Re: Deadlock on performSelectorOnMainThread:

[Jonathan Gillaspie]

Thanks Richard,

We appreciate the fix -- and you are spot-on.  We definitely had at least one 
case where we were flooding this channel which would account for the problem, 
but we also had situations where we had only made a few calls (or maybe a few 
hundred calls) to that method before it locked up, so I think it's possible 
there might have also been some other condition that could trigger it, but 
hopefully it's resolved now.

> learly this shouldn't happen, and I've tried changing the code to avoid this 
> by (as you suggest) making the write non-blocking.
> 
> But ...
> The write sends a single byte to the pipe ... and the pipe buffer is 
> typically 8 or 16 KB in size ... so for the buffer to be full, the threads 
> calling the method are presumably doing so faster than the main thread can 
> handle them.
> That might mean there is a bug in the application too.
> 
> 

--
Jonathan Gillaspie
address@hidden