RTS -I switch

[Simon Marlow]

Bulat Ziganshin wrote:
> Hello Simon,
> 
> Tuesday, November 14, 2006, 1:19:45 PM, you wrote:
> 
> 
>>>i still forget to ask: if my program is idle during long time, is GC
>>>will occur each 0.3 seconds or GHC is smart enough to check that no
>>>much memory was allocated after last GC?
> 
> 
>>You'll only get one GC.  Or at least that's the theory: I remember seeing cases
>>where one GC caused a finalizer to be created, which in turn triggered another
>>idle GC (because the heap had changed), which in turn created a new finalizer,
>>and so on.  If you see something like this, we should investigate.
> 
> 
> well, how about this scenario for GUI program: user thinks and GC
> occurs, then user performs some simple action (say, move cursor to
> another line) and thinks again. new GC will occur at this moment,
> right? and counting that in my program major GC could be up to several
> seconds, this may make program very unresponsive
> 
> well, i don't tried this, just imagined. may be better strategy for
> automatic GC invocation will be checking ratio of used memory/memory
> used after last GC and perform automatic GC, say, only when this ratio
> grows to value more than 1.2?
> 
> (yes, i know that autoGCing can be just turned off)

Yes, that's another strategy we could use.  The main reason I added the idle GC 
was to catch deadlock in fact: deadlock can only be caught by a major GC, and 
many of the tests rely on catching deadlock.  Of course, if the idle GC is 
inconvenient, then you should turn it off.

Cheers,
	Simon