BUNDLE-EXEC(1) BUNDLE-EXEC(1) 1mNAME0m 1mbundle-exec 22m- Execute a command in the context of the bundle 1mSYNOPSIS0m 1mbundle exec 22m[--keep-file-descriptors] 4mcommand0m 1mDESCRIPTION0m This command executes the command, making all gems specified in the [1mGemfile(5)22m][Gemfile(5)] available to 1mrequire 22min Ruby programs. Essentially, if you would normally have run something like 1mrspec0m 1mspec/my_spec.rb22m, and you want to use the gems specified in the [1mGem-0m 1mfile(5)22m][Gemfile(5)] and installed via bundle install(1) 4mbun-0m 4mdle-install.1.html24m, you should run 1mbundle exec rspec spec/my_spec.rb22m. Note that 1mbundle exec 22mdoes not require that an executable is available on your shell's 1m$PATH22m. 1mOPTIONS0m 1m--keep-file-descriptors0m Exec in Ruby 2.0 began discarding non-standard file descriptors. When this flag is passed, exec will revert to the 1.9 behaviour of passing all file descriptors to the new process. 1mBUNDLE INSTALL --BINSTUBS0m If you use the 1m--binstubs 22mflag in bundle install(1) 4mbun-0m 4mdle-install.1.html24m, Bundler will automatically create a directory (which defaults to 1mapp_root/bin22m) containing all of the executables available from gems in the bundle. After using 1m--binstubs22m, 1mbin/rspec spec/my_spec.rb 22mis identical to 1mbun-0m 1mdle exec rspec spec/my_spec.rb22m. 1mENVIRONMENT MODIFICATIONS0m 1mbundle exec 22mmakes a number of changes to the shell environment, then executes the command you specify in full. o make sure that it's still possible to shell out to 1mbundle 22mfrom inside a command invoked by 1mbundle exec 22m(using 1m$BUNDLE_BIN_PATH22m) o put the directory containing executables (like 1mrails22m, 1mrspec22m, 1mrackup22m) for your bundle on 1m$PATH0m o make sure that if bundler is invoked in the subshell, it uses the same 1mGemfile 22m(by setting 1mBUNDLE_GEMFILE22m) o add 1m-rbundler/setup 22mto 1m$RUBYOPT22m, which makes sure that Ruby pro- grams invoked in the subshell can see the gems in the bundle It also modifies Rubygems: o disallow loading additional gems not in the bundle o modify the 1mgem 22mmethod to be a no-op if a gem matching the require- ments is in the bundle, and to raise a 1mGem::LoadError 22mif it's not o Define 1mGem.refresh 22mto 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 o Override 1mGem.bin_path 22mto use the gems in the bundle, making system executables work o Add all gems in the bundle into Gem.loaded_specs Finally, 1mbundle exec 22malso implicitly modifies 1mGemfile.lock 22mif the lock- file and the Gemfile do not match. Bundler needs the Gemfile to deter- mine things such as a gem's groups, 1mautorequire22m, and platforms, etc., and that information isn't stored in the lockfile. The Gemfile and lockfile must be synced in order to 1mbundle exec 22msuccessfully, so 1mbundle0m 1mexec 22mupdates the lockfile beforehand. 1mLoading0m By default, when attempting to 1mbundle exec 22mto a file with a ruby she- bang, Bundler will 1mKernel.load 22mthat file instead of using 1mKernel.exec22m. 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 1m$0 22mor 1m__FILE__22m) and the optimiza- tion can be disabled by enabling the 1mdisable_exec_load 22msetting. 1mShelling out0m Any Ruby code that opens a subshell (like 1msystem22m, backticks, or 1m%x{}22m) 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 1mwith_clean_env 22mmethod 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_clean_env do `brew install wget` end Using 1mwith_clean_env 22mis 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 1mwith_clean_env22m. Bundler.with_clean_env do Dir.chdir "/other/bundler/project" do `bundle exec ./script` end end Bundler provides convenience helpers that wrap 1msystem 22mand 1mexec22m, and they can be used like this: Bundler.clean_system('brew install wget') Bundler.clean_exec('brew install wget') 1mRUBYGEMS PLUGINS0m At present, the Rubygems plugin system requires all files named 1mrubygems_plugin.rb 22mon the load path of 4many24m installed gem when any Ruby code requires 1mrubygems.rb22m. This includes executables installed into the system, like 1mrails22m, 1mrackup22m, and 1mrspec22m. Since Rubygems plugins can contain arbitrary Ruby code, they commonly end up activating themselves or their dependencies. For instance, the 1mgemcutter 0.5 22mgem depended on 1mjson_pure22m. If you had that version of gemcutter installed (even if you 4malso24m had a newer ver- sion without this problem), Rubygems would activate 1mgemcutter 0.5 22mand 1mjson_pure 22m. If your Gemfile(5) also contained 1mjson_pure 22m(or a gem with a dependency on 1mjson_pure22m), the latest version on your system might conflict with the version in your Gemfile(5), or the snapshot version in your 1mGem-0m 1mfile.lock22m. 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 1mgemcutter 22mgem) 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 -rubygems -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 (1mgem0m 1muninstall gem_name22m). November 2018 BUNDLE-EXEC(1)