Skip to content
Tutorials Center
My learning

Topics

Git Reference ยท Objects, Refs and Internals

Commit objects

Reference for the Git commit object: tree, parent, author and committer headers, optional encoding and gpgsig headers, the message, and how to build one by hand.

A commit object ties a snapshot to its history. It stores the ID of one tree (the snapshot), the IDs of zero or more parent commits, who wrote and who committed the change with a time stamp for each, and the message. Nothing else: no diff, no branch name, no list of changed files. Everything you see in git log or git show is derived by comparing trees.

A commit is plain text. Because its ID is computed from that text, changing any part of it (the message, a date, the parent, or the tree) gives a different ID. That is the reason rewriting a commit rewrites all of its descendants.

Terminal
git cat-file -p HEAD        # show the raw commit
git cat-file commit HEAD    # same, without pretty-printing
git commit-tree <tree> -p <parent> -m "message"

Try it

Bash Read a commit as raw text

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with read or cat), it gets the next line from this box.

Bash Build a merge commit by hand with two parents

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with read or cat), it gets the next line from this box.

Format

The object body is a block of headers, one blank line, and the message:

TEXT
tree <tree-id>
parent <parent-id>            (zero or more lines)
author Name <email> <unix-time> <zone>
committer Name <email> <unix-time> <zone>
encoding <name>               (optional)
gpgsig <signature>            (optional, continued lines start with a space)
mergetag <embedded tag>       (optional)

<message>
Header Description
tree The ID of the root tree. Exactly one, always first.
parent One line per parent. None for a root commit, two or more for a merge. The order matters: the first parent is the branch you merged into.
author Who wrote the change, plus the time and zone when it was first written.
committer Who created this commit object, with its own time. It differs from the author after cherry-pick, rebase, am or commit --amend -C.
encoding Only present when the message is not UTF-8 (set by i18n.commitEncoding).
gpgsig, gpgsig-sha256 A signature over the commit; see Signed commits and tags. Continuation lines begin with a single space.
mergetag A copy of a signed tag that was merged with git merge, kept as evidence.

The time is seconds since the Unix epoch in decimal, followed by a zone offset such as +0200 or -0800. Names and addresses are free text; the address is wrapped in angle brackets. The message is everything after the first blank line, stored exactly as written, normally ending with a newline.

Usage notes

  • A commit is not a diff. Parents let Git compute one on demand: git show compares the commit's tree with its first parent's tree.
  • Author versus committer. Set GIT_AUTHOR_* and GIT_COMMITTER_* separately (see Author and committer variables) to control each; git log --format=%an,%cn prints both names.
  • Parent order. HEAD^1 and HEAD^2 follow the parent lines in order. Octopus merges list three or more.
  • Unknown headers are kept. Git ignores headers it does not know, which allows tools to add their own; git fsck still checks the standard ones.
  • Notes live elsewhere. Data added later with git notes is stored on a separate ref and does not change the commit's ID.
  • Grafts and replacements (git replace) change what Git shows, not the stored object.
  • First line convention. The first line is the subject, shown by --oneline. The blank line after it is a convention of tools, not a rule of the format.