gitq 0.0.8

git based patch manager

Git Queues

⚠️ 🪏 Work In Progress 🪏 ⚠️

This is currently a work in progress. Planned features for git-queue include:

• Unapplied patches. If rebase fails due to conflicts, give the user the choice to save commits as patch files to be applied later.

Description

This is my second attempt to make a git-based patch queue tool.

It's similar to git-pq, Quilt, Stacked Git, TopGit, Mercurial Queues, and others.

The goal is to assist the user in composing, and maintaining a well-organized queue of patches against a baseline codebase.

Use cases

Personal Integration Branch

You have multiple topic branches in progress, at various levels of maturity. One branch may be a very early work in progress, another may be in the initial stages of build-and-test in the CI environment, another may be almost done code review and ready to merge. In order to anticipate merge conflicts, functional interactions, and for general convenience, you want to do all your local development builds on an integration branch with all your open work merged together. After changes are made to the topic branches, you want to easily update the integration branch and keep going.

Submission Queue

You have a long series of changes you want to push to an upstream project. The upstream, however does not accept long series of changes. Their code review process looks at one small patch at a time. As each patch goes through the review process, changes are made and the reset of the queue must be updated accordingly.

Long Term Patch Set

You're maintaining a set of patches against an upstream project. For whatever reason, most of these patches are never going to be merged upstream. You need to repeatedly rebase this patch set onto new versions of the project. Essentially, the patches are your source code. You would like to use a version control system to track changes to the patches.

But Why?

None of the existing tools do quite what I want. In particular, Git Queues has the following distinguishing characteristics

• Queues can have multiple baselines.

• A queue is a git branch. All queue information is stored in a single branch. There are no auxiliary refs, no information stored anywhere else. A queue can be pushed and pulled to remotes as an ordinary branch. There is no git queue push or git queue pull.

• Each patch is a git commit. As much as possible, ordinary git commands are used to manipulate patches.

• The commit history of a queue feels like a git commit history. That is, commits are user-curated checkpoints, with user-written commit messages. The history is not a detailed log of every micro-operation that was used to create a queue. The data model is as simple as possible. Git is a "stupid" content tracker, because it tracks versions of the content of a source directory. There is no specific representation in git to, for example, move a file. The contents of the file are simply moved to a new location. Git Queues aims to track the content of a patch series in the same spirit that Git tracks the content of a directory.

git-queue

git-queue manages a queue. A queue is a bunch of patches.

Patches are ordinary, non-merge git commits.

The patches in a queue sit on one or more baselines. Baselines are commits which the queue is based on in terms of git history, but are not part of the queue. Baselines may be added, removed, or refreshed. Commits in the old baselines may be left behind, but the patches in the queue are carried forward.

If there are more than one baselines, then the queue's patches will sit on merge commits brining them together.

If the baselines conflict, the queue will also need to contain user merges.

User merges are merge commits prepared by the user to reconcile conflicting baselines. When the queue is rebased, user merges will be carried forward if they can still resolve conflicts in the baselines. If not, the user will need to prepare new ones. If the baselines no longer conflict, user merges may be left behind during a rebase.

Queue Metadata

A queue is a branch with a .git-queue file at the root, or a bare branch where the metadata is stored in git config (branch.<name>.git-queue) instead. Baselines are recorded in the queue metadata. They specify the shas of the current baseline as well as the branch names they came from. This information is used to rebase the queue.

Storing metadata in a .git-queue file means it travels with the branch — push and pull work normally, and anyone who clones or fetches the branch has everything they need to rebase it.

Bare branches keep the commit history clean — no baseline commits and no .git-queue file appear in the branch, so the history looks exactly like an ordinary topic branch. The tradeoff is that the metadata lives only in local git config and is not pushed to remotes.

Create a bare queue with git queue init --bare, or convert an existing queue with git queue rebase --bare / git queue rebase --no-bare.

git queue SUBCOMMAND [OPTIONS]

git queue init [--bare] BASELINE...

Initialize a queue. Each BASELINE is a branch, tag, or commit. --bare stores metadata in git config instead of a .git-queue file.

git queue rebase [--bare | --no-bare]

Rebase the queue onto its baselines, incorporating any upstream changes. --bare converts to a bare branch; --no-bare converts back to a .git-queue file.

git queue add [--bare | --no-bare] BASELINE...

Add one or more baselines and rebase.

git queue remove [--bare | --no-bare] BASELINE...

Remove one or more baselines and rebase.

git queue continue

Resume a suspended operation.

git queue abort

Abort a suspended operation and restore git to its previous state.

git queue status

Print the current operation status.

git queue tidy

Normalize and rewrite the .git-queue file in the current branch.

git queue commit

Commit changes to a queue to a historiography.

Git tracks changes to files in a commit history. git-queue can track changes to a queue to a history of histories, aka a historiography.

A historiography is itself represented as a git history, but the files being tracked represent a queue, that is a set of changes from a specified baseline.

This is done by serializing the commits in the queue (but not the baselines), into patch files, in almost the same format that git format-patch uses. Unlike git-format-patch, git git-queue records the committer, committer date, and parents of each commit in the patch file. This means that the patch files recorded in a historiography have enough information to reconstruct the exact original commit shas that they were created from.

git-swap

git-swap reverses the order of two adjacent commits, while holding the final content constant.

git-edit

git-edit checks out a specific commit for editing, then replays everything above it when you continue — similar to git rebase -i with edit.

git-split

git-split splits a single commit into two or more commits, while holding the final content constant.

git-squash

git-squash squashes a commit into its parent.

git-drop

git-drop deletes a commit from history, replaying all commits above it.