todrivemanygamesandnavigationapps,butthesesensorsarenotexposedattheGUI - PDF document

todrivemanygamesandnavigationapps,butthesesensorsarenotexposedattheGUIlevel.OnesuchappisGoogleMapsStreetView,whichusesthecompasssensortodetectphoneorientation(withrespecttomagneticNorth)inordertoorienttheon-screenmapcorrectly.Finally,theaccuracyofeventtimingduringbothcaptureandreplayiscrucial.Forexample,aswipethatiscapturedtooslowlywillreplayasasequenceofbuttonpresses,andthetimingbetweenexternaleventsanduser-interfaceactionscanbesimilarlycrucial.Indeed,wefoundthatdelaysasshortasmillisecondscanadverselyimpactthereplayprocess.InthispaperwepresentRERAN(RE cordandR eplayforAN droid),arecord-and-replaytoolthataddressesthesechallengesinthecontextofAndroidapplications.RERANdirectlycapturesandreplaysthelow-leveleventsthataretriggeredonthedevice.ThisapproachallowsRERANtocaptureandplaybackGUIevents,i.e.,touchscreengestures(e.g.,tap,swipe,pinch,zoom),aswellasthosefromothersensorinputdevices(e.g.,accelerometer,lightsensor,com-pass).Incontrast,priorreplaytoolsforAndroid[2],[6],[7],whicharebasedonthekeywordactionparadigm,onlyreplaytouchscreenpressesandcannothandletouchscreengesturesorothersensorinputs.RERANworksinanatural,1non-intrusive,andlow-effortway:duringrecord,acapturecomponentreadstheevents,withnonoticeableimpactonexecution.Thiscontrastswithmanykeyword-action-basedapproaches,whichrequiremanualscriptingoftestcasesratherthandirecttracecollection[3]–[5].Duringreplay,aphone-basedreplayagentfeedstheeventstothephone,andthephoneactsasitwouldifthesensorinputcamefromuserorenvironmentinteraction.RERANiscarefullydesignedtomanagethelargenumberoflow-leveleventsthatoccurduringexecutionandtorecordandreplaythemwithprecisetimings.RERANiscurrentlyunabletoreplaysensorswhoseeventsaremadeavailabletoapplicationsthroughsystemservicesratherthanthroughthelow-leveleventinterface(e.g.,cameraandGPS).RERANalsodoesnotcaptureothersourcesofnondeterminisminapplications,forexampleuseofrandom-numbergeneratorsandaccesstothelesystem.Providingfullydeterministicreplaywouldrequireasignicantlymoreheavyweightsolution,forexample,involvingmodicationstothevirtualmachineand/orcompleteinterpositionbetweentheappandtheunderlyingplatform.Despitetheselimitations,wedemonstratethatRERANisusefulinseveralways.First,wewereabletorecordandsuccessfullyreplayexecutiontracesfor86outoftheTop-100freeAndroidappsonGooglePlay(themainAndroidappdistributionchannel).Moreover,RERANreplaysinnearlyrealtime,withanoverheadofjustover1%comparedtotheoriginalexecution.Second,RERANisapowerfuldebuggingaidbecauseitcancaptureandreplayeventsequencesthatleadtocrashesin1RERANrequiresthedevicetobe“rooted”inordertoexecutethereplayagent,whichwebelieveisareasonableassumptionfortheintendedusersofRERAN,researchersanddevelopers. Fig.1.Distributionofgesturesinthe86appsreplayablewithRERAN.apps,whichhelpsdevelopersinspectsystemstatearoundacrashpoint.Inparticular,wedemonstratehow,usingRERAN,wewereabletoreproduceandreplaybugsinwidely-usedappsincludingFirefox,Facebook,K-9Mail,andQuickofce.Third,RERANhassupportfortimewarpingwhichallowstheexecutionofcertainappstobefast-forwardedinordertoreachaspecicappstatefasterthanintheoriginalexecution,andwithoutanyadditionalmanualinput.Wefoundthatthereplaytimefortime-warpableappscanbereducedby22to68percentcomparedtotheoriginalexecution.Theremainderofthispaperisorganizedasfollows.Sec-tionIIintroducestheAndroidinputeventsandsensorclasses,aswellaseventgeneration,delivery,andprocessing.Sec-tionIIIprovidesanoverviewofourapproachandthechal-lengesthataccuraterecord-and-replayposefortouchscreen-basedsmartphones.InSectionIVwediscusstheexperimentalmethodologywefollowed,theimplementationofRERAN,resultsofrunningitonTop-100GooglePlayapps,andperformanceoverhead.SectionVpresentsthreeapplicationsofusingRERANforresearchanddevelopmenttasks:repeata-bility,reproducingbugs,andtimewarping.InSectionVIwediscusslimitationsofourapproachandrecord-and-replayonAndroidingeneral.SectionVIIdiscussesrelatedwork,andinSectionVIIIwepresentourconclusions.II.ANDROIDINPUTSInthissectionwedescribethediversekindsofinputsthatAndroidappscanaccept,aswellashowtheseinputsarerepresentedinternallyasevents.A.TouchscreenUserActionsSmartphoneappsaredistinguishedinpartbytheexpressivevarietyofuseractionspossiblethroughtouchscreens.Webrieydescribethemostcommonones:Press-and-Release:Theprimaryinputactionisthesim-plepress,inwhichtheusertapsanareaonthescreenandreleasesquickly,e.g.,clickingabuttonortypingtextwiththeon-screenkeyboard.Press-and-Hold:Thepressandholdconsistsoftheusertappinganareaonthescreenandreleasingafterholding,inthesameposition,pastsomethresholdamountoftime.Thisactioncanbeusedtoaccesssecondarymenusorhiddenoptions. Fig.3.TimeseriesofeventsinGasBuddy. Fig.2.Seriesofeventsrepresentingaswipe;Xi,Yiaretheinitialstartingcoordinates,Xm,Ymarethecoordinatesduringmoving(stillholdingdown),andXf,Yfarethenalcoordinates.orientation.Inthefourthphase,theuserbrowsesgaspricesforanindividualstation(seconds46–60),whichrequiresscrollingandnocompass.Finally(seconds61+)theusernavigatesagainthroughthestationsmapusingpinchandzoom,whichgeneratesbothtouchscreenandcompassevents.Thetimeseriesrevealsseveralkeyrequirementsforsuccess-fulreplay:eventsfromallsensors,notonlythetouchscreen,mustbecapturedanddelivered,astheyinuenceappbehavior;andeventsmustbecapturedanddeliveredathighthroughput.WeprovideabroaderquantitativeassessmentoftypicaleventvolumeinTableI.Thetableshowsthetotalnumberofeventsreceivedfromthreeinputdevices,capturedduringa5-minuterunoftheapplisted.Touchscreeneventsdominate,TABLEINUMBEROFINPUTEVENTSPERINPUTDEVICEDURINGA5-MINUTERUN. Appname Touchscreen Proximitysensor Accelerometer Facebook 6,199 701 6 AngryBirds 5,895 675 3 GasBuddy 4,723 273 15 Amazon 4,831 779 4 howevertherearealsoasignicantnumberofeventsfromsensors.Forallappsthephonewaskeptatonatableduringexecution,whichaccountsforthesmallnumberofreportedaccelerometerevents.III.APPROACHOVERVIEWWenowpresentanoverviewofourapproach,itstechnicalchallenges,anditsadvantagesoverpriorrecord-and-replayap-proaches.GoogleprovidestheAndroidSoftwareDevelopmentKit(SDK),whichcontainsvarioustoolstoaidappdevelopersinimplementation,testing,anddebugging.SeveralofthesetoolswereusedinthedevelopmentofRERAN.Thehigh-levelviewispresentedinFigure4.Recordorreplaycanbestartedfromeitherthephone,usingtheAndroidTerminalEmulator[9]appavailableforfreeonGooglePlay,orfromacomputerconnectedtothephoneviaaUSBcable.TheAndroidDebugBridge(ADB),accessiblethroughtheadbcommandintheAndroidSDK,andAndroidTerminalEmulatoractastheinterfacebetweenRERANandthesmart-phone.Commandscanbeissuedtothesmartphone'sOSviatheseshells.Weusedthisabilitybothtorecordeventsandtoreplaythem.A.RecordWeusetheAndroidSDK'sgeteventtool,whichreadsthe/dev/input/event*lesinordertoprovidealivelog(trace)oftheinputeventsonthephone,asshownonthetopofFigure4.Theperformanceoftheappisunaffectedduringlogging.Oncelogginghasended,RERAN'srecordphasetranslatesthecapturedtraceandcreatestheeventtracelethatisreadytobereplayed.Duringtranslation,allthe TABLEIITHE86OUTOFTOP-100APPSINTHEU.S.VIEWOFGOOGLEPLAY(ASOFMAY6,2012)THATRERANCANREPLAY. Appname AdobeAIR CraigslistMobile FeedYourDinoFree IMDB RunningFred VisualAnatomy AdobeFlashPlayer CrimeCity GasBuddy Instagram SincerelyInkCards WeatherBug AdobeReader DailyHoroscope GoogleDrive Kindle SkyMap WeatherChannel AmazonMobile DanceLegend GoogleEarth Maps Skype WhatsAppMessenger AngryBirdsRio DeathRallyFree GoogleMapsStreetView MotoPrint SlackerRadio Where'sMyWater AngryBirdsSeasons Dictionary GooglePlayBooks MoviesbyFlixster SlotCityMachines WordSearch AngryBirdsSpace DocumentsToGo GooglePlayMusic MP3RingtoneMaker StickManBMXStunts YellowPages BackgroundHDWallpapers DragRacing GooglePlayMovies Myxer TalkingTom2 Yelp BaseballSuperstars2012 DragRacing-BikeRacing GooglePlus Netix TED YouTube BBCNews DrawSomething GoogleSearch NBAGametime Textgram Zedge Bible EasyBatterySaver GoogleTranslate OneTouchDrawing TouchNotePostCards Zinio BMXBoy eBay Groupon Picsart Tunewiki BrightestFlashlight ESPNScoreCenter HeartRadio PoolMasterPro Twitter BubbleShoot Evernote HowtoReadThoughts PulseNews UnblockMeFree ColorNote Facebook iFunny RecipeSearch UnicornDash TABLEIIITHEPERFORMANCEANDSPACEOVERHEADOFRERANON4POPULARAPPSAVERAGEDOVER5USERSESSIONS. Appname Run-timeoverhead Tracesize Original Replay Overhead (seconds) (seconds) (%) (KB) Facebook 291.04 294.12 1.05 260.43 AngryBirds 296.81 300.05 1.08 359.76 Dictionary 262.83 264.11 0.48 117.85 GasBuddy 254.14 256.81 1.04 244.62 thephoneviaadb.However,asdescribedearlier,thatsolutiondidnotachievethemicrosecondaccuracyneededforgestures.RERANsupportsselectivereplay,wherebytheusercansuppressentireclassesofevents.Thiscanbeespeciallyhelpfulwhentheuseristryingtoisolatetherootcauseofabug.Itcanalsoreduceoverheadwhenreplayinganappthatneverusesaparticularsensorinput,suchastheaccelerometer.Thecompiledreplayagentisuploadedtothephoneviaadbandtakesatracele(eventsetandtheirtiming)asargument.Inordertoexecuteonthephone,systempermissionsmustbechanged,whichrequiresthephonetoberooted.Thereplayagentrunsinaseparateprocessfromthereplayedapp.D.AchievementsOftheTop-100appsonGooglePlay,wewereabletosuccessfullyreplay86;welistthenamesofthese86appsinTableII.NotethattheseappsspanavarietyofcategoriesfromSocialMediatoProductivitytoGames,demonstratingthewidescopeofRERAN'spracticalutility.InSectionVIwediscussthereasonswhyreplayisnotpossiblefortheremaining14appsinTop-100.E.TimeandSpaceOverheadWemeasuredtimeandspaceoverheadsfor4sampleapps,usingtracesfrom5differentusers.WepresenttheresultsinTableIII.First,wecomparedthecompletiontimeoforiginalexecutionswiththecompletiontimeofreplayedexecutions.Columns2and3showtheaverageruntime,acrossthe5executions,oftheoriginalrunandofthereplayedrun;column4showsthetimeoverhead,near1%inallcases.Webelievethereplayoverheadislargelyduetothefactthateventinjectionisnotinstantaneous.Duringcapture,sometimesaninputdevicereportseventsasoccuringsimultaneously,i.e.,withthesametimestamp.Wewereabletomimicthisbywritingmultipleeventsintothestreamatonce.However,becausetheeventsarebeinginjectedprogramatically,ratherthanbeinggeneratedbythephysicaldevice,capturedsimultaneouseventsareonlyreplayedwithin500microsecondsofeachother.Despitethelag,itcausednonoticeabledifferencesinappbehavior.Thelastcolumnshowstheaveragelogsizeoftheeventscaptured.Asexpected,forthemoreinteractiveappsthatusemanygestures,e.g.,AngryBirds,thelogsizeishigherthanlessinteractiveapps,e.g.,Dictionary.AlltestswereconductedonthephysicaldevicesdescribedinSectionIV-A.V.REPLAYUSESTheabilitytoreplayarecordedexecutionhasmyriadapplications.InthissectionweillustrateseveralscenariosofhowRERAN'srecord-and-replaycapabilitiescanbebenecialtodevelopersandresearchers.A.RepeatabilitythroughTraceReplayRERANisabletoplaybackasetofinputeventsofaspecicsessionexactlyinthesamewayasitwasoriginallyrecorded.Thisincludestouchscreenevents,userproximityevents,aswellaschangesinphoneorientationsensedbytheaccelerometerandcompass.Incontrast,ifauserweretomanuallyrepeatasessiontryingtoreplicatethesameactionsasthepreviousrun,therecouldbeinconsistenciesandabnormalitieswhencomparedtotheoriginalrunduetohumanerrorandinconsistenttiming;thiscouldinturnleadtounfaithfulresultsoranundesiredoutcome.Replaycangreatlyhelpdevelopersandresearcherswhowishtoeliminateinaccurateandtime-consumingapptestingwhenrepetitionofasequenceisneeded.Infact,inourpriorresearch[10],weusedapreliminaryversionofRERANtoeliminatetheneedforconstantinteractionwithtesters:toprolenetworktrafcandotherappcharacteristics,weexamined10functionallyidenticalexecutionsofthesameappbyrecordingauser'sinteraction(execution#1)andreplayingit9times(executions#2through#10). virtualorphysicalkeypadandclickingbuttons.Theseeventsoccuronhumantimescalesandthushavesignicantroomforspeed-ups.Processingcontentreferstotheuserexaminingcontentonthescreen,anddecidingwhichactiontoperformnext,e.g.,pressabutton.ProcessingcontentcanalsotakeplaceafteraccessingdataonexternalserversviatheInternet(e.g.,theCNNandFacebookapps).Thesesituationscanbefast-forwardedbecausethereisalagbetweenthetimethatthecontenthasloadedandthetimetheuserperformsthenextinputaction.Wefast-forwardusingthreerulesthatwehavederivedempiricallyforourphonesandtestapps,butweexpectthegeneralprincipletoholdforothersetups.Therulesareasfollows:(a)iftheTDislessthan0.7seconds,thenwecompressthedelayto0.001seconds;(b)ifthedelayisgreaterthan3.0seconds,wecompressthedelayto3.0seconds,and(c)anydelaybetween0.7and3.0secondsisunchanged.Rule(a)triestoidentifyandfast-forwarddataentryfromtheuser.Thisrule,forexample,allowsdataentrytobegreatlyspedupinappsrequiringheavytextinputsuchasDictionaryandGoogleTranslate.Rule(b)accountsforthelongTDsduringexecutionthatcanoccurwhenauserisviewingcontentonthescreenordecidingwheretonavigatetonext.Forexample,thisoptimizationcanbeseeninthegasstationnderGasBuddy,inwhichtheuserviewsamapofgaspricesbeforeselectingone,creatinglongperiodsofdelay.Rule(c)preservesthespeedoftheexecutioninallothersituations.Despitethesimplicityoftheserules,wewereabletousethemsuccessfullytotime-warpmanyapps.VideodemosofRERANfast-forwardingareavailableonYouTube[1].Wepresenttheresultsoffast-forwardinginTableV.Foreachapp,incolumn2wepresenttheoriginalrun-time,incolumn3wepresenttherun-timewhenfast-forwarding,whilecolumn4showstherun-timereductionasapercentage.Notethat,ingeneral,appscanbespedupanywherefrom4%to68%.Certainapps,however(Quickofce,NPRNews,HomeSwitcher),couldnotbefast-forwardedduetoreasonsthatwewillexplainnext.UsesofFast-forwarding:Fast-forwardingcanshortenthetimetoreachacertainpointintheappwithoutrequiringmanualinput.Forexample,whiletestingafeatureaddedtooneoftheapp'sscreens,adevelopercouldusefast-forwardingtogreatlyreducethetimerequiredtogettothatscreen.Fast-forwardingcanalsobeusedincombinationwithRERAN'sabilitytoreproducebugsinordertospeedupdebuggingtime.Byfast-forwardingthereplay,theamountoftimetoreproducethecrash,i.e.,replaythestepstoreproduce,willbereduced,thussavingthedevelopertime.AsshowninTableV,wealsofast-forwardedthe9appsthathadreproduciblebugcrashes.Ineach,weranourtime-warpedreplayandfoundthat5appshadreducedrunningtime.Theremaining3appssawnoreductionandkepttheirpreviousrunningtime.Thisisduetotheappsnothavingdataentryorlongdelays(sowecouldnotapplyTABLEVAPPRUNTIME,INTHEORIGINALEXECUTIONANDREPLAYEDUSINGFAST-FORWARDING.BUGSREPRODUCEDARESHOWNINGRAY. Appname Run-time Reduction Original Fast-forwarded (seconds) (seconds) (%) Facebook1.9 249.71 162.93 34.75 Dictionary 238.49 144.31 39.49 GasBuddy 263.06 84.72 67.79 BBCNews 277.31 120.32 56.61 Craigslist 308.92 239.54 22.46 ESPNScore 314.22 244.95 22.05 Amazon 298.69 109.15 63.46 Movies 286.16 197.09 31.13 GoogleTranslate 308.66 211.63 31.44 Ankidroid 8.66 8.35 3.58 APVPDFviewer 10.93 8.17 25.25 Firefox 64.22 29.55 53.99 Soundcloud 29.34 25.79 12.10 K-9Mail 32.23 24.84 22.93 Facebook1.7.1 30.31 18.69 38.34 Quickofce 3.12 3.12 0.00 NPRNews 9.91 9.91 0.00 HomeSwitcher 7.99 7.99 0.00 TABLEVITHE14OUTOFTOP-100APPSTHATRERANCANNOTREPLAY. Non-replayablereason Appname RequiresAndroidsensorservice BarcodeScanner,TangoVideoCalls (e.g.,camera,microphone) VoiceSearch,VoxerWalkie-Talkie Dynamicorrandomelements Pandora,BoxingGame,FruitNinja (i.e.,nondeterminism) JewelsStar,ShootBubbleDeluxe Solitare,TempleRun,Tetris UnoFree,WordswithFriends rules(a)or(b)),andinsteadtheTDsstayedintherangeofourdo-not-alterrule(rule(c)),whichkepttheoriginalTD.VI.LIMITATIONSAsmentionedinSectionIV-D,RERANcouldreplay86oftheTop100-appsonGooglePlay.However,therewere14appswecouldnotreplay.Theseappsfellintotwomaincategories:appswhichrequiredanAndroidsensorserviceandappsthatcontaindynamicorrandomelements,asshowninTableVI.Wenowdiscussthemainreasonswhytheseappscannotbereplayed.RequiresAndroidsensorservice:Someappsrequiresen-sorinputthatRERANdoesnotrecord.Forexample,wecannotreplaytheBarcodeScannerappproperlywithoutcapturingtheidenticalbarcodeimagethatisfedasinputfromthecamera.Similarily,wecannotreplayVoxerWalkie-Talkiebecausetheapprequirescapturingaudiodatafromthemicrophone.InAndroid,thesetypesofinputcomefromsensorsthatarenotinstrumentedinthesamewayasthetouchscreenandaccelerometer(thatuse/dev/input/event*).Instead,duetoprivacyandsecurityissues,thesesensorsarenotopenlyaccessibleforcapturingandareprotectedbysystempermissions,e.g.,thedevice'scamera,microphone,andGPSlocation.Currently,wedonotcapturesuchsensors,astappingintosuchsourceswouldrequireanon-trivialextensiontooursystem,ataskweleavetofuturework. GUITAR[27]isaGUItestingframeworkforJavaandWindowsapplications.GUITARgenerateseventsequence-basedtestcasesforGUIapplications.Theirtechniquecangeneratetestcasesautomaticallyusingastructuraleventgenerationgraph.WhileGUITARprimarilytargetsJavadesk-topapplications,ithasrecentlybeenportedtoAndroidbyextendingAndroidSDK'sMonkeyrunnertooltoallowuserstocreatetheirowntestcaseswithapoint-and-clickinterfacethatcapturespressevents.However,GUITARdoesnotsupporttouchscreengestures,e.g.,swipeandzoom,orotherinputde-vices,e.g.,accelerometerandcompass.Inthepast,navigatingthroughaprogrambypointingandclickingwassufcientanduseful—outsideofthemobileworld,theprimaryinputdevicescapturedhavebeenmousemovementandkeyboardkeystrokes.However,assmartphonesbecomemoreadvanced,offeringgreaterfeaturesandinteractivitythroughaddeddevicesensorsandtheiruses,wearguethatinordertofaithfullyreplayuserinteractionallthesesensorsmustbeaccountedfor,notonlyasubsetofonedevice,i.e.,touchscreenpresses.C.DeterministicReplayAsubstantialbodyofworkhasfocusedondeterministicreplaybycapturingandreplayingeventsatthehardware[28],operatingsystem[29],orvirtualmachine[30]levels.Theseapproacheslogeventsthatmightintroducenon-determinism(I/O,threadscheduling,memoryaccesses,etc.)anddelivertheeventsintherightorderduringreplaytoensurethatthereplayedexecutionis(quasi)identicaltotheoriginalrun.Loggingoverheadvariesdependingonthenatureofthebenchmark,andaddingrecordandreplaycapabilitiestoanexistingsystemmightbequiteintrusive.Weexploreadifferentpointinthedesignspace:wedonotaimtoachievedeterministicreplay,butinsteadfocusonaneffectiveapproachandtoolthatusesstandardsoftware,hardware,andisminimallyintrusive,yetneverthelesscanreplay86outoftheTop-100popularapps.VIII.CONCLUSIONSWehavepresentedRERAN,anapproachtorecord-and-replayfortheAndroidplatform.Ourresearchwasmoti-vatedbythenoveltyandpopularityoftouchscreen-basedplatformsandapps,theuniquechallengesassociatedwithreplayingtheseapps,andthebroadapplicabilityofrecord-and-replaytechniqueswhentacklingresearchandpracticaltasksonAndroid.Bydirectlycapturingthelow-leveleventstreamonthephoneandreplayingitwithprecisetiming,RERANcaneasilyreproducecomplexGUIgesturesaswellasothersensorinputs.Theresultisanoninvasiveyetveryeffectiverecord-and-replayapproachthatworksforthevastmajorityofpopular,real-worldappsinGooglePlay'sTop-100.Moreover,wehavedemonstratedthattheapproachcanbeappliedsuccessfullyforrepeatability,bugreproducibility,andexecutiontime-warping.ACKNOWLEDGMENTSWethankShashankKothapalliforhisassistanceincollect-ingusertraces.ThisworkwassupportedinpartbyNationalScienceFoundationawardsCNS-1064646,CNS-1064844,andCNS-1143627,andbyaGoogleResearchAward.REFERENCES[1]“RERAN:RecordandReplayforAndroidVideoDemo,”August2012,http://www.youtube.com/user/RERAN2012.[2]SourceForge,“AndroidGUITAR,”August2012,http://sourceforge.net/apps/mediawiki/guitar/index.php?title=Android GUITAR.[3]TimothyWall,“AbbotframeworkforautomatedtestingofJavaGUIcomponentsandprograms,”August2012,http://abbot.sourceforge.net.