Object database

Git's on-disk storage format is relatively simple.

SHA-1s

Values are just sequences of bytes, and they're indexed by the SHA-1 hash of their content, generated by git hash-object.

Object storage

git init creates the .git directory:

• objects/
  • NN/ -- the first two characters of a SHA-1 hash
    • nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn -- the remaining 38 characters of the SHA-1 hash

git cat-file retrieves stored objects; -t gets type, -p pretty-prints content.

Commits are objects

git commit creates and stores an object, and its SHA-1 hash is what determines the commit's ref. The file contains:

• Header:
  • tree nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn, containing the SHA-1 hash of a tree object.
  • parent nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn references the parent commit. This field is omitted from root commits.
  • author Name <email> epoch +HHMM
  • committer Name <email> epoch +HHMM
• Commit message

Trees are objects referencing other objects

The tree object is a series of lines comprised of the following fields:

• mode (e.g. 10644)
• type (blob, tree)
• object-id containing the SHA-1 hash of the object.
• name contains the filename.

git count-objects provides a summary of the number of objects and consumed disk space.

Tags are objects referencing commit objects

Git has two types of tags:

• Lightweight tags don't have labels and point directly to a commit.
• Annotated tags have labels, which are stored in tag objects.

Annotated tags are stored in the object database with the following fields:

• Header:
  • object nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn
  • type commit
  • tag name
  • tagger Name <email> epoch +HHMM
• Message

Both types of tags are represented as files in the .git/refs/tags directory, where the value of the file is the SHA-1 hash of an object. Lightweight tags contain a reference to their HEAD commit, as there's no additional data that needs to be stored. Annotated tags instead contain a reference to their tag objects, which stores the label.

In short, tags are just labels pointing at commits -- exactly the same as a branch. The difference is in semantics: tags should be considered immutable, whereas branches can change.

Branches

At their core, branches just references to their current commit. The references are stored in .git/refs/heads, indexed by their branch names.

The current branch is stored in the .git/HEAD file:

ref: refs/heads/master

Remote branches can be found in a couple of locations within .git:

• refs/:
  • remotes:
    • origin:
      • HEAD
• packed-refs

$ git show-ref
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn refs/heads/x
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn refs/remotes/x