Git under the hood (minimal)

Git is a content-addressable, immutable, distributed database optimized for tracking filesystem snapshots.

Core ideas:

• Everything is identified by a cryptographic hash (SHA-1 or SHA-256)
• Data is immutable: new content ⇒ new hash ⇒ new object
• Commits store complete tree snapshots, not diffs
• History forms a Merkle DAG of commits

Main object types:

• blob → raw file content
• tree → directory (maps filenames to blob/tree hashes)
• commit → metadata + pointer to a tree + parent commit(s)
• tag → named reference to an object (usually a commit)

Storage layout:

• Objects live under .git/objects/<2-char>/<38-char>
• Loose objects are zlib-compressed individually
• Packfiles group and delta-compress objects for efficiency

Hashes and integrity:

• Object id = hash(content)
• Commit includes hash(tree) and hash(parent)
• Chain of hashes = tamper-evident Merkle DAG

References:

• .git/refs/heads/<branch> → latest commit hash
• .git/refs/tags/<tag> → tagged commit
• HEAD → current branch ref (symbolic)
• Detached HEAD → points directly to a commit

Index (staging area):

• .git/index maps paths → blob hashes + metadata
• Bridge between working directory and next commit
• Enables three-way diff: working dir, index, HEAD

Graph model:

• Commits form a DAG: node = commit, edge = parent
• Merge = commit with multiple parents
• Rebase = rewrite DAG by creating new commits

Remotes:

• Remote = peer repository (not a master)
• Fetch/push sync missing objects by comparing hashes
• Transfer is delta-efficient and stateless

Architectural patterns:

• Content-addressable storage → object immutability
• Composite → trees containing blobs/subtrees
• Merkle DAG → commit integrity and verification
• Symbolic references → HEAD and branches
• Staging buffer → index as write cache
• Eventual consistency → decentralized sync

Mental model: Git = immutable key-value store + DAG of snapshots + symbolic refs.