Bug #12109: Exception during class load can cause partially-loaded class

• Status: Rejected
• Assignee: -
• Target version: -
• ruby -v: ruby 2.2.4p230 (2015-12-16 revision 53155) [x86_64-darwin15]
• Backport: 2.0.0: UNKNOWN, 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: UNKNOWN

[arrtchiu (Matt U)]

I noticed this bug using Rails, which lazy-loads classes (via ActiveSupport.)

Here's a minimal test case:

#!/usr/bin/env ruby

begin
  # simulate loading a class from file at runtime
  eval %q{
    class Foo
      def self.bar
        fail 'bar'
      end

      bar

      def self.baz
        puts 'baz'
      end
    end
  }
rescue => e
  puts "rescued: #{e}"
end
# => rescued: bar

Foo.baz
# => undefined method `baz' for Foo:Class (NoMethodError)

EXPECTED
Either:

• No exception - finish adding the method to the class, or
• 'undefine' the class, at least making it obvious that the class didn't load completely (or in the case of rails/activesupport, allow it to be reloaded on-demand later)

[shyouhei (Shyouhei Urabe)]

• Description updated (diff)

You can write virtually anything inside a class definition. I don't think it's practically possible to revert all operations inside of a require (that needs a full ACID properties I guess).

Then on the other hand is it a wise idea to ignore exceptions? I don't think so. Because there should be a way for a class author to bail out the creation of a class on some reason. That is not a well-doing, but should there be a way to do so at least.

I heard that chef scripts intentionally raises from inside their definitions, to abnormally quit a provisioning. That sort of things sometimes happen.

[marcandre (Marc-Andre Lafortune)]

• Status changed from Open to Rejected

Closing as this is not a bug.

I don't see any change of the sort happening as both of your expectations make no sense. Please understand that when loading a file, ruby executes them. The statements are executed in order. The def could be changed to define_method which is not a keyword but an instance method of Module.

Is it rails that rescues the exception when autoloading? I can't remember. If so, patch that behavior if you really want to.