rbenv: Surviving without gemsets

Question

TL;DR

• Don't bother with gemsets; multiple versions of a gem may be installed concurrently.
• When necessary, specify which version to execute using $ gem-based-binary _version_ args notation.
• Use bundle exec when you have a Gemfile specifying the version.

gem install rails -v 3.2.13 rails _3.2.13_ new Project2 cd Project2 bundle exec rails server 

---

UPDATE: 2015-06-04

I wrote this question three years ago. Partly, it was based on a false assumption, and partly the situation has changed since then. With appreciation to @indirect for his original answer, I want to call attention to @kelvin's newer (less upvoted) answer, summarized above.

My false assumption: Only a single version of a gem could be installed at a time, hence the need for gemsets to isolate the namespace. Not true. Multiple versions of a gem may be installed concurrently. The most recent one will be used when invoked from a command line, unless you have a Gemfile specifying the version constraints and invoke the command via bundle exec, or specify the version as its first argument.

See also How can I call an older version of a gem from the commandline? re: the underscore-version notation.

---

Original question:

I have multiple projects going on using different versions of Rails. I have a workflow (described below) for creating projects using specific versions of rails, and keeping the projects isolated from each other. I'd like to experiment with other workflows, in particular, using rbenv instead of RVM, but it's not clear how to do so.

QUESTION: What is the best current practice for creating multiple rails projects, each using a different version of rails, when making use of rbenv and bundler, as opposed to rbenv-gemset or rvm?

USE CASE: I have two rails projects, called ProjectA and ProjectB. ProjectA is developed using one version of rails ("RailsA"), whereas ProjectB uses a different version ("RailsB"). How do I manage having both versions installed?

THE GEMSETS APPROACH: When I first started with Rails development, I used RVM. In addition to supporting multiple, concurrent installations of ruby, RVM supports having multiple Named Gem Sets. Each project has its own independent collection of gems (including rails itself) called a gemset:

rvm gemset create RailsA rvm gemset use RailsA # RailsA.  Note: My question is not version-specific. gem install rails --version 3.0 rails new ProjectA cd ProjectA rvm --rvmrc use `rvm current` vi Gemfile bundle install cd .. ## Now do the same for ProjectB rvm gemset create RailsB rvm gemset use RailsB gem install rails --version 3.2 rails new ProjectB cd ProjectB rvm --rvmrc use `rvm current` vi Gemfile bundle install 

Note: The very creation of the project folders should be done (IMHO) by a rails new command using the desired version of rails, since the skeleton files change from version to version. (Perhaps I should revisit this premise?)

THE BUNDLER APPROACH: I've been playing with using rbenv instead of RVM, but I don't understand the workflow as clearly. In the README.md, Sam Stephenson writes that "rbenv does not ... manage gemsets. Bundler is a better way to manage application dependencies." There is a plugin (rbenv-gemset) for getting the same results as rvm's gemsets, but Sam clearly favors using Bundler instead. Unfortunately, he doesn't elaborate on what the workflow would look like. Even the Bundler website doesn't explicitly connect all the dots of how to isolate one project from another. Several blogs and gists come to the rescue, suggesting the following ~/.bundle/config file:

--- BUNDLE_PATH: vendor/bundle 

(BTW, I'm not sure what the "---" is about. The docs make no mention of it and it doesn't seem to make a difference.)

This effectively gives each rails project its own gemset, storing the gems in ProjectX/vendor/bundle/. In fact, rails itself will be (re-)installed there, making the project completely independent of the rest of my environment, once I run bundle install.

But the elephant in the room is the chicken-and-egg problem of creating the rails project folder in the first place!! In order to create the ProjectA folder using RailsA, I need to install rails (and its numerous dependencies) first. But when I want to create ProjectB, I must then switch to using RailsB. Without gemsets, I must do some serious upgrading/downgrading. Not cool.

A possible solution is simply not to worry about what version of rails I use to create the ProjectX folder. If I then use rails 3.0 to create a 3.2 project, I could just manually create the app/assets tree. But that just irks me. Ain't there a better way?

Answer 1

Most people solve this by installing the rails gem first via gem install rails. If you refuse to do that for some reason, you can opt out of the automatic bundling that Rails attempts to do for you. This will work completely regardless of your ruby management system.

mkdir myapp cd myapp echo "source :rubygems" > Gemfile echo "gem 'rails', '3.2.2'" >> Gemfile bundle install --path vendor/bundle bundle exec rails new . --skip-bundle 

When prompted, type "y" to replace your Gemfile with the default Rails one (or not, as you prefer). Then, once it's done:

bundle install 

You're done, and you have boostrapped a new rails app with the version of your choice without installing the rails gem into rubygems.

Answer 2

Suppose you have rails 3.1.0 installed, but you want to create a new project using rails 3.2.13 which is not installed.

Let's say you want the new project to be in ~/projects/Project2.

gem install rails -v 3.2.13 cd ~/projects rails _3.2.13_ new Project2 

This will create the Gemfile for you, locked to the version of rails you specified on the command-line.

I deliberately omitted the idea of keeping a separate copy of gems for the new project, because that goes against the Bundler philosophy, which is to have all gems installed in one place. When you run rails, Bundler will pick the correct gem versions automatically from that central location. That means a project can share gems instead of installing a fresh copy for itself. (Note, however that each version of ruby you install will have its own gems. This is a good thing because native extensions likely won't work across ruby versions.)

You do have to be a bit more aware, because most commands, like rake, will load the newest version of rake that you have installed. You'll need to run bundle exec rake ... to make sure the correct version is loaded. Usually I'll run bundle exec for all commands except rails. You can create an alias to make it shorter (I use bex). To automate this with gem executables, you can use rbenv-binstubs, but you still have to be aware that running non-gem executables like ruby and irb won't automatically use the Gemfile.

Sidenote: rails new will run bundle install, which will check for the newest version of the dependencies. If you want bundler to try to use currently installed gems that satisfy the dependency requirements, you can skip the bundle install with rails new --skip-bundle, then run bundle check in the app dir.

Sidenote 2: suppose you want to use a ruby version for Project2 (e.g. 2.1.8) that's different from the default (e.g. 2.3.0). In that case, running gem install as specified above will install the gems under 2.3.0, which is a waste of time because you'll need to install the gems again under 2.1.8. To solve that problem, you can force the commands to use the preferred version via environment variable:

RBENV_VERSION=2.1.8  gem install rails -v 3.2.13 cd ~/projects RBENV_VERSION=2.1.8  rails _3.2.13_ new Project2 echo 2.1.8 > Project2/.ruby-version 

You could use rbenv shell to set the variable, but I only recommend that if you don't want rbenv to auto-switch based on .ruby-version files for the duration of that shell. It's very easy to forget that you have the variable set, and when you cd to a different project, it won't be using the version you expect.