[Inline] Add bundler/inline to allow using Bundler without a Gemfile

[segiddins]

Closes rubygems/bundler-features#47.

Enables scripts such as the following to work on their own, as long as bundler is installed:

#!/usr/bin/env ruby

require 'bundler/inline'

gemfile do
  source 'https://rubygems.org'
  gem 'json'
  gem 'nap', require: 'rest'
  gem 'cocoapods', '> 0.pre'
end

puts Pod::VERSION

[segiddins]

Those 2 spec failures look like API timeouts to me.

[TimMoore]

I restarted the failing jobs.

[indirect]

Er, just to double-check, in the example script, you would still have to require 'cocoapods', right?

[segiddins]

@indirect right now, using bundler/inline's gemfile method will also call require on the environment, just as if you'd somehow done a require 'bundler/require' afterwards.

[indirect]

There is no bundler/require, though. :) There's a bundler/setup, but that means you still need to explicitly call require.

[segiddins]

Ah my bad. It'd be like calling require 'bundler/setup'; Bundler.require

[indirect]

In order to encourage reasonable growth of a script into a multi-file project, can we leave out the Bundler.require part? Requiring the world up front is an anti-pattern that I wish Rails didn't encourage. :'(

[segiddins]

André and I just agreed that not auto-requiring in this circumstance would make users do a redundant gem 'foo' && require 'foo' for every gem that they include in the inline gemfile, which is probably worse than (slightly) encouraging en-mass requiring.

[indirect]

Can you document this somewhere, preferably with examples? :D (Maybe even on the Bundler website?!)

[andremedeiros]

This is really nice! 👍

[cstrahan]

@segiddins Excellent! This new feature should greatly simplify the way we do Ruby packaging in NixOS. Thanks!

[schneems]

Requiring the world up front is an anti-pattern

Do you have more thoughts on this written down somewhere? I would think we want to preload to take advantage of CoW with forking webservers.

[cstrahan]

@schneems I believe the anti-pattern that @indirect is referring to is the fact that - assuming you don't require what you use - your tests (as just one example) can't be run without requiring everything up front. So while the act of requiring everything up front is great (with respect to CoW), you should still require what you need where you need it.

[segiddins]

This is not really a relevant discussion for this particular feature, which is targeted at single-file scripts.

[schneems]

Thank you @cstrahan, makes sense. Thanks for the feature @segiddins i cede this thread back to you.