Hi,

This solves a long standing bug (#5234) where a TaskObserver.Code was
triggered by a breakpoint being hit on a Task that it didn't originally
register with.

The problem was that the BreakpointAddresses didn't have the original
Task against which the TaskObserver.Code was registered. And originally
it was thought to not be such a big deal if a TaskObserver.Code
interface saw all breakpoint events in a Proc. It could check that it
was given the right Task in its updateHit() anyway. But this required
some ugly workarounds in some cases which are now not necessary anymore.

We now have a little static final helper class CodeObserver that records
the original Task and TaskObserver.Code so that notifyCodeObserver() can
check whether the Task (and observer) were registered for this address.
Note that it still must report zero, and not -1 if there are any
breakpoints installed, even if it wasn't for this particular Task
because the ptrace state machine depends on that to know whether or not
to go past an installed breakpoint on the current address. The patch
also marks the BreakpointAddresses class and (helper) methods as package
private.

frysk-core/frysk/proc/live/ChangeLog
2008-04-21  Mark Wielaard  <mwielaard@redhat.com>

    * BreakpointAddresses.java: Mark class and methods package private.
    (CodeObserver): New final static helper class.
    (addBreakpoint): Take CodeObserver as argument, not
    TaskObserver.Code.
    (removeBreakpoint): Likewise.
    (getCodeObservers): Return collection of CodeObservers, not
    TaskObserver.Code.
    * LinuxPtraceProc.java (BreakpointAction.run): Install or remove
    BreakpointAddresses.CodeObserver instead of bare TaskObserver.Code.
    * LinuxPtraceTask.java (notifyCodeBreakpoint): Check Task against
    BreakpointAddresses.CodeObserver.
    * TestTaskObserverCode.java (testMultiTaskUpdateCalledSeveralTimes):
    Resolved bug #5234.

The testcase testMultiTaskUpdateCalledSeveralTimes now PASSes.
No regressions.

Committed and pushed,

Mark