Decouple CAS process management from integration tests

[hannahwhy]

This PR contains an another approach to getting Aker's test suite running in ActiveSupport 4.0 (and better future-proofing it). My previous approach was https://github.com/NUBIC/aker/pull/30; it should be ignored.

This approach uses a pair of URL generation programs to set ports for all test-related services. These generated URLs are paired with tasks to start a CAS callback and server manager. Finally, the CI script has been changed to download, start, and stop CAS servers.

[hannahwhy]

I have run this new configuration using

• CI_RUBY=1.9.3 ACTIVESUPPORT_VERSION=as_3.2
• CI_RUBY=2.0.0 ACTIVESUPPORT_VERSION=as_4.0

Both test suite runs passed.

CI_RUBY=jruby ACTIVESUPPORT_VERSION=as_4.0 had one error:

SSL_connect SYSCALL returned=5 errno=0 state=SSLv3 read finished A (OpenSSL::SSL::SSLError)
/Users/dwy870/.rvm/rubies/ruby-2.0.0-p247/lib/ruby/2.0.0/net/http.rb:918:in `connect'
/Users/dwy870/.rvm/rubies/ruby-2.0.0-p247/lib/ruby/2.0.0/net/http.rb:918:in `block in connect'
/Users/dwy870/.rvm/rubies/ruby-2.0.0-p247/lib/ruby/2.0.0/timeout.rb:52:in `timeout'
/Users/dwy870/.rvm/rubies/ruby-2.0.0-p247/lib/ruby/2.0.0/net/http.rb:918:in `connect'
/Users/dwy870/.rvm/rubies/ruby-2.0.0-p247/lib/ruby/2.0.0/net/http.rb:862:in `do_start'
/Users/dwy870/.rvm/rubies/ruby-2.0.0-p247/lib/ruby/2.0.0/net/http.rb:857:in `start'
./features/support/mechanize_test.rb:35:in `/^I have logged into CAS using "([^\"]*)" \/ "([^\"]*)"$/'
features/cas.feature:56:in `Given I have logged into CAS using "mr296" / "mr296"'

I don't know if this is related to the changes in this PR.

[hannahwhy]

Strike that. Looks like I wasn't using JRuby after all for that test run. Actually using JRuby results in this:

rake aborted!
wrong exec option: [:out, :err]

I keep forgetting that mucking about in Process is dangerous territory when you cross the Ruby -> JRuby line. I'll look for the best workaround.

[hannahwhy]

OK, now things seem to be happy (ish) on JRuby.

The spawned CAS processes still don't die when they're told to die; it looks like there's something funny with Process.spawn on JRuby. (Said processes correctly die on MRI 1.9/2.0.) We might just have to rely on Jenkins to clean up the mess for the JRuby case.

@rsutphin: any comments?

[rsutphin]

I have not tried running it myself, but the code looks reasonable to me.

[hannahwhy]

Alright, I'll merge this.

[hannahwhy]

Hey, neat:

https://public-ci.nubic.northwestern.edu/job/Aker/877/

[rsutphin]

Nice work.