- author = "Mark Harman",
- title = "Scaling Genetic Improvement and Automated Program Repair",
- booktitle = "International Workshop on Automated Program Repair (APR'22)",
- year = "2022",
- editor = "Maria Kechagia and Shin Hwei Tan and Sergey Mechtaev and Lin Tan",
- address = "Internet",
- month = "19 " # may,
- publisher = "ACM",
- note = "Invited keynote",
- keywords = "genetic algorithms, genetic programming, genetic improvement, GI, Automated Program Repair, APR, Search Based Software Engineering, SBSE",
- isbn13 = "978-1-4503-9285-3/22/05",
-
URL = "
https://research.facebook.com/publications/scaling-genetic-improvement-and-automated-program-repair/",
-
URL = "
https://research.facebook.com/file/987372605236188/Scaling-Genetic-Improvement-and-Automated-Program-Repair.pdf",
-
DOI = "
doi:10.1145/3524459.3527353",
-
video_url = "https://youtu.be/uAU_ySdEvks",
- size = "7 pages",
- abstract = "techniques and research directions for scaling genetic improvement and automated program repair, highlighting possible directions for future work and open challenges",
-
notes = "Video Intro/Chair: Lin Tan, Purdue University,
USA
Scalability. Build effort repair, 20 minutes too long. New feature build cost problem. Meta = Facebook, 2900000000 users. 100000 commits per week. Cloud infrastructure for both development and emulator testing. Search based testing (SBST) at scale. Sapienz (testing) SSBSE 2018 https://engineering.fb.com/2018/05/02/developer-tools/sapienz-intelligent-automated-software-testing-at-scale/ Repair. Yue Jia, Ke Mao, Nadia Alshahwan.
Alexandru Marginean SapFix \cite{Marginean:2019:ICSE} 12 week PhD \cite{Marginean_10137954_thesis_redacted} internship at Facebook. Immediate impact (12 weeks v 12-15 years (uni). Impact on 3 billion users.
Sapienz and Infer Peter O'Hearn. Fixing null pointer exceptions. Developers modifying some automatically generated Sap-Fix bug fixes.
Developer cannot deal with 100000000 lines of code, automated systems can.
New feature build cost. Interaction between new features: feature interaction problem. Automated mock => genetic improvement solution. Automated acceptance testing. Digital twin, dove-dove cyber-cyber digital twin, hierarchy, precision v speed tradeoff.
WES \cite{Ahlgren:2020:GI} detection of problem versus making it harder to cause problem. Software technology is different from physical systems. Bot emulate real users in Facebook system. Restrictions on using social graph like road traffic speed humps. Avoids rebuilding system between tests of new parameters. Dynamic co-adaptation coevolution between system bad software robots and good bots, both emulating users. GI 43 minutes: then Q+A Q: Westley Weimer
48 Q: Sergey Mechtaev
52: Q Lin Tan deployment in industry? A: No fundamental block to use in industry. Need better system level test. Unit tests often give good coverage.
56: Q Lin Tan progress. A: Fault localisation (sometimes easy), developers want to take ownership of (automatically generated) code change.
999900a001.pdf http://program-repair.org/workshop-2022/ APR@ICSE Part of \cite{Kechagia:2022:APR}",
