git-run

A tool for managing multiple git repositories

  • Tag all the things! gr @work foo will run the command foo in all the paths tagged @work.
  • Auto-discovery of git repositories for easy setup and tag management.
  • gr does not reinvent any git operations: instead, it passes through and runs any unknown commands. All your git-fu will still work! e.g. gr @work git fetch is the same as running git fetch in each all the paths tagged @work.
  • Built-in commands for common pain points:
    • status for one-line summaries of repos (modified, behind/ahead, tags)
  • Extensible via plugins and middleware: REST API/Connect-style request handlers (function(req, res, next) { ... }

  • Looking for more committers! Let me know if you're interested; gr currently meets my fairly limited needs but I know can become even more useful given a larger core team and/or a stronger vision around how it can support usage in a team working on multiple repos.
  • 0.4.0: Added several usability improvements and bug fixes, courtesy of @nichtich (better handling of missing directories, support for simple paths). Added a fix that improves errors related to directory permissions, courtesy of @pnxs.
  • 0.3.0: Switched from #foo to @foo for tags; while the #foo syntax looks cool, most shells will treat it as a comment unless the tag is surrounded by quotes. Looking back at the design, I'd rather go for usability over pretty looking commands. Updated the documentation to match this change.

gr works by tagging directories with tags:

gr +@work ~/mnt/gluejs ~/mnt/microee

After this, you can run any commands on the tagged directories simply by prefixing them with gr @work. For example:

gr @work status

Outputs (actual output is colorized):

~/mnt/gluejs           2 modified [ahead 2]      @work
~/mnt/microee          Clean                     @work

E.g. path, modified, ahead/behind remote, tags. Alternatively, you can use plain git commands for a more verbose report:

gr @work git status -sb

Outputs:

in ~/mnt/gluejs

## glue2
 M lib/runner/package-commonjs/index.js
 M index.js

in ~/mnt/microee

## master

gr doesn't do any command rewriting, or introduce any new commands - I like git as it is.

First, install Node.js. Node.js adds the node and the npm command. On Ubuntu you need the nodejs-legacy package.

Next, to install gr (the name was already taken on npm):

npm install -g git-run

You may need to prefix this with sudo.

Use the auto-discovery feature to set up tags quickly:

gr tag discover

Auto-discovery searches all paths under your home directory, generates a list, and opens it in your default console editor.

It will look like this:

# Found the following directories with `.git` directories.
#
# Please add any tags you want by adding one or more `@tag`
# entries after the path name.
# For example:
#   ~/foo @work @play
# will tag ~/foo with @work and @play.
#
~/foo
~/bar/baz

Add tags after each path, save the file, and exit.

Your tags are now set up!

Verify with gr status or gr tag list. Use gr @work status or gr @work ls -lah to see how commands are executed. (status is a built-in command; ls -lah is not, so it is run in each of the paths.)

You can run auto-discovery multiple times. It makes tag-related bulk changes quite easy.

To add tab completion:

  • open your ~/.zshrc or ~/.bashrc (~/.bash_profile on OS X)
  • add the line . <(gr completion) at the end of the file
  • then, to apply this change to your current session, run source ~/.zshrc (or source ~/.bashrc or source ~/bash_profile)

Now, when you type gr <tab>, you'll see the list tags you've created. If you notice any bugs, let me know via an issue.

Some examples:

COMMANDTASK
gr @work git fetch and then gr @work statusUpdate all my work repos. This fetches the newest information from the remote, and then prints a one-line-at-a-time summary.
gr @work git diff or gr @work git diff --cachedSee diffs
gr @work jshint . --exclude=**/node_modulesRun jshint
gr @write makeRebuild all my writing via make
gr @work npm lsList install npm modules
gr @work git --no-pager log --decorate --graph --oneline -n 3Print a graph-like log

Of course, I don't actually type these out; I'm using zsh aliases instead. grd is for diff, grdc is for diff --cached; grl is for the log. For example, in .zshrc:

alias grs="gr status"
alias grl="gr git --no-pager log --decorate --graph --oneline -n 3"

You can set up similar aliases for bash; Google is your friend here.

Usage:

gr <options> <targets> <cmd>

Currently, there is just one option: --json, which switched to a machine-readable output and is used for integration tests.

Targets can be paths or tags. For example:

gr ~/foo ~/bar status
gr @work ls -lah
  • Path targets should be directories.
  • Tags refer to sets of directories. They managed using the tag built-in.

If you are using a scripting language that uses # for comments, you can also write tags as -t foo.

If no targets are given, then all tagged paths are used. For example, gr status will report the status of all repositories.

