Downloaded 25 times
Uploaded byMarcel Bruch
2010 06-24 karlsruher entwicklertag
This document discusses Eclipse code recommenders, which leverage collective intelligence in software development. Eclipse code recommenders observe how frameworks are commonly used and extended to provide intelligent code completion, usage-driven Javadocs, code example recommendations, and smart bug detection. They work by mining development data to extract patterns on how frameworks are reused and then surface those recommendations to developers. The system aims to help developers by recommending what others frequently do based on past usage. It analyzes collected data on framework usage to build recommender models and provide suggestions to developers.
2010 06-24 karlsruher entwicklertag
- 1. Eclipse Code Recommenders “IDE2.0” Collective Intelligence in Software Development Leveraging the Wisdom of the Crowds Marcel Bruch
- 3. Eclipse Code Recommenders All Frameworks Code Code Eclipse Code Recommenders – do what most people do! (where appropriate) Developers Who Called “new Text()” Method Also Called setText() addListener() setLayout() setFont() 89% 53% 99% 35% What Do Developers Ultimately Override After Extending This Class? 100 % overwrote PreferencePage.createContents() 96 % overwrote PreferencePage.performOk() 5 % overwrote PreferencePage.performCancel()
- 4. “Other developers frequentlybought used the following methods…” INTELLIGENT CODE COMPLETION
- 5. Code Completion… What doesthe developer need, i.e., which methods should the code completion present to the user?
- 6. All 164 Methodsof Text ?
- 7.
- 8. Code Completion… What doesthe developer need, i.e., which methods should the code completion present to the user?
- 9. Intelligent Code Completion … or just the three missing ones?
- 10.
- 11. Why just onewhen you can have templates…
- 12. How it works– in a nutshell… Framework If you extend A, you should call B.c Framework Framework Recommender … Model Instantiation 1 Instantiation n extract facts about how the framework is reused <extends:A> <extends:A> <overrides:A.b> <overrides:A.d> … <calls:B.c> <calls:B.c> … … Mining
- 13. How it works– in a nutshell…
- 14. How it works– in a nutshell… Observation Recommendations
- 15. “What do DevelopersUltimately buy override after extending this class?” USAGE-DRIVEN JAVADOCS
- 16. Some Facts AboutThe Documentation of Overridable Methods overridable 2074 overridden 623
- 17. What Clients DoWith Your API… overridden 623 documented* 155 478 *: documented as overridable by using phrases like “clients may/should/must override/extend this method”
- 18. What documentation wouldyou expect when subclassing a Dialog?
- 19. What Javadoc givesto you… A dialog is a specialized window used for narrow-focused communication with the user. Dialogs are usually modal. Consequently, it is generally bad practice to open a dialog without a parent. A modal dialog without a parent is not prevented from disappearing behind the application's other But what are the windows, making it very confusing for the user. If there is more than one modal dialog is open the second one should be parented offDialog? hot-spots of of the shell of the first one otherwise it is possible that the OS will give focus to the first dialog potentially blocking the UI.
- 20. What Code Completionoffers… Which of the 56 methods should we override?
- 21.
- 22.
- 23. This way, orthat way? Subclassing Patterns for ViewerSorter
- 24. How it works? Clustering in a nutshell http://upload.wikimedia.org/wikipedia/en/2/29/Bernoulli-Mixture-Model.gif
- 25. “The good onesin the potty, the bad ones in… the problem view.” SMART BUG DETECTION – DRIVEN BY REAL USAGES
- 26. What’s wrong withthis code?
- 27. At runtime yourerror log shows… An error has occurred. See error log for more details. org.eclipse.core.runtime.AssertionFailedException null argument:
- 28. At runtime yourerror log shows… So what have we missed?
- 29. If all ofyour colleagues do it – why don’t you?
- 30. How it works– a conceptual view That’s “strange”
- 31. How it works– a conceptual view
- 32. Just one formula… x= {observed method calls on a variable} E(x) = {exact similar usages for variables of type x in the code base} A(x) = {almost similar usages, i.e., x+1 method call} | E ( x) | strangeness( x) 1 | E ( x) | | A( x) |
- 33. Find your bugs…during development!
- 34. Find your bugs…during development!
- 35. “Why is GoogleCodesearch not ‘google for code search’? ” CODE SEARCH ENGINES – LIFT OFF!
- 36. Navigating the APIjungle… How do I get an IStatusLineManager?
- 37. When asking GoogleCodesearch No IDE integration Manual query creation Text-based retrieval Ignores prior knowledge
- 38. Code Example Recommender Query
- 39.
- 40.
