Weird New libjava Testsuite Failure (PR java/14104)

[Ranjit Mathew]

Andrew Haley wrote:
> Andrew Haley writes:
>  > Ranjit Mathew writes:
>  >  > 
>  >  > Any insights into the problem or at least how to
>  >  > try to reproduce it reliably?
>  > 
>  > I'll have a look.
> 
> Okay, so it's even weirder than I thought.  If I rename libjava.jar to
> libjava.0jar in order to make it run first in the testsuite, it
> doesn't fail either!
> 
> We have a real Heisenbug here...

Oh yes - and it does not seem dependent on the order
either - if I rename "libjava.jar" to say "libjava.jacks123"
so that it is still executed in the same order, it doesn't
fail - however, if I rename it back to "libjava.jar", it
duly fails! I am able to consistently reproduce this
over several runs. What the...


> FWIW, there is a way to catch this: install a SEGV signal handler that
> either sends the process SIGSTOP or goes into an infinite loop.  I'll
> do it.

Sorry, I didn't understand this fully. Are you trying to
bypass GCC's ICE reporting?

Ranjit.

-- 
Ranjit Mathew          Email: rmathew AT hotmail DOT com

Bangalore, INDIA.      Web: http://ranjitmathew.tripod.com/