Modern code review: A case study at Google

Caitlin Sadowski, Emma Söderberg, Luke Church, Michal Sipko, Alberto Bacchelli
[doi] [ISBN] [Google Scholar] [DBLP] [Citeseer] [url]
Read: 23 September 2020

Proceedings of the 40th International Conference on Software Engineering: Software Engineering in Practice
ICSE-SEIP '18
Gothenburg, Sweden
Association for Computing Machinery
New York, NY, USA
Pages 181-190
2018
Note(s): google, usability
Papers: potvin:cacm:2016, sadowski:cacm:2018

Code review goes back to heavyweight practices in the ’70s but this paper looks at modern, lightweight review as practiced at Google since the early days. Modern code review is

  • informal
  • tool-based
  • asynchronous
  • focused on reviewing code changes

Although finding bugs is important, much more important are

  • normative effects: bringing consistency to the codebase
  • educational effects: making sure that more than one person knows the code
  • improving test quality
  • accident prevention
  • gatekeeping to protect other team’s codebases

The paper highlights the importance of the relationship between the author of a change and the reviewer. Several relationships can exist and affect the expectations of the review.

  • Project lead: education, maintaining norms
  • New team members: education, maintaining norms
  • Other team members: education, accident prevention
  • Readability reviewers: maintaining norms
  • Other teams: gatekeeping

Some distinctive features of Google’s approach and tools (compared with OSS, Microsoft and other review processes)

  • the first review comments come fast
  • changes are smaller
  • often just one reviewer (most other processes have two)
  • clearly defined expectations from review
  • the tool recommends reviewer choice based on ownership, review load, availability, etc.
  • integration of static analysis checks (sadowski:cacm:2018)

One thing I did not understand about the paper was why they used snowball sampling for the interviews they performed. Wikipedia says that this is normally used for hidden populations like drug addicts that are hard to identify but it is easy to generate a list of reviewers, change authors, etc.


  • Why Google stores billions of lines of code in a single repository [potvin:cacm:2016]