RFR: 8293067: (fs) Implement WatchService using system library (macOS) [v8]

[Michael Hall]

>> 
> 
> This still doesn’t appear to show local errors for me. I ran a modified LotsOfEvents 5 times which should amount to actually 25 test runs including concurrently. Without an error.
> 
>     

Since timeouts were previously mentioned, for my modified test of 5 LotsOfEvents For overflow polling uses slightly less elapsed millis than FSEvents averaging 32165.8 to 32227.4.
But for the queuing test it is noticeably worse, 39012.2 to 25067.4

> newov <- c(32256,32258,32216,32216,32191)
> newq <- c(25072,25074,25064,25064,25063)
> mean(newov)
[1] 32227.4
> mean(newq)
[1] 25067.4
> oldov <- c(32180,32180,32167,32167,32135)
> oldq <- c(39014,39014,39010,39010,39013)
> mean(oldov)
[1] 32165.8
> mean(oldq)
[1] 39012.2

(jconsole didn’t show the gc difference that the bitcoin monitor did, no significant differences were apparent, although polling still showed slightly higher CPU)