bundle-exec - Man Page

Execute a command in the context of the bundle

Synopsis

bundle exec [--keep-file-descriptors] command

Description

This command executes the command, making all gems specified in the [Gemfile(5)][Gemfile(5)] available to require in Ruby programs.

Essentially, if you would normally have run something like rspec spec/my_spec.rb, and you want to use the gems specified in the [Gemfile(5)][Gemfile(5)] and installed via bundle install(1) bundle-install.1.html, you should run bundle exec rspec spec/my_spec.rb.

Note that bundle exec does not require that an executable is available on your shell's $PATH.

Options

--keep-file-descriptors: Passes all file descriptors to the new processes. Default is true from bundler version 2.2.26. Setting it to false is now deprecated.

Bundle Install --binstubs

If you use the --binstubs flag in bundle install(1) bundle-install.1.html, Bundler will automatically create a directory (which defaults to app_root/bin) containing all of the executables available from gems in the bundle.

After using --binstubs, bin/rspec spec/my_spec.rb is identical to bundle exec rspec spec/my_spec.rb.

Environment Modifications

bundle exec makes a number of changes to the shell environment, then executes the command you specify in full.

make sure that it's still possible to shell out to bundle from inside a command invoked by bundle exec (using $BUNDLE_BIN_PATH)
put the directory containing executables (like rails, rspec, rackup) for your bundle on $PATH
make sure that if bundler is invoked in the subshell, it uses the same Gemfile (by setting BUNDLE_GEMFILE)
add -rbundler/setup to $RUBYOPT, which makes sure that Ruby programs invoked in the subshell can see the gems in the bundle

It also modifies Rubygems:

disallow loading additional gems not in the bundle
modify the gem method to be a no-op if a gem matching the requirements is in the bundle, and to raise a Gem::LoadError if it's not
Define Gem.refresh to be a no-op, since the source index is always frozen when using bundler, and to prevent gems from the system leaking into the environment
Override Gem.bin_path to use the gems in the bundle, making system executables work
Add all gems in the bundle into Gem.loaded_specs

Finally, bundle exec also implicitly modifies Gemfile.lock if the lockfile and the Gemfile do not match. Bundler needs the Gemfile to determine things such as a gem's groups, autorequire, and platforms, etc., and that information isn't stored in the lockfile. The Gemfile and lockfile must be synced in order to bundle exec successfully, so bundle exec updates the lockfile beforehand.

Loading

By default, when attempting to bundle exec to a file with a ruby shebang, Bundler will Kernel.load that file instead of using Kernel.exec. For the vast majority of cases, this is a performance improvement. In a rare few cases, this could cause some subtle side-effects (such as dependence on the exact contents of $0 or __FILE__) and the optimization can be disabled by enabling the disable_exec_load setting.

Shelling out

Any Ruby code that opens a subshell (like system, backticks, or %x{}) will automatically use the current Bundler environment. If you need to shell out to a Ruby command that is not part of your current bundle, use the with_unbundled_env method with a block. Any subshells created inside the block will be given the environment present before Bundler was activated. For example, Homebrew commands run Ruby, but don't work inside a bundle:

Bundler.with_unbundled_env do
  `brew install wget`
end

Using with_unbundled_env is also necessary if you are shelling out to a different bundle. Any Bundler commands run in a subshell will inherit the current Gemfile, so commands that need to run in the context of a different bundle also need to use with_unbundled_env.

Bundler.with_unbundled_env do
  Dir.chdir "/other/bundler/project" do
    `bundle exec ./script`
  end
end

Bundler provides convenience helpers that wrap system and exec, and they can be used like this:

Bundler.clean_system('brew install wget')
Bundler.clean_exec('brew install wget')

Rubygems Plugins

At present, the Rubygems plugin system requires all files named rubygems_plugin.rb on the load path of any installed gem when any Ruby code requires rubygems.rb. This includes executables installed into the system, like rails, rackup, and rspec.

Since Rubygems plugins can contain arbitrary Ruby code, they commonly end up activating themselves or their dependencies.

For instance, the gemcutter 0.5 gem depended on json_pure. If you had that version of gemcutter installed (even if you also had a newer version without this problem), Rubygems would activate gemcutter 0.5 and json_pure <latest>.

If your Gemfile(5) also contained json_pure (or a gem with a dependency on json_pure), the latest version on your system might conflict with the version in your Gemfile(5), or the snapshot version in your Gemfile.lock.

If this happens, bundler will say:

You have already activated json_pure 1.4.6 but your Gemfile
requires json_pure 1.4.3. Consider using bundle exec.

In this situation, you almost certainly want to remove the underlying gem with the problematic gem plugin. In general, the authors of these plugins (in this case, the gemcutter gem) have released newer versions that are more careful in their plugins.

You can find a list of all the gems containing gem plugins by running

ruby -e "puts Gem.find_files('rubygems_plugin.rb')"

At the very least, you should remove all but the newest version of each gem plugin, and also remove all gem plugins that you aren't using (gem uninstall gem_name).

Info

June 2024