From: Gavin Kistner <gavin.kistner@...>
Date: 2006-09-07T03:13:31+09:00
Subject: What Not to Do (a cautionary tale)

I wrote a little code generation system for my company. It's a small
script that reads in an XML file and an ERB template, feeds the XML file
to the template, and writes out the result to a file.
 
To help those in the company who don't know XPath, I extended REXML with
a couple convenience methods that let you search for elements by name
and/or attributes.
 
class REXML::Element
  # Find all descendant nodes with a specified tag name and/or
attributes
  def find( tag_name='*', attributes_to_match={} )
    self.each_element(
".//#{REXML::Element.xpathfor(tag_name,attributes_to_match)}" ){}
  end

  # Find all child nodes with a specified tag name and/or attributes
  def kids( tag_name='*', attributes_to_match={} )
    self.each_element(
"./#{REXML::Element.xpathfor(tag_name,attributes_to_match)}" ){}
  end

  ...
end

This worked very nicely. Template writers could simply do something
like:
  In this file I see the following classes: <%=
    root.find( 'class' ).map{ |el| el.attributes[ 'name' ] }.join( ', '
)
  %>

All was fine under 1.8.4.

I just upgraded to 1.8.5 and all templates using #find broke. After a
bit of sleuthing, it looks like REXML changed the Document#root method
in a way that caused it to call Elements#[], and that method uses
@element.find, mixed in from Enumerable. My Element#find was overriding
Enumerable#find, and all hell broke loose as a result.

Moral of the story (pick one or more):
[ ] Don't extend someone else's class with common method names
[ ] Don't upgrade your Ruby environment unless you need to