[4]Hewlet-PackardCompany,“HPFunctionalTesting,”August2012.[5]IBM,“RationalRobot,”August2012,www.ibm.com/software/awdtools/tester/robot/.[6]D.Amaltano,A.R.Fasolino,andP.Tramontana,“Aguicrawling-basedtechniqueforandroidmobileapplicationtesting,”inICSTW'11,pp.252–261.[7]Nodobo,“NodoboCapture:MobileDataRecordingforAnalysingUserInteractionsinContext,”2011.[8]AndroidDevelopers,“SensorEvent,”August2012,http://developer.android.com/reference/android/hardware/SensorEvent.html#values.[9]GooglePlayAndroidapps,“AndroidTerminalEmulator,”August2012,https://play.google.com/store/apps/details?id=jackpal.androidterm.[10]X.Wei,L.Gomez,I.Neamtiu,andM.Faloutsos,“Proledroid:multi-layerprolingofandroidapplications,”inMobiCom'12,pp.137–148.[11]M.Jovic,A.Adamoli,D.Zaparanuks,andM.Hauswirth,“Automatingperformancetestingofinteractivejavaapplications,”inAST'10,pp.8–15.[12]J.FlinnandZ.M.Mao,“Candeterministicreplaybeanenablingtoolformobilecomputing?”inHotMobile'11,pp.84–89.[13]J.Mickens,J.Elson,andJ.Howell,“Mugshot:deterministiccaptureandreplayforjavascriptapplications,”inNSDI'10,pp.159–174.[14]DanielHerdingandChristianSpannagel,“Jacareto,”August2012,http://sourceforge.net/apps/mediawiki/jacareto/.[15]MatthewPekar,“Pounder-JavaGUITestingUtility,”August2012,http://pounder.sourceforge.net/.[16]DakshinamurthyKarra,“Marathon,”August2012,http://www.marathontesting.com/.[17]P.A.BrooksandA.M.Memon,“Automatedguitestingguidedbyusageproles,”inASE'07,pp.333–342.[18]A.C.R.Paiva,J.a.C.P.Faria,N.Tillmann,andR.A.M.Vidal,“Amodel-to-implementationmappingtoolforautomatedmodel-basedguitesting,”inProceedingsofthe7thinternationalconferenceonFormalMethodsandSoftwareEngineering,pp.450–464.[19]J.Steven,P.Chandra,B.Fleck,andA.Podgurski,“jrapture:Acap-ture/replaytoolforobservation-basedtesting,”inISSTA'00,pp.158–167.[20]L.B.KaraandT.F.Stahovich,“Animage-based,trainablesymbolrecognizerforhand-drawnsketches,”inComputers&Graphics,vol.29,no.4,pp.501-517,2005.[21]AndroidDevelopers,“Layouts,”August2012,http://developer.android.com/guide/topics/ui/declaring-layout.html.[22]S.H.Khandkar,S.M.Sohan,J.Sillito,andF.Maurer,“ToolSupportforTestingComplexMulti-TouchGestures,”inACMInternationalConferenceonInteractiveTabletopsandSurfaces,pp.59–68.[23]S.H.KhandkarandF.Maurer,“ADomainSpecicLanguagetoDeneGesturesforMulti-TouchApplications,”inProceedingsofthe10thWorkshoponDomain-SpecicModeling,pp.2:1–2:6.[24]AndroidDevelopers,“UI/ApplicationExerciserMonkey,”August2012,http://developer.android.com/tools/help/monkey.html.[25]——,“MonkeyRunner,”August2012,http://developer.android.com/guide/developing/tools/monkeyrunner concepts.html.[26]GoogleCode,“Robotium,”August2012,http://code.google.com/p/robotium/.[27]AtifMemon,“GUITAR,”August2012,guitar.sourceforge.net/.[28]S.Narayanasamy,G.Pokam,andB.Calder,“Bugnet:Continuouslyrecordingprogramexecutionfordeterministicreplaydebugging,”inISCA'05,pp.284–295.[29]S.M.Srinivasan,S.Kandula,C.R.Andrews,andY.Zhou,“Flashback:alightweightextensionforrollbackanddeterministicreplayforsoftwaredebugging,”inUSENIXAnnualTechnicalConference,2004,pp.29–44.[30]G.W.Dunlap,S.T.King,S.Cinar,M.A.Basrai,andP.M.Chen,“Revirt:enablingintrusionanalysisthroughvirtual-machineloggingandreplay,”inOSDI'02,pp.211–224.