Hi,

Tim found a bug and created a testcase for when two Tasks were
unfortunate enough to be simultaneously stepping the same instruction
using "reset stepping" at the exact same moment.
http://sourceware.org/bugzilla/show_bug.cgi?id=5082
This fortunately triggered a sanity check or really unpredictable
behavior would result. Unfortunately the sanity check was actually
wrong. When reset-stepping an instruction we keep all threads running,
temporarily remove a breakpoint instruction by putting the original
instruction back, step over it and put the breakpoint instruction back.
But if multiple Tasks are stepping the exact same instruction we should
wait till all have stepped before putting the breakpoint instruction
back.

There is a related bug in the out-of-line stepping case which is tracked
in http://sourceware.org/bugzilla/show_bug.cgi?id=5148 but for which
there is no testcase yet. It is much harder to trigger since we almost
never use out-of-line-stepping since we don't have a proper instruction
parser yet (http://sourceware.org/bugzilla/show_bug.cgi?id=4762).

frysk-core/frysk/proc/ChangeLog
2007-10-09  Mark Wielaard  <mwielaard@redhat.com>

    Fixes bug #5082
    * Breakpoint.java (stepping): Removed, state is Task specific.
    (reset_stepping_tasks): New field.
    (prepareStep): Check Instruction properties not stepping.
    Adjust reset_stepping_tasks and only reset when zero.
    (stepDone): Likewise.
    (stepAbort): Likewise.

frysk-core/frysk/hpd/ChangeLog
2007-10-09  Mark Wielaard  <mwielaard@redhat.com>

    * TestBreakpoints.java (testHpdBreakMultiThreadedContinue): Don't 
    mark unresolved bug #5082. Expect Task terminating messages.

This should fix all occurrences of IllegalStateException "Already
stepping" that sometimes popped up. It does make the window of
opportunity of missing a breakpoint a little bigger though for other
running Tasks.

Tim, I adjusted your testcase a little to catch the various Task X
terminating/terminated statements. I wasn't sure there were just extra
debug output or whether they should be explicitly matches. If I made the
wrong assumption please adjust it to how the fhpd should report these.

Cheers,

Mark