JDWP errors cause "old" VMs to crash

[stephan]

Inspecting various crashes that specifically occurred on MacOS (see OtdtOnMac), it turns out that our debugger may created JDWP requests, that on modern VMs fail with a proper error code, whereas older VMs (1.5) seem to crash (in native code) in these situations.

Thus we should clean-up the debugger to strictly avoid any erroneous requests, rather than hoping to catch resulting exceptions.

[stephan]

r21657 solves a first instance of this issue:

1.5 JVM could not correctly place method entry breakpoint:

• using a JavaMethodEntryBreakpoint on o.o.Team.finalize() the VM actually placed this breakpoint on j.l.Object.finalize().
• thus every Object.finalize() incorrectly triggered the TeamBreakpointListener. Somewhere downstream JDIStackFrame.getUnderlyingThisObject() created an illegal JDWP request, causing the JVM to abruptly die, where a 1.6 JVM would gracefully report a JDWP error (not sure what the exact error would be).

Instead of JavaMethodEntryBreakpoint we're now using a JavaMethodBreakpoint with entry==true. I don't see the big difference, but the new version works on 1.5 JVMs (so far tested on Linux-GTK only).

Also pass a valid linenumber (extracted from Team.java by the script genIOOTBreakpoints).

[stephan]

already r21619 solved a JDWP issue. No further issues known at this time.