Cloning Git in Go

Gitgo began as an experiment: how would programming today be different if Git had been written in Go instead of C?

The result: an implementation of Git that is compatible with existing Git repositories, but simpler, more portable, and fast enough to be used as a general-purpose data store. With Gitgo, Git is no longer just for managing source control. Go makes Git a practical choice for content distribution, distributed build systems, establishing consensus, and more.

Transcript

1. Cloning Git in Go
   Aditya Mukerjee
   Risk & Systems Engineer at Stripe
   

   View Slide

2. Why write Gitgo?
   •Git operations from Go applications
   •Portability
   •Simplicity
   •…why not?
   @chimeracoder
   

   View Slide

3. The dirty secret about Git
   •Git wasn’t always meant to be a version control system
   •Git is a decentralized, userland filesystem for preserving historical
   state and synchronizing data
   @chimeracoder
   

   View Slide

4. What We’ve Enabled
   •Backup
   •Content management
   •Distributed build systems
   •Establishing consensus
   @chimeracoder
   

   View Slide

5. Have you ever seen a “friend’s” computer that looked like
   this?
   @chimeracoder
   

   View Slide

6. @chimeracoder
   107724969b2af751a1ffd74e3f38d0d76cb8aa55
   

   View Slide

7. @chimeracoder
   commits
   trees
   blobs
   objects
   

   View Slide

8. Discovery #1: Layered interfaces are more flexible than union types
   @chimeracoder
   commits
   trees
   blobs
   objects
   

   View Slide

9. Bootstrapping Git in Go
   “We can just use the real .git directory for our tests, right?”
   @chimeracoder
   …nope.
   

   View Slide

10. hash(“I love Go”) != hash(“I love Go!”)
    @chimeracoder
    Packfiles
    

    View Slide

11. Reference: Unpacking Git Packfiles
    binarydiff(1, 2, 3)
    1. “I love Go”
    2. “I love Go!”
    3. “You love Go, and so do I!”
    “I love Go” + “!”
    

    View Slide

12. Discovery #2: Pay attention to interface contracts
    @chimeracoder
    zlib data
    zlib data
    

    View Slide

13. @chimeracoder
    type Reader interface {
    Read(p []byte) (n int, err error)
    }
    Object 1 Object 2 Object 3 Object 4 …
    

    View Slide

14. • Discovery #3: Use concurrency to define state machines, and use
    channels for the input/output
    @chimeracoder
    f1
    f3
    f2 g
    g1
    if
    f
    g2
    

    View Slide

15. • Discovery #4: Treat errors as values, and ask how they behave
    @chimeracoder
    httpdir.Open()
    ParsePackfile()
    GitFetch()
    

    View Slide

16. • Discovery #5: Well-architected interfaces perform very well
    @chimeracoder
    

    View Slide

17. #1: Layered interfaces are more flexible than union types
    #2: Pay attention to interface contracts
    #3: Use concurrency to define state machines, and use channels for
    the input/output
    #4: Treat errors as values, and ask how they behave
    #5: Well-architected interfaces perform very well
    @chimeracoder
    

    View Slide

18. Further References
    • “Unpacking Git Packfiles”: https://codewords.recurse.com/issues/three/unpacking-git-packfiles/
    • Gitgo: https://github.com/ChimeraCoder/gitgo/
    • “Git from the Inside Out”: https://codewords.recurse.com/issues/two/git-from-the-inside-out
    

    View Slide

19. Aditya Mukerjee
    @chimeracoder
    https://github.com/ChimeraCoder
    

    View Slide