Is there any harm in using instance variables inside an RSpec Matcher like this? Will I keep garbage values from use to use?

InstanceVariablesInRSpecMatcher.rb

    RSpec::Matchers.define :eventually_redirect_to do | target_url |
      def collect_redirect_urls_from(url)
        redirect_urls = []
        response = RestClient.get(url) do | response, request, result, &block |
          redirect_urls.push(response.headers[:location]) if response.code == 301
          response.return!(request, result, &block)
        end
        redirect_urls
      end

      @redirected_urls = []

      match do | request_url |
        @redirected_urls = collect_redirect_urls_from(request_url)
        @redirected_urls.include? target_url
      end

      failure_message do | actual |
        where_it_redirected_clause = @redirected_urls.empty? ? "redirected nowhere" : "only redirected to #{@redirected_urls}"
        "expected response to redirect eventually to #{target_url}, but #{where_it_redirected_clause}"
      end

      failure_message_when_negated do | actual |
        "expected response never to redirect eventually to #{target_url}"
      end

      describe do
        "checks eventual redirect response"
      end
    end

Also, line #4 doesn't need to capture the result. It is an unnecessary variable set.

That bug was fixed a long time ago. It's totally fine to use instance variables inside a DSL-defined custom matcher. In fact, chain almost always uses instance variables:

http://www.rubydoc.info/gems/rspec-expectations/RSpec/Matchers/DSL/Macros#chain-instance_method

Note, however, that the @redirected_urls = [] line isn't doing what you think it is -- that's because the define block is essentially evaluated as a class body, while the individual match, etc blocks define instance methods and will run with a self of an instance of the class. So that line is defining an instance variable on a different object compared to the rest of the @redirect_urls uses.

It turns out not to matter in your case because the match logic is always run before the failure_message logic, so @redirect_urls always has a value when referenced in failure_message, but if you wanted to keep something like that (e.g. because you find it easier to reason about or whatever), I'd recommend changing from an instance variable to an instance variable -- after all, all the blocks are closures and retain access to local variables.

What version of RSpec are you using? I ask because in your example, describe should be description. In older rspec 2 versions, describe was unfortunately monkey patched onto every object, which meant that it was available in any context such as this. On RSpec 3 (and probably later RSpec 2 versions, although I didn't test this just now) you'll get a NameError since we've been so much more intentional about eliminating monkey patching.

So the fact that you are able to use describe like that suggests you are on an older RSpec 2 version, which may still have the instance variable bug you referenced. I'd highly recommend upgrading to RSpec 3 as there are so many bug fixes, new features and other improvements.