Git
Unusual repositories
What Pip shows for repositories off the usual path — reftable, shallow and partial clones, very large ones, worktrees and submodules — and why.
Ref and object formats#
- Reftable
- A repository that keeps its refs in reftable (
extensions.refStorage = reftable) leaves only placeholder files where Pip reads, so it says nothing rather than guess. The Git tab shows “This repository uses a ref format Pip cannot read.” with the folder; the row shows no branch, arrows or operation, and the tray menu has no git line. Repositories marks it Refs Pip can't read, and of the git verbs in Quick actions only Show Git tab and Open folder remain. git itself reads reftable, so Check with git in Repositories still answers. - SHA-256
- Supported everywhere a SHA-1 repository is. The ids are simply longer, and copying a detached HEAD copies all 64 characters.
- Borrowed objects
- A clone made with
git clone --referencekeeps its commit graph in the repository it borrows from, which Pip does not read. Until the clone writes one of its own (git commit-graph write --reachable), a branch that differs from its upstream readsnot synced, and the Branches review marks every branch Unknown: no commit graph and says why above the list.
Shallow and partial clones#
- Shallow clone
- git ignores the commit-graph in a shallow clone, so Pip cannot count ahead and behind there: a branch that differs from its upstream says
not syncedon the row and “Differs from origin/main, count unknown” on the Git tab. For the same reason the falling-behind notification never fires, and the Branches review marks branches Unknown: shallow clone. History stops where the clone began, and says so. - Partial clone
- The files Pip reads — refs, the index, the commit-graph — are complete in a partial clone, so rows and the Git tab behave as usual. History asks git not to download missing objects. Its log counts the lines every commit changed, so in a clone that left file contents on the server History usually stops at git's own error — “git couldn't read the history: …” — and a single diff at “git couldn't show this diff: …”, rather than downloading anything. git 2.44 and later honour that request; an older git may still fetch the missing objects from the remote, never with a prompt, using credentials you already have. It is the one case where something other than Fetch reaches the network.
Very large repositories#
The scan reads every repository every few seconds, so its readings have ceilings. Past one, Pip says it did not count rather than show a number it never took; the readings you open on demand reach further.
- Ahead and behind
- Counted from git's commit-graph only while the graph is under 16 MiB in all — about a quarter of a million commits — and the two tips are within 10,000 commits of each other. Past either limit the row says
not syncedand the Git tab says “Differs from origin/main, count unknown”. A graph that has not caught up with a fetch says the same in any repository — Fetch and sync explains. - Conflicts
- The scan counts conflicted files only while the index file is under 16 MiB; above that the Stats tab says conflicts not counted. The Git tab still lists them when it opens, from an index of up to 64 MiB, and marks the card Read when the tab opened.
- Working copy
- Above 250,000 tracked files the card says it is too large to compare, with the count, instead of trying. A comparison that runs past two seconds stops and says how far it got, and the search for untracked files stops after 50,000 entries; either count then reads at least.
- Check with git
- git gets 10 seconds. A tree it cannot finish in that time gets “Too big for git to check in 10 seconds.”
- Branches review
- It reads the first 4,096 branches and says when there are more. A comparison that would run past its limits reads Unknown: history too deep or Unknown: graph too large.
Worktrees, bare repositories and submodules#
- Linked worktrees
- Each worktree is its own row in Repositories, and its HEAD card says Linked worktree. When a repository has more than one, the Git tab adds a Worktrees card marking this one, the main worktree, and any that are locked or have their folder missing. They share one repository, so one fetch runs at a time across them, a fetch from any of them shows on all, and one mute covers them all.
- Bare repositories
- Pip finds a repository by its
.gitentry, which a bare repository does not have, so a server whose folder is a bare repository shows no git. Worktrees made from one read normally, and the Worktrees card lists them with no main worktree — though their HEAD card says Branch rather than Linked worktree unless the bare repository's own folder is called.git. - Submodules
- The innermost
.gitwins, so a server running inside a submodule shows the submodule's own branch, and its HEAD card says Submodule. Fetch never goes into submodules, Check with git counts one only when its recorded commit has moved, and the Repository card counts them and says how many are not initialised.
Ownership and safe.directory#
git refuses to work in a repository owned by another user unless your safe.directory setting allows it, and Pip lets git make that check rather than step around it. Its file readings run no git, so rows, the tray, the Git tab's cards, Repositories and notifications are unaffected.
The three things that run git show git's refusal as git worded it: “Could not fetch: detected dubious ownership in repository at '…'” on the Remote card, and the same line after “git couldn't check this working tree:” or “git couldn't read the history:”. If you trust the folder, add it with git config --global --add safe.directory followed by its path, as git itself suggests. Pip never changes your git config.
Unborn branches and unusual upstreams#
- No commits yet
- A branch with no commits — a fresh
git init, or an orphan branch — shows its name on the row with nothing after it. The HEAD card and Repositories say No commits yet, there is no History button, and in place of a pull request the Remote card says “Make a first commit to open a pull request.” - A local upstream
- A branch that tracks another local branch (
remote = .) reads Tracks local branch develop on the Remote card and is counted against it as usual. A fetch never moves it, and neither the upstream-deleted nor the falling-behind notification watches it. The row still saysupstream goneif you delete the branch it tracks; the tray menu does not. - An upstream URL
- A branch whose remote is a URL rather than a named remote reads Tracks a branch fetched by URL: and the URL. git keeps no remote-tracking branch for one, so there is nothing to compare, and the row shows the branch alone.
Did this page not answer it? Tell us what you were looking for — that is how Pip's documentation gets written.