accelerating se research adoption with analysis bots
play

Accelerating SE research adoption with Analysis Bots - PowerPoint PPT Presentation

Accelerating SE research adoption with Analysis Bots https://github.com/AnalysisBotsPlatform Ivan Beschastnikh, Mircea F. Lungu, Yanyan Zhuang U. of British U. of Groningen U. of Colorado Columbia Colorado Springs Canada Wanted: SE


  1. Accelerating SE research adoption with Analysis Bots https://github.com/AnalysisBotsPlatform Ivan Beschastnikh, Mircea F. Lungu, Yanyan Zhuang U. of British U. of Groningen U. of Colorado Columbia Colorado Springs Canada

  2. 
 Wanted: SE research impact • SE research community developing numerous analysis techniques and tools that use them • But, practitioner adoption is a struggle “many tools are so hard to configure, they prevent you from doing anything.” [1] “I find that the information they provide is not very useful, so I tend to ignore them.” [1] “if it disrupts your flow, you’re not gonna use it.” [1] [1] Brittany Johnson et al. “Why Don’t Software Developers Use Static Analysis Tools to Find Bugs?” ICSE 2013 Ivan Beschastnikh University of British Columbia 2

  3. Each paper creates its own zoo • (Stereotypical) SE research evaluation workflow: • Identify a set of projects/revisions • Run the tool on revisions • (Optional) Submit bug reports to projects e.g., [1] • Write evaluation section [1] Boyuan et al. “Characterizing and Detecting Anti-patterns in the Logging Code” ICSE 2017 Ivan Beschastnikh University of British Columbia 3

  4. petting Each paper creates its own zoo Ivan Beschastnikh University of British Columbia 4

  5. petting Each paper creates its own zoo - Tool runs on project in controlled environment - No involvement of the project community - Tool not used by target user population Ivan Beschastnikh University of British Columbia 5

  6. petting Each paper creates its own zoo End-result: slow tool adoption and low research impact - Tool run in controlled environment - No involvement of the project community - Tool not used by target user population Ivan Beschastnikh University of British Columbia 6

  7. Going into the wild Ivan Beschastnikh University of British Columbia 7

  8. Going into the wild: it’s hard! How will developers find the research tool? • • Requires evangelism/advertising How will developers use the research tool? • • Requires end-to-end working artifact How will they want to • use the tools? • Easy to install • Simple to configure • Immediate value • … Ivan Beschastnikh University of British Columbia 8

  9. Going into the wild: it’s hard! How will developers find the research tool? • • Requires evangelism/advertising How will developers use the research tool? • • Requires end-to-end working artifact How will they want to • use the tools? • Easy to install • Simple to configure • Immediate value • … Ivan Beschastnikh University of British Columbia 9

  10. Rethinking SE research transfer • How do we make it easier (for researchers)? Ivan Beschastnikh University of British Columbia 10

  11. Rethinking SE research transfer • How do we make it easier (for researchers)? …What if developers came to us? Ivan Beschastnikh University of British Columbia 11

  12. Rethinking SE research transfer • How do we make it easier (for researchers)? …What if developers came to us? Idea: Build a one-stop-shop platform for hosting research tools (analysis bots) Ivan Beschastnikh University of British Columbia 12

  13. Analysis bots platform vision Analysis bot := partial automation of some SE task Key features • Low barrier to entry • For bot users/developers • Fine-grained bot control • Developer in control • Openness • Anyone can use/build a bot Ivan Beschastnikh University of British Columbia 13

  14. Analysis bots platform vision Analysis bot := partial automation of some SE task Key features • Low barrier to entry • For bot users/developers • Fine-grained bot control • Developer in control • Openness • Anyone can use/build a bot …like iRobot! Ivan Beschastnikh University of British Columbia 14

  15. How would it work? Twitter Mail Chat Bots Dashboard Developers Bot Runner Bot Upload Bot Ranker Pull requests Project Issues Development Reporter API Bot creators Mediam Versions, Issues, … (you?) Social coding platforms Ivan Beschastnikh University of British Columbia 15

  16. How would it work? Twitter Mail Chat Bots Dashboard Developers Bot Runner Bot Upload Bot Ranker Pull requests Project Issues Development Reporter API Bot creators Mediam Versions, Issues, … Social coding platforms Ivan Beschastnikh University of British Columbia 16

  17. How would it work? Twitter Mail Chat Bots Dashboard Developers Bot Runner Bot Upload Bot Ranker Pull requests Project Issues Development Reporter API Bot creators Mediam Versions, Issues, … Social coding platforms Ivan Beschastnikh University of British Columbia 17

  18. How would it work? Twitter Mail Chat Bots Dashboard Developers Bot Runner Bot Upload Bot Ranker Pull requests Project Issues Development Reporter API Bot creators Mediam Versions, Issues, … Social coding platforms Ivan Beschastnikh University of British Columbia 18

  19. Defining the right APIs is key Twitter Mail Chat Bots Dashboard Developers Bot Runner Bot Upload Bot Ranker Pull requests Project Issues Development Reporter API Bot creators Mediam Versions, Issues, … Social coding platforms Ivan Beschastnikh University of British Columbia 19

  20. Analysis bots benefits • Naturalistic evaluation • Improved replication of academic artifacts • Deals with aspects of tool delivery that tool developers don’t want to deal with • Quick iteration to understand what works and what doesn’t • Developers get to subscribe and use the latest and greatest SE research (Survival of the fittest: most useful tools receive attention and contributions) Ivan Beschastnikh University of British Columbia 20

  21. Open questions • Trust/discovery: how to bootstrap uptake of the platform? • Preventing abuse: what are the appropriate and inappropriate ways in which analysis bots can interact with developers? • Usability: how can analysis bots be designed to prevent information overload for developers? • Economics: How and where should these bots be hosted? • Design: What are the right platform APIs and reputation system? Ivan Beschastnikh University of British Columbia 21

  22. Related work • Google: Tricorder [1] and Shipshape [2] (analysis infrastructure) • Academia: CodeDrones [3] • More academia: Testing-as-a-service [4] and AppInspector [5] • OpenSource: Various bots on GitHub; Imageoptimiser bot [6] • Other ‘bots’: Hubot programmable bot [7] [1] Sadowski et al. Tricorder: Building a Program Analysis Ecosystem. ICSE 2015 [2] https://github.com/google/shipshape [3] Acharya et al. Code Drones. ICSE 2016 [4] Candea et al. Automated Software Testing As a Service. In SoCC, 2010. [5] Gilbert et al. Automated Security Validation of Mobile Apps at App Markets. MCS, 2011 
 [6] https://github.com/imageoptimiser [7] https://hubot.github.com/ Ivan Beschastnikh University of British Columbia 22

  23. Analysis bots vision SE research adoption a consistent challenge; requires concentrated effort Vision: Analysis Bots platform lowers barriers for developers and researchers Bot discovery Centralized repository - + Trust A robust reputation system - + Information overload Common APIs and runtimes - + https://github.com/AnalysisBotsPlatform Ivan Beschastnikh University of British Columbia 23

Recommend


More recommend