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

[Michael Hall]

> On Nov 29, 2022, at 2:47 AM, Maxim Kartashev <mkartashev at openjdk.org> wrote:
> 
> FSEventStreamCreate

This is sounding somewhat similar to the kqueue watcher I attempted. It was suggested this was due to maxing out file descriptors. Which seems plausible. What I would see was that suddenly the kernel would cease sending events at all.

https://developer.apple.com/documentation/coreservices/1443980-fseventstreamcreate <https://developer.apple.com/documentation/coreservices/1443980-fseventstreamcreate>
The number of seconds the service should wait after hearing about an event from the kernel 

Note this indicates it is still based on receiving events from the kernel. The following was my code, it shows originally I did no polling at all but changed it to do polling thinking this was a thread deadlock issue. But with polling my thread continued to run, it was the kernel that totally stopped sending events. Again this seems potentially similar to what is being described for errors here. It occurred in trying mine on the LotsOfEvents test. Maybe there would be some way you could check for this situation? The errors all seem to be timeouts when things stop happening, or I suppose happen but are missed?  

-(void)		watcherThread: (id)sender
{
	int					n;
	int 				idle_cnt = 0;
	
	struct kevent64_s   ev;
    struct timespec     timeout = { 5, 0 }; // 5 seconds timeout.
	int					theFD = queueFD;	// So we don't have to risk accessing iVars when the thread is terminated.
	struct timespec		nullts = { 1, 0 };
		
    while( keepThreadRunning )
    {
		NSAutoreleasePool*  pool = [[NSAutoreleasePool alloc] init];
		NS_DURING
//			NSLog(@"going into kevent");
			n = kevent64( queueFD, NULL, 0, &ev, 1, 0, &timeout );
//			n = kevent64( queueFD, NULL, 0, &ev, 1, 0, &nullts );
//			n = kevent64( queueFD, NULL, 0, &ev, 1, 0, NULL);
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/nio-dev/attachments/20221129/53d9cca9/attachment.htm>