Git Internals: Objects, Refs, and the DAG Under the Hood

git commit feels like saving a file. Under the hood, Git stores snapshots as hashed objects in .git/objects/ a Merkle DAG where branches are just movable pointers.

Four object types

Everything keyed by SHA-1 (or SHA-256 in newer repos):

blob a1b2c3... → "console.log('hi')"
tree d4e5f6... → { "README.md": blob_a1b2, "src/": tree_789 }
commit 789abc... → { tree: d4e5f6, parent: 456def, msg: "init" }

Refs are labels

• refs/heads/main → latest commit on main
• refs/remotes/origin/main → last fetched remote tip
• HEAD → usually ref: refs/heads/main

Branches aren't folders. They're 40-character filenames containing one hash.

# Peek under the hood
git cat-file -p HEAD
git ls-tree HEAD
git rev-parse main

The DAG

Commits form a directed acyclic graph. Merge commits have two parents. Rebase rewires parent pointers same patches, new commit hashes.

    A---B---C  main
         \
          D---E  feature

git merge feature creates commit F with parents C and E.

Staging area (index)

The index is a tree snapshot of your next commit between working directory and repository. git add updates index entries; git commit freezes index into a tree object.

Why this matters

• Reflog orphaned commits survive until GC
• Cherry-pick applies patch, new commit hash
• Shallow clone truncates history, not objects
• Partial clone lazy-fetches missing blobs

[!TIP] git gc packs loose objects. Nothing is truly deleted until unreachable and garbage-collected.

Takeaway

Git is a content-addressed snapshot store with cheap branching. Files are blobs. Directories are trees. History is a DAG. Everything else rebase, merge, stash is pointer manipulation on that graph.