[Javapathfinder-devel] Re: Endless loops and POR?

[Peter C. M. <pcm...@em...>]

Yes, but before I do, I just want to think a few minutes about the 
livelock detection - it seems awfully close

-- Peter

On May 2, 2005, at 11:01 AM, Willem Visser wrote:
> I meant in POR transitions, but the same will happen in atomic steps.
>
> This is a pathological case I admit, but it is a problem - if someone
> writes a program with a meaningless endless loop, then even under a
> normal VM this will be bad. However if someone writes a meaningless
> endless loop in one thread and a bug in another thread then we'll never
> find the bug if the meaningless thread gets scheduled first.
>
> I think the counter would be good - it will fix this problem. But set 
> it
> real high so that we don't unnecessarily break por-steps.
>
>> -----Original Message-----
>> From: Peter C. Mehlitz [mailto:pcm...@em...]
>> Sent: Monday, May 02, 2005 10:49 AM
>> To: wv...@em...
>> Cc: jp...@em...
>> Subject: Re: Endless loops and POR?
>>
>> You mean POR transitions or atomic sections? You're right that we
>> currently have no "progress" check, but why wouldn't a normal VM loop
>> endlessly in this case? Or do you mean starvation, where another
> thread
>> would exit, but a daemon does endless loops w/o boundary steps so the
>> first one doesn't get scheduled? Right, that would be a problem
>>
>> should be easy to implement though, a simple counter should do. Can
>> actually be requested from outside (ThreadInfo.yield())
>>
>>> Do you know whether there is a flag/option to break loops in atomic
>>> steps? If not then we still have the bug that you could cycle
> forever
>>> in
>>> a loop that accesses no external variables and doesn't change state
> -
>>> no
>>> this is rare, but someone found it :-)
>