbeautomated)

TimeEffort(andsenseofeffort)Whatwasdone(andwhy,bywhome)MoodDiscussionsanddecisions

Someisbetterthannone;enoughisbetterstill;thereistoomuch

4.10

ProgrammerCredosThree Virtues

AccordingtoLarryWall(1),theoriginalauthorofthePerlprogramminglanguage,therearethreegreatvirtuesofaprogrammer;Laziness,ImpatienceandHubris

1. Laziness: The quality that makes you go to great effortto reduce overall energy expenditure. It makes you writelabor-saving programs that other people will find usefuland document what you wrote so you don't have toanswer so many questions about it.

2. Impatience: The anger you feel when the computer isbeing lazy. This makes you write programs that don'tjust react to your needs, but actually anticipate them. Orat least pretend to.

3. Hubris: The quality that makes you write (andmaintain) programs that other people won't want to saybad things about.

(1)Quotedfrom"ProgrammingPerl",2ndEdition,O'Reilly&Associates,1996

5.1

Debugging

—GraceHopper'sBugReport

5.2

DefectsAgainRecall:

A defect in a software system is a quality level (for somequality)thatisnotacceptable.

Wefocuson defectsprimarily

ThoughrobustnessisalsokeyMorestability,i.e.,doesn'tcrash

functionalCorrectness

5.3

WhatIsDebugging?Debugging is the modification of code to remove (ormitigate)correctnessdefects.

Wedon'tcountmissingfunctionalitydefectsDebuggingstartsafterapurporteddetection

Input:aresultoftestingorabugreportWeallowmitigation

NotproperlyfixingthebugButenoughsoit'slessdamagingMuststillinvolvecodemodification

Otherworkaroundsdon'tcount!

5.4

FunctionalLandscape(Enhanced)

5.5

DebugCycleInput:Anindicationofadefect

Stabilise—MakereliabilyrepeatableIsolate(orlocalise)—TothesmallestunitExplain—What'swrongwiththecodeRepair—ReplacethebrokencodeTest—Verifythefix

CheckforRegressionsMaskedbugsNearbybugs

5.6

IndicationAnindicationofadefectisatangiblerecordofabehaviourcontrarytothe(explicitorimplicit)functionalspecificationinadesignatedsituation.

Keyparts:Situation

Preferably,sufficientlydescribedforreplicationExpectedBehaviourWitnessedBehaviour

Typicallywithsomeexplanationwhyit'swrongOftenveryvague

5.7

Indication?Oftenveryvague

Programcrashedsometimeduringthistest

Actually,onlyon !OpenOfficeonUbuntuwon'tprint

Tuesdays

FromJohnRegehr," "ClassicBugReports

5.8

StabiliseBugsareoftenverysituationdependent

Preciseinput+stateOS,hardwareSequenceofactionsLengthofoperating

Astabilsedbugisreliablyrepeatablepreferrablywithminimalsufficientconditions

5.9

Isolate(Localise)Bugsareoftenverylocal

SingleLOCSingleroutineParticularclass

Theydon'thavetobe!Communicationpointsarevulnerable

Adefectisisolatedifyouhaveidentifiedtheminimumsubsystemnecessarytoexhibitthedefectforantriggerinputandsitutation

5.10

Explain&RepairExplainingthebug

YoucanarticulatethemechanismofthebugYourbugtheory

YoucanmanipulatethebugTriggeroravoiditProducevariantsPredictitsbehaviourFixit

RepairingthebugModifyingthecodesothedefectiseliminatedMaynotbepossible!

5.11

TestPostfix

YouneedtoverifyYourtheoryYour*executionofthefix

YouneedtoguardagainstUnintendedconsequences!

"New"bugsariseBugsinthefix

ThefixisincompleteThefixtriggersaregression

Maskedbugs

5.12

CheckNearbyBugscomeinfamilies

SimilarmistakesYoudiditonce,youmighthavedoneittwicePersistentmisunderstandingwithmultiplemanifestations

ClusteredmistakesSomebugshidden

AcrashconcealsmuchSomeroutinesarebroken

Lotsofdebt!Abugisapredictorofmorebugs!