- 41. “Why is GoogleCodesearch not ‘google for code search’? ” LEVERAGING USER FEEDBACK TO IMPROVE CODE SEARCH ENGINES
- 42. Leveraging User Feedback ... ... ... ... 42
- 43. Refine the rankingbased on feedback
- 44. Leveraging User Feedback 1 2 Support 2 Vector 1 Machine 3 3 44
- 45. Stackoverflow.com – Reloaded! WHATCAUSED THAT STACKTRACE?
- 46. So far wehave… Intelligent Code Completion Usage-Driven Javadocs Code Example Recommender Smart Bug Dectection
- 47. But what ifyou get this?
- 48. Debugging – theold way
- 49. Debugging – theold way
- 50. But more oftenwe get…
- 51. How can wefix that? Don’t stacktraces share some commonalities? ExceptionType Feature (evaluates identical exception type) Message Similarity Feature (counts of similar words) org.eclipse.swt.SWTException: Invalid thread access at org.eclipse.swt.SWT.error(SWT.java:3884) at org.eclipse.swt.SWT.error(SWT.java:3799) at org.eclipse.swt.SWT.error(SWT.java:3770) at org.eclipse.swt.widgets.Widget.error(Widget.java:463) at org.eclipse.swt.widgets.Widget.checkWidget(Widget.java:355) at org.eclipse.swt.widgets.ProgressBar.setSelection(ProgressBar.java:317) at org.eclipselabs.cr.stacktraces.demo.ThreadDemo$1.run(Unknown Source) at java.lang.Thread.run(Thread.java:619) TraceDiff Similarity Feature (counts similar stackframes per trace)
- 52. Having a centralexception repository
- 53.
- 54. “IDE 2.0? Leveragingthe wisdom of the (Eclipse) community…” LAST.FM 4 ECLIPSE
- 55. How It WorksToday Framework If you extend A, you should call B.c Framework Framework Recommender … Model Instantiation 1 Instantiation n extract facts about how the framework is reused <extends:A> <extends:A> <overrides:A.b> <overrides:A.d> … <calls:B.c> <calls:B.c> … … Mining
- 56. How it WorksTomorrow Framework If you extend A, you should call B.c Framework Framework Recommender … Model Instantiation 1 Instantiation n extract facts about how the framework is reused <extends:A> <extends:A> <overrides:A.b> <overrides:A.d> … <calls:B.c> <calls:B.c> … … Mining
- 57. Leveraging your IDE’sbuild environment framework context project indication environment object usage selected proposals
- 58. “I’ve seen muchbut why do you call it IDE 2.0 ?” WRAP UP
- 59. From IDE 1.0to IDE 2.0
- 60. From Web 2.0to IDE 2.0 Let’s map the Web 2.0 principles to IDE 2.0 to see how this analogy fits: 1. The Web as Platform… 2. Data as “Intel inside”… 3. Harnessing Collective Intelligence… 4. Rich User Experiences… 5. Lightweight Programming Models…
- 61. Milestones … 01.09.2010 01.12.2010 21.03.2009 Moving to Stacktrace v0.1 Release Eclipse.org Recommender 01.07.2010 v0.1 Example Code 01.10.2010 v0.1 Bugs 01.02.2011 Recommender v0.1 Last.FM Recommender
- 62. Resources • Contact • Marcel Bruch bruch@cs.tu-darmstadt.de • • Project homepage • http://www.stg.tu-darmstadt.de/research/core/ • Download Release “LISBON” (V.0.2.105) • http://www.stg.tu-darmstadt.de/research/core/download • Source Code • http://eclipselabs.org/p/code-recommenders/
- 63. Credits – innumerablehours have spent… Peter Schroeder Daniel Glöckner Johannes Lerch Maik Görtz Mohsen Parisay Jan Kassens Johannes Born Sebastian Kasten Daniel Staesche Sebastian Ahlfeld David Kalnischkies Sebastian Wörner Boyan Yurukov Jan Stolzenburg Florian Jakob Nico Wombacher Dirk Kröhan Christopher Mann Julius Rückert Steffen Remus Sebastian Proksch Sinem Emeröz Veronika Kostadinova Nikolay Shindov Tomasz Kalbarczyk Michael Novotny Jens Krause Michael Kutschke Gary Fritz Kristijan Madunic Sheip Dargutev David Schuld Sebastian Denel Johannes Kastl Roman Getto Daniel Brandtner Christian Kilb Florian Nöll Laura Altmüller Dennis Sänger Tjark Vandommele Dennis Siebert Markus Migenda Minh Hoang Nguyen Peter Sinzig Paul Schatygin Jan-Michael Heller Sascha Nordquist
- 64.
- 65. Summary Intelligent Code Completion Usage-Driven Javadocs Code Example Recommender Smart Bug Dectection
- 66. Affiliation Prof. Dr. MiraMezini Software Technology Group Fachbereich Informatik Technische Universität Darmstadt