Git Multirepo Ledger
Not a mock-up. Every figure above was read by the extension from a real .git, in a directory
of invented repositories that scripts/make-demo-workspace.ts builds — so you can run it and
get the same board.
Every other repository list tells you where each repository is. This one tells you when it last
moved — and sorts by it.
VS Code's own Repositories row already carries the branch and the ahead/behind figures, and so does
GitLens's. Neither carries the last commit: not on the row, not in the tooltip, not in the children
beneath it. And neither will sort by it — VS Code offers discovery order, name or path; GitLens
offers discovered, last-fetched or name. "Which of these moved, and when" is the question that
starts a working day, and nothing in the editor answers it.
The second half is where the repositories are. git.repositoryScanMaxDepth defaults to 1, and
git.scanRepositories refuses an absolute path outright, so a repository two levels down — or in a
directory you have not opened — is invisible to both. This walks for .git at any depth, beneath
the open folders and beneath directories you name.
That is the whole of what this does. It is a status board, not a git client: no staging, no
committing, no pushing, no merging, no rebasing, no conflict resolution. Nothing it runs can
change your working tree, your branches, or any commit you have made.
One thing it runs is not a read: git fetch, offered per row and for the whole board, and
disabled until you turn it on. It writes remote-tracking refs, FETCH_HEAD and objects — and
nothing else, which is why it is the only write here. It exists because ahead and behind are
measured against remote-tracking refs: those are exactly as old as your last fetch, so without it a
row can read in sync having asked its server nothing for a month. It never merges, pushes or
prunes. When a repository wants credentials the fetch fails immediately rather than waiting on a
prompt that cannot be answered, and you are shown git's own words and the exact command.
The row
One repository per row, three lines, nothing to expand — so twenty repositories are one pass of the
eye rather than twenty clicks.
- The repository name, how far it has diverged from its upstream, and — dimmed at the end —
how old that divergence figure is.
- The last commit: when, then its subject. This is the line the extension exists for, and the
list is sorted by it.
- What HEAD is doing —
main, or detached at 7c86ebf, or rebasing 1/3 onto main — and a
marker when the repository is not an ordinary one: worktree, submodule, bare, shallow.
Two things the row deliberately does not do
It never renders an unestablished figure as zero. ↑2 ↓1 is computed against a remote-tracking
ref that is exactly as old as the last fetch, so the row says how old that is and stays silent when
.git/FETCH_HEAD cannot prove even an attempt. A branch with no upstream reads no upstream, one
whose upstream is gone reads gone, and a working tree that has not been read yet is not the same
as a clean one. Absence of measurement never looks like a measured zero.
It has no branch count. No tool surveyed carries one, and a number you would not act on is not
a field.
A repository that will not answer — ownership git does not trust, a corrupt index, no commits yet,
a directory that vanished mid-scan — gets a row that says so, why, and the exact command that
failed, so you can retype it yourself.
The history pane
Below the list, filling in for whichever row is selected: that repository's recent commits — hash,
age, author, subject, and chips for the branches and tags that point at each one. One git log per
page, not one per commit.
Commits that exist on no remote this repository knows about are marked only here, and only when
that was actually established: the check costs a second process and runs only when the row already
said there was something ahead, so an unmarked commit means "asked, and no", never "did not ask".
Click a commit and it expands into the files it changed. Click a file and it opens in the
editor's own diff, against its parent — for a repository the editor has never opened. That needs
this extension to serve the blobs itself, because VS Code's own git: URIs resolve only for
repositories the built-in Git extension has already opened, which excludes every repository this
board exists to show.
A merge says it is a merge and names its parents rather than showing a diff that would be true
against one side and misleading against the other.
Settings
| Setting |
Default |
What it does |
multirepoLedger.additionalRoots |
[] |
Absolute directory paths scanned for repositories in addition to the open folders. This is the normal way to use the extension: the directory your repositories live in is usually not the one you have open. |
multirepoLedger.exclude |
[] |
Absolute paths of repositories to leave out entirely — a mirror, a vendored checkout, anything you keep but never work in. |
multirepoLedger.maxDepth |
32 |
How deep below each root the search for .git descends. A stop against a symlink cycle or a home directory, not a way to make the scan cheaper. |
multirepoLedger.dirty.enabled |
true |
Show which repositories hold uncommitted work. The one read that walks the working tree, so it costs a second git process per repository on screen. While off, the row says nothing there rather than showing a zero. |
multirepoLedger.concurrency |
0 |
How many repositories are read at once. 0 derives it from what the machine says it can run in parallel. |
multirepoLedger.history.pageSize |
50 |
Commits per page in the history pane. |
multirepoLedger.badge |
unpushed |
What the number on the Activity Bar icon counts. Every option is also a chip on the board, so one click shows exactly what the badge counted. A count of none draws no badge rather than a 0. |
multirepoLedger.forge.enabled |
false |
Open merge and pull request counts, via the gh and glab CLIs. Off by default: it reaches the network. While off, neither tool is invoked. The GitHub half is exercised against a real gh; the GitLab half is not yet, and fails to silence rather than to a guess. |
multirepoLedger.fetch.enabled |
false |
Offer a Fetch button on each row and in the view title. Off by default: it reaches the network. It runs git fetch and nothing else — no merge, no push, no prune — so it can update what the remote knows without touching your working tree, your branches or any commit you have made. |
multirepoLedger.fetch.concurrency |
4 |
How many repositories are fetched at once. Unlike the read concurrency this cannot be derived from your machine: the limit belongs to a server the extension cannot see. |
Privacy
Everything except multirepoLedger.forge.enabled is a local read of your own repositories. With the
forge setting on, gh and glab are run as subprocesses and talk to whatever hosts your
repositories point at, using credentials those tools already hold; no token is ever asked for or
stored, and there is no telemetry and no account. With it off, nothing leaves the machine.
Licence
MIT — see LICENSE.