Short form:

@tag            List directories associated with "tag"
@tag <cmd>      Run a command in the directories associated with "tag"
-t <tag> <cmd>  Run a command in the directories associated with "tag"
+@tag           Add a tag to the current directory
-@tag           Remove a tag from the current directory
+@tag <path>    Add a tag to <path>
-@tag <path>    Remove a tag from <path>

Long form:

tag add <tag>   Alternative to +@tag
tag rm <tag>    Alternative to -@tag
tag add <t> <p> Alternative to +@tag <path>
tag rm <t> <p>  Alternative to -@tag <path>
tag list        List all tags (default action)
tag discover    Auto-discover git paths under ~/

Example:

gr +@work ~/bar

The command can be either one of the built-in commands, or a shell command. For example:

gr @work status
gr ~/foo ~/bar ls -lah

To explicitly set the command, use --:

gr ~/foo -- ~/bar.sh
gr @work -- git remote -v

Tags can also be specified more explicitly. For example gr -t work -t play is the same as gr @work @play.

gr tag ..
  add <t>         Add a tag to the current directory
  rm <t>          Remove a tag from the current directory
  add <t> <path>  Add a tag to <path>
  rm <t> <path>   Remove a tag from <path>
  list            List all tags (default action)
  discover        Auto-discover git paths under ~/

gr list        List all known repositories and their tags

gr status       Displays the (git) status of the selected directories.
gr status -v    Runs "git status -sb" for a more verbose status.

gr config ..
  get <k>       Get a config key (can also be a path, e.g. "tags.foo")
  set <k> <v>   Set a config key (overwrites existing value)
  add <k> <v>   Add a value to a config key (appends rather than overwriting)
  rm <k> <v>    Remove a value from a config key (if it exists)
  list          List all configuration (default action)

gr help        Show this help
gr version     Version info

TODO:

  • bootstrap: bootstraps a set of repositories from a config file.

Generally speaking, you need to do two things:

  1. install the plugin globally via npm: npm install -g foo
  2. configure gr to use the plugin: gr config add plugins foo

The new commands should now be available.

Plugins are functions which are invoked once for each repository path specified by the user. This makes it easier to write plugins, since they do not need to handle executing against multiple repository paths explicitly.

Plugins are treated a bit like a REST API: they are defined as "routes" on the gr object.

Each plugin consists of an index file which is loaded when gr is started, and which should add new "routes":

module.exports = function(gr) {
  // set up new commands on the gr object
  gr.use(['foo', 'help'], function(req, res, next) {
    console.log('Hello world');
    req.exit(); // stop processing
  });
  gr.use('foo', function(req, res, next) {
    console.log(req.argv, req.path);
    req.done(); // can be called multiple times
  });
};

Of course, req and res in the handlers are not HTTP requests, but rather objects representing the target directory (a regular object) and process.stdout.

Each "route" is called multiple times, each with one path. Thus, assuming @work matches two paths, gr @work status is translated into multiple indvidual function calls; one for each directory/repository tagged @work.

  status({ path: '/home/m/foo', argv: ... }, process.stdout, next);
  status({ path: '/home/m/bar', argv: ... }, process.stdout, next);

There are three ways to stop processing:

  1. Call res.done(). This means that the command should be called again for the next path. This is useful for processing commands that target directories.
  2. Call res.exit(). This means that the command is complete, and gr should exit. For example, we don't want to show a help text multiple times if the user calls gr @work help.
  3. Call next. This means that the current handler does not want to handle the current request. Similar to how Connect works, this is mostly used for writing middleware or falling back on a different action.

The req object describes the current request.

  • req.argv: the command line arguments passed to the command, excluding ones that have already matched. For example: given app.use(['foo', 'add'], ...) and gr foo add bar, argv = [ 'bar' ].
  • req.config: the configuration object used by gr; allows you to read and write configuration values (see the code for details).
  • req.path: the full path to the repository directory for this call
  • req.gr: the instance of gr (see index.js)
  • req.format: The desired output format, either human or json.

The res object controls

  • res.done: Call this function when you have completed processing the task.

The next function is used if you decide not to handle the current request. Calling next will make the next matching request handler run. If you encounter an error, call next(err) to output the error.

Middleware are functions that extract additional metadata.

  • req.git.remotes: TODO - a hash of remotes (for example: { origin: 'git...'}). Extracted via the git middleware.

Here are some plugin ideas:

  • extend auto-discovery beyond git
  • run tests
  • run jshint / gjslint (only modified?)
  • do npm / package.json linting
  • generate docs
  • fetch and report status as one action
  • report npm versions (and whether the version of the package on npm is up to date)
  • generate a changelog (between tagged versions)
  • check that npm modules are up to date
  • run npm link on all modules
  • generate a list of authors
  • generate a list of licences
  • xargs compatibility
  • Ability to confirm each command (to make it possible to skip)
  • Ability expose other statuses (e.g., npm outdated)

If you write a plugin, make sure to add the gr keyword to (in package.json). This makes it easy to find plugins by searching npm by tag.

Also, file a PR against this README if you want to have your plugin listed here.

(This is just a random idea) Using "meta-tags" to target commands based on git ls-files.

  • clean: Clean working directory - in other words, no tracked files are modified; no untracked files exist.
  • untracked: Has files that are not tracked (but that have not been added to tracking)
  • modified: Has files that are tracked and modified (but that have not been staged)
  • deleted: Has files that are tracked and deleted
  • staged: Has files that are staged for commit (but that have not been committed)
  • unmerged: Has files that have not been merged
  • unclean: Does not have a clean working directory.

For example: gr @clean git fetch

  • http://manpages.ubuntu.com/manpages/jaunty/man1/mr.1.html
  • https://github.com/fabioz/mu-repo
  • http://source.android.com/source/developing.html