5.13

BugReportsToWONTFIXSometimes,afixisn'tgoingtohappen

ThebugistoosmallOrinsignificantOrambiguous

ThebugistoobigItwouldchangetoomuchbehavior

WhichsomepeoplerelyonOtherdebtincreasestherisk

Thebutistoohard

6.1

Optimising

6.2

ResourcesSize

RunningspaceAtalllevesl

PersistenceandtransmissionCode

TimeResponsevs.thourghput

Instantvs.OverallWall/CPUTime/Instructions

6.3

WhatIsOptimisation?Optmisiationisatransformationofcodeintosufficientlyfunctionally equivalent code that has "better" resourceutilisation.

"Sufficientlyfunctionallyequivalent"Userobservable/desirablebehaviourispreservedUptosomepointItmaybespecialisedtoacertainparticularscenario

ResourceutilisationTypeandPatternmustbespecified

6.4

Where?

6.5

TuningTrade-OffsTimeforSpace(andthereverse)PerformanceforReadability(andthereverse)

AndothercomprehensionqualitiesNotalwaysatradeoffforalgorithmicimprovements

OrfatremovalPerformanceforCorrectnessPerformanceforCost

6.6

TuningAlternativesBuyMoreandFasterHardwareUsetheOptimiserBettercompilers/frameworks/librariesInputmanipulation

"It'sslowwhenIdothis""Don'tdothat!"

6.7

TuningSafetyTuningisrisky

Evenoptimisationcanberisky!It'seasytomakecodefast

BymakingitincorrectIt'seasytomodifythecodealot

AndnotimproveperformancemuchOrmakeworse

6.8

TuningAs(Performance)DebuggingInput:Anindicationofaperformancedefect

Stabilise—MakereliabilyrepeatableIsolate(orlocalise)—Tothesmallestunit

USEAPROFILER!TESTCASESARECRITICALExplain—What'swrongwiththecode

Repair—Replacethe"slow"codeTest—Verifytheimprovements

CheckforSufficiency(Wasthatenough?)Trade-offs(e.g.,spaceconsumption)(Correctness)Bugs

7

ComplexitySlides

8.1

Boehm'sEvidence

FollowingslidesderivedfromMakingSoftware,Chapter10

8.2

ReadingPapersThesepapersarechallenging!

EvenmassagedabitforthepractitionerLotsoftechnicaljargonandtechniquesSummarizingavastliteratureChallengingstatsandpresentations

Don'tpanic!ThesearereadandrereadFirstreadingshouldfocusonkeypointsLaterreadingsshouldfocusontheevidence

8.3

TheRoleOfArchitectureKeychallenge(Boehm,MakingSoftware,Chp10)

Howmuchshouldyouinvestinarchitecture?Analogytobuilding

Wepaythearchitect10%ofthecostofabuildingWeshouldspend10%oftheprojectbudgetonarchitecture

Isthisenough?Howwouldweknow?

Note:statisticallygeneralconclusionsmaynotapplyinyourcase!

8.4

Bohem'sResearchQuestions:"Byhowmuchshouldyouexpectthecostof

makingchangesorfixingdefectstoincreaseasafunctionofprojecttimeorproductsize?"

"Howmuchshouldyouinvestinearlyarchitectingandevidence-basedprojectreviewsbeforeproceedingintoproductdevelopment?"

8.5

EconomiesCommoditymanufacturingexhibitseconomiesofscale

Making1chipmaybemuchmoreexpensivethan1000Theunitcostdiminishesasthenumberofunitsincreases

Softwareend-unitcostsare(canbe)zeroCheaptomakeacopy!

Installation&configurationmaynotbeSofocusonlinesofcodeorbitsoffunctionality

SoftwareexhibitsdiseconomiesofscaleTheunitcostrisesasthenumberofunitsincreases

Potentiallyexponential!Pgs166-167esp.useful

8.6

CostRatiosWhat'stheratioofcosttofixearlyvs.late?

1970s1inrequirementsto≈100postdelivery

19811:100forlargecodebases

But1:5forsmall(2,000-5,000LOC)1996survey

(70-125):12000s

Someevidenceofreductionfrom1:100to1:20Orevenflat(for1millionlinecodebase)

8.7

CostRatios(ForCoursework!)What'stheratioofcosttofixearlyvs.late?

Thinkofyourcoursework!Beforedeployment(akasubmission)

SmallfixesarecheapEsp.inthecurrencyofthecourse,i.e.,points

Afterdeployment(akasubmission)Even"small"fixesareexpensive(orimpossible)

Courseworkbuildsoverthesemester!Soproblemscanbuildup

8.8

TwoStrategiesAvoidlatebugsMakefixinglatebugscheaperFailuretodobothkillstheproject

FailuretodoonemaybemitigatedbytheotherAllouractivitiesshouldaimforthis

Thuswewantarchitecturesthatprecludesomebugsconfinetheeffectsofallbugs

8.9

TwoArchitectureBreakers(Pg376)"20%ofthedefectsaccountfor80%ofthecosts"

"these20%are...duetoinadequatearchitecture..."Twosortsofcosts

DirectcostsOpportunitycosts

TwoexamplebigfailurestheOSarchitecturedidn'tsupportfail-overwhenprocessorsfailed

lackedakeyfunctionalityassumingallmessagesareshort

thusborkingon1millioncharactermessages

8.10

TradeOffsMoreupfrontarch

Costs!Runsriskofoverruns

SincelesstimeforeverythingelsePotentially,gettingarchright

Reducesreworktime

Note,changingrequirementscankillgettingitright

8.11

SweetSpots

8.12

Summary(Pg403)"...thegreatertheproject'ssize,criticality,andstability,the greater the need for validated architecture feasibilityevidence.

"veryverysmalllow-criticalityprojectswithhighvolatility,thearchitectingeffortsmakelittledifference"

Note:Thereareothercostdrivers;checktheassumptions!

9.1

CreationClasses

9.2

MargaretHamilton

NexttotheApolloprojectnavigationsoftwarecode

9.3

AbstractDataTypes(1)Adatatypeis

asetofvalueswithassociatedoperations

Anabstractdatatypeisadatatypecharacterisedentirelybytheoperations

independentofimplementationdetailsAconcretedatatype(orimplementationofanADT)is

arepresentationofasetofvalueswithparticularimplementationsoftheoperations

9.4

ADTExample:Lists

9.5

ADTExample:IntegersIntegers

AbstractValues:All(orfinatesubsetsof)integersOperations:+,-,*,/,<,=,>

Concrete(representation)32bit2s-complementor64bit2s-complementorarraysorfloats(Javascript)or

9.6

ADTExample:DatabasesCRUD

Create-Read-Update-DeleteVeryabstract!

SomeMappings(from ):Operation SQL HTTP DDSCreate / writeRead(Retrieve) read/takeUpdate(Modify) / writeDelete(Destroy) dispose

Wikipedia

INSERT PUT POSTSELECT GETUPDATE PUT PATCHDELETE DELETE

9.7

WhyADTs?Separateinterfaceandimplementation

TheclientonlyneedstothinkabouttheinterfaceUptoapoint!Abstractionsleak

SimplerandclearerFocusesonthesemanticsofthetype

Tellsyouthecriticalfunctionality

9.8

UnitsOfCreationA code unit is a syntactically complete(able) chunk of aprogramthatenactssomebehavior.

LineofCodeBlockorcontrolstructureRoutines/procedures/functions/methods/operationsClasses/modulesPackages/modules

Individualroutinesandclassesarewhatgetunittested