RE: Regression testing on GCJ 3.3

[Øyvind Harboe <oyvind dot harboe at zylin dot com>]

> > This happens under GCJ 3.2. I can't yet test in GCJ 3.3, because it 
> > crashes for other reasons which are being followed up on.
> 
> Can you please write a real testcase that compiles everywhere out of 
> the box ?

I'd like to, but:

I have not yet gotten my dirty little mittens on a machine where
the problem happens(or rather I was only able to borrow the computer for
an hour). It is fairly rare, and I've tested numerous 
Windows 98/NT/2000/XP configurations.

Even if I do manage to put together a testcase, it should be tweaked
such that it happens on *all* machines if at all possible. This could
be a latent bug on all configurations. It just happens to manifest 
itself on all the time on particular computers.


I posted the info to draw upon any battlefield experience you guys
might posess. 



My personal, completely unsubstantiated, theory is that this is somehow
related to the virtual memory system and the garbage collector. I've
seen mentions of the VirtualAlloc() Win32 function in conjunction with
the garbage collector and the early work on Win32 JNI work. Memory
could be freed and then paged out on the particular machine it is
crashing
on.


I've got a personal grudge against this bug. It has caused serious
damage
to the perception of Java with my client(and some rather heated
outbursts).

Don't misunderstand me, GCJ is great! Without it, Java wouldn't even be
considered for the project I'm working on.

Øyvind