bundler-inject

bundler-inject is a bundler plugin that allows a developer to extend a project with their own personal gems and/or override existing gems, without having to modify the Gemfile, thus avoiding accidental modification of git history.

Installation

Add these lines to your application's Gemfile:

plugin 'bundler-inject'
require File.join(Bundler::Plugin.index.load_paths("bundler-inject")[0], "bundler-inject") rescue nil

Additionally, commit a bundler.d/.gitkeep file, and add /bundler.d to your .gitignore file. This will be one of the locations developers can place their overrides.

Usage

Once the above lines are in the Gemfile, subsequent bundle commands will attempt to evaluate extra gemfiles from two locations if they are present.

• global - ~/.bundler.d/*.rb
• project - $PROJECT_DIR/bundler.d/*.rb

For example, a developer may prefer the pry gem over irb, but can't use it because the project Gemfile doesn't have pry in it. Instead of modifying the Gemfile and hoping they don't commit the changes, they can create a bundler.d/developer.rb file with the contents set to gem "pry". From then on the developer can use pry, even though the project didn't state it explicitly.

Since this example developer likely always wants to use pry, it would be preferable to specify this as a global choice in a ~/.bundler.d/developer.rb file instead.

override_gem

override_gem is an extra DSL command that allows overriding existing gems from the Gemfile. A useful example of this is if you are making a change to a dependent gem, and temporarily want to override the existing gem definition with, a git or path reference.

For example, there is a project with a Gemfile with gem "foo" in it. We want to fix a bug in foo, and we intend to make a pull request to upstream, but in the interim we want to test our foo change with the project. Instead of modifying the Gemfile and hoping we don't commit the changes, we can create a bundler.d/developer.rb with the contents set to override_gem "foo", :git => "https://github.com/me/foo.git". bundler-inject will output a warning to the screen to make us aware we are overriding, and then it will use the new definition.

override_gem will raise an exception if the specified gem does not exist in the original Gemfile.

ensure_gem

ensure_gem is an extra DSL command similar to override_gem, and primarily meant for the global override file.

One issue with the global file is that it specifies a new gem with gem, but that gem already exists in the project you will get a nasty warning. Conversely, if it specifies an override with override_gem, but the gem does not exist in the project you will get an exception. To deal with these issues, you can use ensure_gem in your global file.

ensure_gem works by checking if the gem is already in the dependency list, and comparing the options specified. If the dependency does not exist, it uses gem, otherwise if the options or version specified are significantly different, it will use override_gem, otherwise it will just do nothing, deferring to the original declaration.

What is this sorcery?

While this is technically a bundler plugin, bundler-inject does not use the expected plugin hooks/sources/commands. To understand how this works and why these were not used, it's useful to understand how bundler passes over your Gemfile.

For bundle install/bundle update, bundler makes two passes over the Gemfile.

On the first pass, bundler executes your Gemfile in a Bundler::Plugin::DSL context. This class is a special subclass of Bundler::Dsl, where nearly all of the usual DSL methods like gem and gemspec are a no-op, and the plugin DSL method does it's thing. So, after the first pass, bundler sees only your plugin gems, and will then install them into .bundle/plugin.

On the second pass, bundler executes your Gemfile in a Bundler::Dsl context, where the usual DSL methods like gem and gemspec do their thing, and the plugin DSL method is instead a no-op. This is the pass most people think of.

On the very first install of the plugin, between the two passes, bundler will load your plugin to see what kind of features it has, whether hooks, sources, or commands, and will store this information in the .bundle/plugin/index file. It would seem that since the plugin code is loaded this would be an opportune time to do what we need, but unfortunately this only happens on initial install. On subsequent bundle update calls, bundler sees that the index file exists, and doesn't need to load the plugin, since all of the information is in the index.

Complicating the matter further, for bundle check/bundle exec, bundler makes only one pass over the Gemfile in a Bundler::Dsl context.

One immediate question on your mind may be, why not just define a fake hook or source, and then put the code in there. Unfortunately, the problem is that any of hooks, sources, or commands that will trigger the load of the plugin occur after the Gemfile has already been passed over, preventing us from adding to the DSL, as well as loading extra gemfiles.

As such, the earliest we can possibly trigger our plugin load in all cases is immediately after we declare the plugin in the Gemfile. This is why we need that magic line after the plugin line in the Gemfile.

Once we have the ability to trigger a load of our plugin, whether direct via the magic line, or automatically on first plugin installation, then the plugin can manipulate the Bundler::Dsl to add the override_gem method and trigger the eval of extra gemfiles.

Development

Development of bundler plugins can be a little strange, with a few gotchas.

1. bundler installs your gem into the .bundle/plugin directory of the target project.
2. plugin "bundler-inject", :path => "/path/to/bundler-inject" doesn't work as expected since bundler needs to "install" your gem into .bundle/plugin, and thus doesn't know how. To get around this, use :git => File.expand_path("/path/to/bundler-inject").
3. If you are using :git (which also applies to :path with the workaround above), bundler will only consider committed code. Therefore, you must commit your code in a temporary commit if you want it to be picked up.
4. bundler plugins are copied to .bundle/plugin only on first install, and then updated only if a change is detected. Unless you have a :ref or a changing version number, bundler will think your gem hasn't changed and will not update it, even if you commit something. To force bundler to pull in your changes, you will have to rm -rf .bundle/plugin.

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/ManageIQ/bundler-inject.

License

This project is available as open source under the terms of the Apache License 2.0.