Skip to content

Commit

Permalink
docs: Add a Jujutsu from first principles doc.
Browse files Browse the repository at this point in the history
This document is WIP. 

This came up recently in Discord as a explanation was requested without referencing Git, which
all our other documents do.
  • Loading branch information
PhilipMetzger committed May 5, 2024
1 parent af95c78 commit 0e209d4
Show file tree
Hide file tree
Showing 2 changed files with 62 additions and 0 deletions.
60 changes: 60 additions & 0 deletions docs/jj_first_principles.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,60 @@
# Jujutsu from first principles (without Git)

## Preface

Why does Jujutsu exist and which problems does it solve? This document tries to
answer both of these questions while expanding on the design in a user-friendly
way.

At its core Jujutsu is [Version Control System][vcs] which scales to huge
repositories at [Google scale][billion-lines]. Many design choices are
influenced by the concurrent commits happening in Googles Monorepo, as there
are always multiple people working on the same file(s) at the same time.

## Core Tenets

Jujutsu's core tenets are:

* User-friendliness: Making the working copy a commit is simpler. This is
how the project started.
* The "repository", so the commit graph is the source of truth. The working
copy is just one way of editing commits.
* All operations must be able to scale to Google-scale repos (lots of commits
, lots of files): Laziness is important, must avoid accessing data
unnecessarily.
* Having as few states as possible.
* Make it incredibily hard to lose work in your repository.
* Allow concurrent edits on any commit, pending or finished.
* Make a "stacked diffs" workflow as easy as possible.
* Git-interop: Git is everywhere. We need to have good interop to be adapted.
* Pluggable storage: Must be easy to integrate with different commit storage,
virtual file systems and more.

## Base design

The initial base design is to be a conceptually simpler Mercurial, as
automatically snapshotting the working copy simplifies the UX of the
command-line interface by a huge amount and avoids many bad states.

By also choosing to operate by default on the history of the repository (
just called the "the Graph" from now on) instead of files, all history
modifying commands can be done at any point. This is a major improvement on
other version control systems as they need to re-apply a single patch on each
new ancestor before finishing the Graph rewrite. Since the Graph can be changed
at any point, the working copy cannot contain any state depending on it, thus
we have the working-copy commit, which just is another commit from the Graph's
point of view.


### Change-IDs and Changes

Since Jujutsu is oriented around a "stacked diffs" kind of workflow, which
primarily work on individually versiond patch sets, some kind of container is
needed, this is what a Change is. They are provided with a unique id to address
them easily.

TODO: expand on change-ids and "working on the graph/repo" instead of a commit


[billion-lines]: https://www.youtube.com/watch?v=W7*TkUbdqE&t=327s
[vcs]: https://en.wikipedia.org/wiki/Version_control
2 changes: 2 additions & 0 deletions mkdocs.yml
Original file line number Diff line number Diff line change
Expand Up @@ -86,6 +86,8 @@ nav:

- FAQ: 'FAQ.md'

- 'Jujutsu from first principles': 'jj_first_principles.md'

- "CLI Reference": 'cli-reference.md'

- Testimonials: 'testimonials.md'
Expand Down

0 comments on commit 0e209d4

Please sign in to comment.