📊
git stats

A tool for getting aggregated commit stats.

git-stats parses log information to get stats about the files changed, additions, and deletions. For example:

$ git stats -r origin..HEAD
Author           Commits  Changed Files  Insertions  Deletions  Net Δ
Luke Hsiao            67            117       +2616      -1126  +1490
dependabot[bot]       31             62        +203       -267    -64
Total                 98            179       +2819      -1393  +1426

Reviewer/Tester  Commits
Luke Hsiao             1

Install

Git Stats reads your repository directly with gitoxide, a pure-Rust implementation of git.

From crates.io

cargo install git-stats --locked

Since you are compiling from source anyway, you can opt into CPU-specific optimizations that the prebuilt binaries cannot use:

RUSTFLAGS="-C target-cpu=native" cargo install git-stats --locked

Or, if you use cargo-binstall:

cargo binstall git-stats

From the AUR

On Arch Linux, install from the AUR using your preferred helper (e.g. paru or yay):

paru -S git-stats       # builds from source
paru -S git-stats-bin   # prebuilt binary

The two packages conflict with each other, so only one may be installed at a time.

With mise

mise can install a prebuilt binary directly from the GitHub release, verifying its checksum and SLSA provenance:

mise use -g github:lukehsiao/git-stats

Usage

A tool for getting aggregated commit stats

Usage: git-stats [OPTIONS] [revision-range]

Arguments:
  [revision-range]  Show only commits in the specified revision range [default: HEAD, or ::@ with jj]

Options:
  -e, --email            Show the email address of each author
  -r, --reviews          Show who reviewed/tested commits based on `Acked-by`, `Tested-by`, and `Reviewed-by` git trailers
  -s, --sort <SORT>      What column to sort the stats table by. Names sort by byte value, so uppercase comes before lowercase. The reviewers table always sorts by commits [default: commits] [possible values: author, commits, files, insertions, deletions, net]
      --reverse          Reverse the sort order of the stats table: counts ascending, authors Z to A
  -a, --author <AUTHOR>  Limit the commits output to ones with author header lines that match the specified pattern (regular expression)
      --since <SINCE>    Limit the commits output to ones more recent than a specific date
      --until <UNTIL>    Limit the commits output to ones older than a specific date
      --vcs <VCS>        Which version control system reads the repository and interprets [revision-range] [default: auto] [possible values: auto, git, jj]
  -h, --help             Print help (see more with '--help')
  -V, --version          Print version

Jujutsu

In a jj workspace, colocated or not, git stats takes a revset instead of a git revision range. Change IDs, bookmarks, and your revset-aliases all work:

$ git stats 'trunk()..@'

• Default range. Without a range, it reports ::@, the working-copy commit and its ancestors. Like any jj command, it snapshots the working copy first, so uncommitted edits count. An empty working-copy commit counts as one commit with no changes; use ::@- to leave it out.
• Requirements. jj 0.36 or newer must be on your PATH. jj selects the commits, and gitoxide reads them from jj's git store.
• Conflicts. A conflicted commit, and a commit with a conflicted parent, take their counts from jj's own diff, as jj diff --stat shows them. jj stores a conflict as a git tree of .jjconflict-* directories, which a git diff would count as files.
• Forcing git. --vcs git reads a colocated workspace as a git repository with git revision ranges, for scripts that pass ranges like origin..HEAD. git stats also uses git when GIT_DIR is set, as it is inside git hooks.

Notes

A few behaviors are worth knowing, mostly where reading git natively differs slightly from the porcelain:

• Revision ranges. Single revisions, A..B, and A...B are supported, and each endpoint can be spelled any way gitoxide understands (refs, short hashes, @{n}, and so on). More exotic gitrevisions(7) forms are not interpreted.
• --since / --until. Accepts ISO 8601, RFC 2822, unix timestamps, and relative dates like "2 weeks ago". This is a subset of git's full approxidate, and filtering is on the committer date.
• --author. Matches against the mailmap-resolved author (Name <email>), whereas git log --author matches the raw author header. These differ only for repositories that carry a .mailmap.
• Empty names and emails. An author or reviewer with an empty name shows as their email, and each email gets its own row. Without --email, git shortlog instead groups them all under one blank name; with it, git shows <email>. With --email, a contributor without an email shows as just the name, where git shortlog -e shows Name <>.
• Merge commits count as commits but contribute no line or file changes, matching the default of git log --numstat.
• Shallow clones work: commits at the shallow boundary diff against the empty tree, exactly as git log --numstat shows them. When the range reaches the boundary, a warning on stderr notes that the truncated history makes the stats differ from a full clone's.
• Submodules count like git log --numstat: adding or repointing one is a single changed file whose pointer line is the diff, and a file swapped with a submodule is one changed file too. Renaming one without repointing it is one changed file with no line changes, as git pairs the rename.
• Partial clones (git clone --filter=blob:none) fail with an error that says the clone lacks objects, because gitoxide does not fetch missing objects the way git does. Running git log --numstat over the same range first fetches them.
• Line counts follow diff.algorithm as git does, but gitoxide's diff implementations can pick a different diff than git's on some files, most visibly on large rewrites, so insertions and deletions can differ from git log --numstat. gitoxide has no patience algorithm and uses histogram instead. File and net counts are unaffected.
• SHA-256 repositories (git init --object-format=sha256) are not supported, as gitoxide cannot read them yet.
• Commit encodings other than UTF-8 are not converted: the encoding header is ignored and non-UTF-8 author names render with replacement characters, where git log would re-encode them. Grouping uses the converted text, so names that differ only in bytes that are not valid UTF-8 can share a row.