What is the purpose of vendor/bundle? Heroku tells me to remove it

Question

Upon pushing some changes to Heroku, I noticed a warning about vendor/bundle (see WARNING below).

What is the purpose of this directory if, according to the warning, it should be "removed" from Git tracking?

Why isn't vendor/bundle automatically .gitignore'd by default by Rails?

Should I run bundle pack? (Is it actually bundle package??)

What are the pros and cons around bundle pack (relative to both development and production)?

To make this even more confusing, there's a popular blog post, by Ryan McGeary, titled "Vendor Everything" Still Applies that strongly argues for running bundle install --path vendor and echo 'vendor/ruby' >> .gitignore and packaging gems in vendor/cache by running bundle package. Any light shed on this relative to my other concerns would be greatly appreciated.

Thank you.

-bash> git push production master
...

-----> Heroku receiving push
-----> Ruby/Rails app detected
-----> WARNING:  Removing `vendor/bundle`.
       Checking in `vendor/bundle` is not supported. Please remove this directory
       and add it to your .gitignore. To vendor your gems with Bundler, use
       `bundle pack` instead.
-----> Installing dependencies using Bundler version 1.2.1
       Running: bundle install --without development:test --path vendor/bundle --binstubs bin/ --deployment
       Using rake (0.9.2.2)
       Using i18n (0.6.0)
       ...

Answer 1

If you have the vendor/bundle directory in your project then at some point you must have run the bundle command with the --path vendor/bundle argument. This will load the files for all your project's gems (listed in Gemfile) into the vendor/bundle directory in your local project rather than to a system gem location. You would do this to completely isolate a project's gems from any other project.

Bundler is good at resolving all dependencies so it isn't necessary to use the --path but some people choose to do so as they wish to keep their gems separate and organised with their project. It also means that bundler on your local machine is set up the same way that Heroku uses bundler.

With this option you are still downloading all gems from the rubygems servers every time you run the bundle command.

bundle package takes it a step further and actually downloads the original gem files from rubygems and caches them into the vendor/cache directory. This means that you no longer need a connection to rubygems to run the bundle command as it will use the packaged files as the source. If you need to update a gem version then it will need to connect to rubygems to collect the new version the first time it is requested. Using bundle package will of course require additional disc space which may or may not be an issue depending on the situation. It would also increase deploy time and bandwidth requirements every time you pushed to Heroku.

Heroku runs the bundle command every time you git push, reading your Gemfile.lock and installing the gems required for the application to work. By default the --path vendor/bundle option is used. This is so that each application has a set of gem files separate from all other apps on Heroku. If you have the vendor/bundle directory in your source control and you push it to Heroku then you can see that there is the potential for significant conflict as it then attempts to load gems into vendor/bundle directory which already exists. If it is pushed then Heroku removes the vendor/bundle directory before it runs bundle install to remove these potential conflicts. If this is the case then you will be wasting deploy time and bandwidth by leaving vendor/bundle under version control, it's better to add it to your .gitignore.

If you want complete control over your gems on Heroku then use the bundle package command and make sure that the vendor/cache directory is under source control. When Heroku runs bundle install it will use the contents of vendor/cache as the gem source rather than using rubygems. Whether this is useful or not will be a matter of personal preference, the type of app that you are building and how often you update your gems. The Ryan McGeary post suggests that using bundle package is useful in case an old gem becomes unavailable at some point in the future. This would appear to be a bigger issue to projects/apps which are not regularly kept up to date.

From my perspective, I generally use --path vendor/bundle to keep my local setup as close as possible to Heroku's. I put vendor/bundle into my project's .gitignore file, and I don't package gems, as my projects are updated relatively regularly.

Rails has a very limited .gitignore file. You are effectively expected to build up what you need yourself, which is why vendor/bundle is not included by default.

I assume that Heroku means bundle package when they say bundle pack.