Recently we had to install our rails app on a box, but we didn't have root access. The machine had ruby 1.8.6 and rubygems 0.9.6 (ouch). We can live with Ruby 1.8.6; but to run modern rails, we needed modern rubygems.

The rubygems install page has a helpful section on how to install locally without root access. Note: be sure to add export GEM_HOME=~/rubygems/gems to your .bashrc or your .bash_profile or else the next time you log in, your system gets confused again.

$ mkdir rubygems
$ cd rubygems-a.b.c
$ export GEM_HOME=~/rubygems/gems
$ ruby setup.rb --prefix=~/rubygems

Unfortunately, this didn't go as smoothly as I hoped. The result:

" `gem_original_require': no such file to load -- rubygems/exceptions"

and many other similar error messages.

The issue was that since the server installation already had an old version of rubygems installed, my local gem script was calling up the old rubygems files, but using my new GEM_HOME. A big mess.

What needs to happen is that your new gem executable needs to load your new rubygems files, and work with your new GEM_HOME. There may be a better way to do this, but I have found a workable solution for my case. It involves messing with the executable files that rubygems installs.

If you take a look at your gem executable, you'll see it is just a text file, a ruby program that executes itself.

$ cat  ~/rubygems/bin/gem

#!/usr/local/bin/ruby
#--
# Copyright 2006 by Chad Fowler, Rich Kilmer, Jim Weirich and others.
# All rights reserved.
# See LICENSE.txt for permissions.
#++

require 'rubygems'
require 'rubygems/gem_runner'
require 'rubygems/exceptions'

...

If you are familiar with shell scripts, you'll know that the first line (the shebang) specifies the binary to use to process the file. In this case, its #!/usr/local/bin/ruby. In general this is great. Ruby loads up the file and runs the script and magic happens. But what if the require 'rubygems' line is loading up the wrong rubygems?

We need to tell ruby where to find our custom rubygems libraries. Luckily, ruby has an option for this, -I path/to/lib. So, all we need to do is open up our gem executable and modify the shebang line to include this option.

$ vim  ~/rubygems/bin/gem

#!/usr/local/bin/ruby -I ~/rubygems/lib
#--
# Copyright 2006 by Chad Fowler, Rich Kilmer, Jim Weirich and others.
# All rights reserved.
# See LICENSE.txt for permissions.
#++

require 'rubygems'
require 'rubygems/gem_runner'
require 'rubygems/exceptions'

This tells the ruby interpreter to start the process with our custom path in the include path. If we have a newer version of the rubygems library there, they get required instead of the old versions. I can now install gems locally into my GEM_HOME without needing root access like so:

$ gem install mongrel

But that's not all. mongrel also installs its own executable, and when you try to run it, it will require rubygems. And yes, you guessed it, it will find the old version of rubygems and get completely confused. So again, we need to go into the executable file ~/rubygems/gems/bin/mongrel_rails and modify the shebang line.

As far as I can tell, you would need to do this for each and every gem that installs its own executable.

Update: after further consideration, using the $LOAD_PATH environment variable would probably be more robust, but my app is already installed. Perhaps you will let me know how it goes with $LOAD_PATH