Re: RFR: JDK-8218754: JDK-8068225 regression in JDIBreakpointTest

[Gary Adams]

On 2/12/19, 7:14 AM, David Holmes wrote:

On 12/02/2019 10:11 pm, Gary Adams wrote:

Yes, see the revised webrev, we do have to guard against
multiple calls to dispose, e.g. catch and ignore VMDisconnectException.
We don't need to guard against a null vm. That would only exist if
the vm was never initialized and we were shutting down.

Okay. Revised webrev is better in that regard.

One more round to add the import for VMDisconnectedException.

  Webrev: http://cr.openjdk.java.net/~gadams/8218754/webrev.01/

There is a race condition between the debuggee and the debugger process.
In the original endDebugee, it attempted to do 2 things in parallel.
By calling dispose then waitFor, in most cases the debugee would finish
and report status before all the dispose operations completed.

But some tests would fail if the dispose happened quicker than the debuggee

could report final exit status.

The tests based on JDIBreakpointTest on  the other hand don't really

care about the the final exit status. Once all the events have been seen and handled, the test wants to shutdown the session.

I'm still not seeing how JDIBreakpointTest started hanging/timing-out after you switched the order of dispose and waitFor. Does dispose affect the debuggee process or the debugger process?

The JDIBreakpointTest would timeout on the endDebugee call to waitFor.
There was no reason for the debugee to exit at that point.
The call to dispose, provided the debugee with it's reason to exit.
Not sure which operation induces the debugee to exit.

Thanks,
David

On 2/12/19, 6:59 AM, David Holmes wrote:

Hi Gary,

On 12/02/2019 8:08 pm, gary.ad...@oracle.com wrote:

The recent change to JDK-8068225 changed the order of operations
in Debugee.endDebugee() to wait for the debugee to exit before
disposing of the vm on the debugger side of the connection.
For the tests based on JDIBreakpointTest the debuggee exit
status is not used and the tests relied on the
debugger side dispose operation to end the test.

Since JDIBreakpointTest already includes a call to wait for
the debugee, if does not need to use endDebuggee()
to dispose and wait for the debugee to finish.

I agree that potentially calling waitFor twice seems pointless. But how did the reordering of vm.dispose() and waitFor() cause all these tests to hang if they were waiting anyway? Does vm.dispose() have an effect on destroying the process?

Also what concerns me is that dispose() is not resilient the way that endDebuggee is:

 public void dispose() {
    vm.dispose();
 }

versus

    public int endDebugee() {
        if (vm != null) {
            try {
                vm.dispose();
            } catch (VMDisconnectedException ignore) {
            }
            vm = null;
        }

Do we need to be concerned with a null VM or getting VMDisconnectedException?

Thanks,
David

Testing in progress. The vm/mlvm tests are included in tiers 2, 3 and 6.

diff --git a/test/hotspot/jtreg/vmTestbase/vm/mlvm/share/jdi/JDIBreakpointTest.java b/test/hotspot/jtreg/vmTestbase/vm/mlvm/share/jdi/JDIBreakpointTest.java --- a/test/hotspot/jtreg/vmTestbase/vm/mlvm/share/jdi/JDIBreakpointTest.java +++ b/test/hotspot/jtreg/vmTestbase/vm/mlvm/share/jdi/JDIBreakpointTest.java

@@ -1,5 +1,5 @@
  /*

- * Copyright (c) 2011, 2018, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2011, 2019, Oracle and/or its affiliates. All rights reserved.

   * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   *

* This code is free software; you can redistribute it and/or modify it

@@ -359,7 +359,7 @@
          }.go();

          if (!debuggee.terminated())
-            debuggee.endDebugee();
+            debuggee.dispose();

          debuggee.waitFor();
          return true;