[RFA/JVMTI] SingleStep notification

[Keith Seitz]

Tom Tromey wrote:

> Keith> I share your reservations about the performance hit this entails. So
> Keith> perhaps a little tweaking is in order. Perhaps the "debug" interpreter
> Keith> should be extended to include not only JDWP but also JVMTI?
> 
> My view is that we're only writing this for the sake of JDWP.
> But perhaps that's just my first reaction :)
> Any idea how much this slows things down?

Okay, I think that I have that tweak figured out. I was right: something 
was using _Jv_InterpMethod::run instead of _Jv_InterpMethod::run_debug, 
and we had the two versions of NEXT_INSN confusing the interpreter. 
[_Jv_CompileMethod didn't take into account debug vs non-debug.]

So I have done what I mentioned above: We presume that when any JVMTI 
environment is created, we will use the debug stuff from there on out. 
There's no going back. If no JVMTI environment is ever created, it uses 
the code the faster code path, skipping JVMTI altogether.

Any less offensive?

Keith

ChangeLog
2007-01-29  Keith Seitz  <keiths@redhat.com>

         * include/jvmti-int.h (JVMTI): Declare member "enabled".
         * jvmti.cc (JVMTI): Add member "enabled".
         (_Jv_GetJVMTIEnv): Mark JVMTI enabled.
         * interpret.cc (_Jv_InterpMethod::ncode): Use JVMTI::enabled
         instead of gnu::classpath::jdwp::Jdwp::isDebugging.
         (_Jv_CompileMethod): If JVMTI is enabled, use run_debug
         instead of run to compile the method.
         * interpret-run.cc [DEBUG] (NEXT_INSN): Add JVMTI single step
         notification.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: jvmti-singlestep2.patch
Type: text/x-patch
Size: 4678 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/java-patches/attachments/20070130/c4e859cc/attachment.bin>