5 ms·

Is it really “billions”?

• cameronbrown 7y ago
  Yes.
• monocasa 7y ago
  Yeah. I think it's even billions at HEAD too, not some metric about total changes.
  • aantix 7y ago
    Is this strictly code written at Google or dependency loc included as well?
    • skj 7y ago
      The vast majority is not brought in from outside and was written in-house.
• CydeWeys 7y ago
  Don't underestimate how many of these changes are entirely automated. There's a lot of versioned data and configuration stored in the monorepo, not just human-authored code commits.
  • kinkrtyavimoodh 7y ago
    Billions is talking about LOC. As for automated changes, yes there are lots of them but from the pov of managing a VCS they are as 'legit' as manual changes.
    • CydeWeys 7y ago
      A surprising amount of the code is autogenerated, too. Protobuf stubs, @AutoValue classes, compile-time dependency injection, etc.
  • hinkley 7y ago
    The first “big” project I worked on was a monorepo, and it ended up somewhere in the neighborhood of 300k loc/sloc, I can no longer recall which, but over a third of that was generated code. This was a datapoint on a theory I have that projects where all the code is being actively stewarded (instead of abandoned code nobody understands) have less than about 15k (human) lines per developer. If Google’s ratio of generated code is near that, then with 40k developers they’re not far from that ratio.
    • forwhomst 7y ago
      I currently work at a company with a monorepo where they generate artifacts like protocol buffers libraries for Go and python, and check in those artifacts. That's not the way Google likes to do it. At Google they just check in the proto file and build the library artifacts at build time with a rule like https://docs.bazel.build/versions/master/be/protocol-buffer.html#proto_library https://docs.bazel.build/versions/master/be/protocol-buffer....