Sorry for not having a chance to read through draft sooner
(out+swamped). I like the spirit of Fred's suggestions --
especially killing ManagedObject etc. I'm not sure they
all fit together when done out, but I'm sure Anthony will
figure this out for us :-)
My main attitude-only comment is that in the 6 years since
all this first started, j.u.c has moved toward a much stronger
focus on very lightweight tasks (ForkJoin, actors on top of
work-stealing etc) for which we offer LESS control than seen
in jsr236 spec. For example, the main new executor, ForkJoinPool,
has only one tuning parameter -- the target parallelism level
(which normally means the target number of threads in pool,
which in turn is usually taken as default #cores on machine).
This is in part because other aspects of tuning are
pretty much useless when you are running 100million tasks/sec.
While the coarser-grained EE usages might still benefit from the
customization offered by ThreadPoolExecutor (as presented
as ManagedExecutor), I'm not positive of that.
-Doug