Publish breed agent set changes for use by the NW extension

[nicolaspayette]

This pull request addresses #399. It was originally just a commit on the 5.0.x branch (523b890) but I reverted it (0fe90a8) and turned it into a pull request in order to address the following questions by @SethTisue with further commits and comments (to come):

• what about World3D? (changes to World often require corresponding changes to World3D)
• where is the test coverage?
• this code runs every time a turtle or link is born or dies, right? is there any performance impact on our benchmark suite?
• (what does scala.collection.mutable.Publisher look like? I don't know anything about that class in particular, there's some pretty janky code in the Scala stdlib, so I hesitate and/or investigate whenever adding a dependency on some new part of it)

[TheBizzle]

So... am I supposed to merge this right now, or is it waiting on more things to come?

[nicolaspayette]

Waiting for more.

[nicolaspayette]

• what about World3D? (changes to World often require corresponding changes to World3D)

d7ae29e

• where is the test coverage?

cea8012
9f35a47

• (what does scala.collection.mutable.Publisher look like? I don't know anything about that class in particular, there's some pretty janky code in the Scala stdlib, so I hesitate and/or investigate whenever adding a dependency on some new part of it)

https://github.com/scala/scala/blob/v2.9.2/src/library/scala/collection/mutable/Publisher.scala
https://github.com/scala/scala/blob/v2.9.2/src/library/scala/collection/mutable/Subscriber.scala

I'm not sure it's the most well-maintained part of the Scala stdlib, but it seems like fairly straightforward code to me. Also, I've already used them for the plotting and drawing layer event sourcing in master. If there was a problem with them (I don't think there is) we'd need to address it there as well.

• this code runs every time a turtle or link is born or dies, right? is there any performance impact on our benchmark suite?

Here are benchmarks from my laptop from before and after the change:

                before  after   diff     pct
      Ising     3.648   3.328   -0.32    -8.77%
   Flocking     2.568   2.508   -0.06    -2.34%
       Team     2.525   2.49    -0.035   -1.39%
Bureaucrats     2.415   2.385   -0.03    -1.24%
   GridWalk     5.207   5.153   -0.054   -1.04%
         BZ     5.342   5.302   -0.04    -0.75%
       Fire     0.175   0.174   -0.001   -0.57%
    FireBig     4.074   4.065   -0.009   -0.22%
  GasLabNew     4.559   4.559    0        0.00%
 PrefAttach     4.584   4.587    0.003    0.07%
       CA1D     4.548   4.551    0.003    0.07%
  GasLabOld     3.201   3.216    0.015    0.47%
   VirusNet     1.266   1.272    0.006    0.47%
   Termites     3.119   3.135    0.016    0.51%
 GasLabCirc     4.68    4.709    0.029    0.62%
     Wealth     3.688   3.711    0.023    0.62%
    Erosion     4.053   4.113    0.06     1.48%
ImportWorld     0.985   1.006    0.021    2.13%
       Life     5.137   5.339    0.202    3.93%
   Heatbugs     2.711   2.85     0.139    5.13%
       Wolf     3.747   4.022    0.275    7.34%
       Ants     4.358   4.863    0.505   11.59%

I am, however, skeptical of those results: two of the most impacted models, Heatbugs and Ants, have no agent births or deaths except in setup, and it doesn't seem like this could account for the difference. Those benchmarks, however, where obtained by only waiting for the first "good" result. I think I'm going to heed Seth's advice regarding #403 (comment), bring back the infinite loop, and let the benchmarks run overnight. Doing this for both the "before" and the "after" brings us to Saturday. Stay tuned...

[SethTisue]

you've addressed my other concerns, so just waiting on the new performance numbers now.

[nicolaspayette]

Here are new numbers, after running both branches over night:

             old    new    diff     pct
      Ising  3.74   3.609  -0.131  -3.50%
     Wealth  3.641  3.593  -0.048  -1.32%
   Flocking  2.493  2.478  -0.015  -0.60%
   Termites  3.078  3.06   -0.018  -0.58%
       CA1D  4.496  4.484  -0.012  -0.27%
   Heatbugs  2.771  2.77   -0.001  -0.04%
    FireBig  4.058  4.06    0.002   0.05%
   GridWalk  5.142  5.146   0.004   0.08%
ImportWorld  1      1.002   0.002   0.20%
       Wolf  3.651  3.659   0.008   0.22%
 GasLabCirc  4.604  4.616   0.012   0.26%
 PrefAttach  4.517  4.53    0.013   0.29%
  GasLabNew  4.47   4.485   0.015   0.34%
Bureaucrats  2.369  2.379   0.01    0.42%
       Fire  0.173  0.174   0.001   0.58%
       Team  2.474  2.49    0.016   0.65%
   VirusNet  1.248  1.259   0.011   0.88%
       Ants  4.243  4.284   0.041   0.97%
    Erosion  3.963  4.002   0.039   0.98%
  GasLabOld  3.155  3.198   0.043   1.36%
         BZ  5.227  5.309   0.082   1.57%
       Life  5.024  5.221   0.197   3.92%

Life is the only one that still seems negatively affected, but since it's a model that has no turtles/links births/deaths whatsoever, I would assume that we can chalk it up to a benchmarking artifact.

If no one has further concerns, I am requesting merging.

[SethTisue]

LGTM

[TheBizzle]

Why mark the trait as abstract? And why does the trait exist at all? If you want to parameterize Publisher over the type of your case object, you can just say Publisher[SimpleChangeEvent.type].

[nicolaspayette]

@TheBizzle: It seems like your comment on SimpleChangeEventPublisher went away when I pushed a new commit (?) (OK, I see it now...)

Anyway, here's my answers:

Why mark the trait as abstract?

An abstract trait is, of course, redundant. I probably started with an abstract class and forgot to remove abstract when later changing to a trait. (Or maybe I was momentarily confused... it has been known to happen to me...)

And why does the trait exist at all? If you want to parameterize Publisher over the type of your case object, you can just say Publisher[SimpleChangeEvent.type].

Well, seems like I just learned a new thing about Scala! Thanks! 1ad6dc9