Re: [rvm-research] Non-disturbing Segmentation Faults in Benchmarks

[Steve B. <Ste...@an...>]

Hi Jan,


On 27/03/2009, at 2:29 AM, Jan Sinschek wrote:
> 1) I have noticed that using the -X:verbose option shows segmentation
> faults at a number of tests in different versions of Jikes.

> All tests run without problems in spite of these faults. Is this
> behaviour intended?

Yes.  We use traps to handle certain exceptions (such as array bounds  
checks).  Exceptions are sometimes used idiomatically by programmers  
(eg for detecting overflow of an array and then growing that array).   
This is normal behavior.   The -X:verbose flag is there to allow you  
to see activity such as this.

> 2) A less important question:  my own machine, a virtualized Ubuntu  
> 8.10
> running on a IA-32 in Sun VirtualBox seems to be unable to perform I/ 
> O,
> failing with IOExceptions whenever a test needs data. I could be more
> specific on that is  nobody can tell of likewise behaviour. If have  
> set
> 777 access for all relevant directories that exist prior to the  
> test. An
> Ubuntu 8.04 installation set up to compare the behaviour worked fine
> (but I am not sure whether or not this was before or after installing
> updates.) I have not collected these data.

I've not heard of such a problem before :-/   Sounds like it may be  
related to your setup.   Have you tried running exactly the same test  
with a different JVM?

--Steve