Use actor based parallelization

[mbj]

• Fixes longstanding issues with nested parrallel execution. (good thing)
• More responsive UI as progress reports happen in main thread with time shedule and not per mutation state event callback. (good thing)
• Allows worker self registration (groundwork for distributed killing)
• Performance is on-par with parallel. (good thing)
• Removes frame throttle logic from reporters (good thing).
• No need to make Mutant::Mutation instances serializable, reduces building partial result objects to make them transmittable over parallel's interfaces. (good thing)
• More code. (bad thing)
• 100% line coverage (I know I know, its not MT covered fully right now) but better than master ;)

[plexus]

Mutant configuration:
Matcher:            #<Mutant::Matcher::Config match_expressions=[<Mutant::Expression: Yaks*>] subject_ignores=[] subject_selects=[]>
Integration:        rspec
Expect Coverage:    100.00%
Jobs:               4
Includes:           ["lib"]
Requires:           ["yaks"]
Available Subjects: 150
Subjects:           150
Mutations:          5543
Kills:              21
Alive:              303
Runtime:            21.52s
Killtime:           1.48s
Overhead:           1351.55%
Coverage:           6.48%
Expected:           100.00%
Active subjects:    1
Yaks::CollectionMapper#call:/home/arne/github/yaks/yaks/lib/yaks/collection_mapper.rb:9 mutations: 90
- rspec:Yaks::CollectionMapper
...............
(15/90) 100% - killtime: 1.01s runtime: 5.67s overhead: 4.67s
/home/arne/.gem/ruby/2.1.3/gems/parallel-1.3.3/lib/parallel.rb:61:in `rescue in work': Parallel::DeadWorker (Parallel::DeadWorker)
        from /home/arne/.gem/ruby/2.1.3/gems/parallel-1.3.3/lib/parallel.rb:58:in `work'
        from /home/arne/.gem/ruby/2.1.3/gems/parallel-1.3.3/lib/parallel.rb:269:in `block (4 levels) in work_in_processes'
        from /home/arne/.gem/ruby/2.1.3/gems/parallel-1.3.3/lib/parallel.rb:439:in `with_instrumentation'
        from /home/arne/.gem/ruby/2.1.3/gems/parallel-1.3.3/lib/parallel.rb:268:in `block (3 levels) in work_in_processes'
        from /home/arne/.gem/ruby/2.1.3/gems/parallel-1.3.3/lib/parallel.rb:263:in `loop'
        from /home/arne/.gem/ruby/2.1.3/gems/parallel-1.3.3/lib/parallel.rb:263:in `block (2 levels) in work_in_processes'
        from /home/arne/.gem/ruby/2.1.3/gems/parallel-1.3.3/lib/parallel.rb:136:in `block (2 levels) in in_threads'

:(

[mbj]

@plexus See it terminated ;)

[plexus]

\o/ lol

[mbj]

@plexus My whole point was to make the error visible in the first step. I really hate systems that do not blow up. The complex interactions between the inner (killfork isolation) and the outer one (for parallelization) did hide this error before.

[mbj]

@plexus The current state of this branch implements everything I wanted it to do. It includes the concurency semantics of master plus per worker reporting and some parallel related bugfixes (via removing it). If you wanted you could run it to give it some testing.

[dkubb]

Does Undefined need to be a module? Would it be more lightweight to use Undefined = BasicObject.new or something?

[mbj]

@dkubb You get a sane #inspect for free this way.

[mbj]

But I agree Module has too much public interface (I initially used Class). I'll use an anonymous object instance with an explicit #inspect defined.

[dkubb]

@mbj I finished a pass over the code.