Re: SableVM on ARM Zaurus

Grzegorz B. Prokopski wrote:
> On Wed, 2005-12-01 at 16:28 -0500, Chris Pickett wrote:
> 
>>I have recently been getting segfaults with --enable-debugging-features 
>>and --with-profiling=yes turned on, but it just says the function name 
>>(e.g. _svmf_interpreter ()), without line number information.  Is there 
> 
> 
> Are you getting these segfaults even --without-profiling ?  I guess not.
> It might be interaction between gcc's profiling and inlined engine.
> Try using direct and with profiling.  If that solves the problem,
> I guess we should make direct the default with profiling.
> 
> 
>>something extra I need to do, such as use '-g', or do you guys get line 
>>number info in SableVM with segfaults like this?
> 
> 
> I don't think gcc's profiling was designed to give us stack traces on
> segfaults, but maybe there is some method.  I assume that you _do_ get
> the full profile output at the end, when there was no segfault?

I'm not profiling, I'm trying to debug threading issues.  I turned on 
profiling because I wasn't even getting line number information in debug 
mode (switch interpreter).  These bugs seem to appear whenever I slow 
down certain bits of the VM that aren't in critical sections.

I would guess that threading is also the source of recent problems and 
regressions (esp. since they're intermittent).  There were some changes 
to threading in AWT on the java-gui branch of gcc recently, that might 
fix things, but you have to wait for upstream to merge them.

All I care about at the moment is getting line numbers in debug mode 
with the switch interpreter.

Chris