Re: Rack test spec_rack_runtime fails on JRuby due to timing

[Christian Neukirchen <chneukirchen@gmail.com>]

On Mon, Jan 18, 2010 at 7:38 AM, Ryan Tomayko <r@tomayko.com> wrote:
> On Sun, Jan 17, 2010 at 10:20 AM, Charles Oliver Nutter
> <headius@headius.com> wrote:
>> Hey, the following test currently fails on JRuby, but not for a very
>> good reason:
>>
>>  specify "should allow multiple timers to be set" do
>>    app = lambda { |env| sleep 0.1; [200, {'Content-Type' => 'text/
>> plain'}, "Hello, World!"] }
>>    runtime1 = Rack::Runtime.new(app, "App")
>>    runtime2 = Rack::Runtime.new(runtime1, "All")
>>    response = runtime2.call({})
>>
>>    response[1]['X-Runtime-App'].should =~ /[\d\.]+/
>>    response[1]['X-Runtime-All'].should =~ /[\d\.]+/
>>
>>    Float(response[1]['X-Runtime-All']).should > Float(response[1]['X-
>> Runtime-App'])
>>  end
>>
>> The test here is trying to time that the "All" timer is longer than
>> the "App" timer. Unfortunately, when running on JRuby, the All time is
>> often below 1ms, which is shorter than the granularity for our Time
>> class (JVM only provides current time to 1ms granularity). As a
>> result, the test usually fails.
>>
>> This came up because a friend of JRuby is working to prepare a Gentoo
>> ebuild, and wants to get all related packages running green.
>>
>> I'm not sure the best way to fix this; the difference in execution
>> time between the "App" and the "All" wrapper are very small, and a
>> timing test like this is probably going to be unpredictable. I have
>> made a JRuby-specific patch here to use the JVM's nanosecond timer
>> (which can be used for deltas but not absolute times), but it's
>> obviously JRuby-specific:
>>
>> http://gist.github.com/279475
>>
>> Running the tests with this modification shows the "All" time to range
>> around 0.00015s and the "App" time around 0.00003s.
>>
>> Thoughts?
>
> Perhaps changing the assertion to be:
>
>    Float(response[1]['X-Runtime-All']).should >=
> Float(response[1]['X-Runtime-App'])
>
> ... or removing the assertion entirely.

... or introducing/moving a second sleep.

> Thanks,
> Ryan
>
-- 
Christian Neukirchen  <chneukirchen@gmail.com>  http://chneukirchen.org