git update-server-info

What dumb-HTTP servers need run after each push.

🎯 What & why

git update-server-info rewrites the static metadata files that a dumb-HTTP server needs to advertise refs and packs. Run it on the remote after every push when the server has no Git CGI of its own.

🧠 Mental model

Dumb HTTP is just Apache serving files. Without server-side Git, the client cannot ask 'what refs do you have?' -- so we pre-bake the answer into info/refs and objects/info/packs and let the static server hand those files out.

🛠️ Synopsis

git update-server-info [-f | --force]

Regenerates `info/refs` (one line per ref: SHA TAB refname) and `objects/info/packs` (one line per packfile). Runs in the current repo (typically a bare repo behind the web root).

🎚️ Switches & options

Flag	What it does
`-f, --force`	Rewrite the info files even when they look up to date.
`(implicit) info/refs`	Generated under `<gitdir>/info/refs` for ref discovery.
`(implicit) objects/info/packs`	Generated alongside packfiles so clients can enumerate them.
`(implicit) post-update hook`	The shipped sample hook in `hooks/post-update.sample` is just `exec git update-server-info`.

💡 Use cases

Bare repo served by plain Apache/nginx with no git http-backend CGI.
Mirror or backup repo synced via rsync that is also published over HTTP.
USB-stick or filesystem repo handed to a viewer that only speaks dumb HTTP.
Recovering after info/refs has been deleted or gone stale.

🧪 Examples

Refresh after a push

git update-server-info

Force a rewrite

git update-server-info -f

Enable the shipped hook

mv hooks/post-update.sample hooks/post-update

Publish a bare mirror

git clone --bare repo.git && cd repo.git && git update-server-info

🎓 Recommendations

Use the smart protocol (git http-backend CGI) instead -- it is faster and needs no server-info dance.
If you must stay dumb, enable the post-update hook so refreshes happen automatically.
Serve from a bare repo; running this in a working tree just to publish is a smell.

🪤 Common pitfalls

Forgetting to run it after a push leaves clients seeing the OLD refs -- silent staleness.
Dumb HTTP cannot push without git http-push, which is itself almost never wanted.
info/refs is plaintext; anyone who can read it sees every ref name in the repo.

🔗 Related modules

📝 Quiz

Hit each option, then Check answers. Score is recorded; Next is always open.