Hi Daniil,
I have the same question and also suggestion to remove the referenced
debuggee fields dummyF and finDummyF as they have no other uses.
Also, a minor suggestion is to reorder the parameters of the
MockReferenceType
constructor the same as in the createObjectReference() method.
It looks more natural to me.
Hi Jerry,
We have to reply in the open.
I also tend to do this mistake.
Let's fix each other.
I've added the svc-dev list back.
Thanks,
Serguei
On 4/20/18 18:10, Gerald Thornbrugh wrote:
Hi Daniil,
Your changes look good but I have a small nit.
Since the code no longer uses the DEBUGGEE_METHODS[i][1] elements in the
array due to your change, should they be removed?
Thanks for doing this!
Jerry
On 04/20/2018 04:47 PM, Daniil Titov wrote:
Hello,
Could you please help with reviewing this change? I need 2 reviewers.
Thanks!
Best regards,
Daniil
On 4/19/18, 4:52 PM, "Daniil Titov" <daniil.x.ti...@oracle.com> wrote:
Please review the change that affects two tests:
-
fromTonga/nsk/jdi/ClassType/invokeMethod/invokemethod009/TestDescription.java
-
fromTonga/nsk/jdi/ObjectReference/invokeMethod/invokemethod006/TestDescription.java
Both tests try to emulate the case when the method with a
parameter is invoked in the target VM via JDWP while the class
corresponding to the parameter type is not yet loaded. The problem
with these tests is that when they invoke the method in the target VM
they pass null as an argument and in case of null
com.sun.tools.jdi.ValueImpl.prepareForAssignment(Value,
ValueContainer) method does a quick return without any additional check.
39 static ValueImpl prepareForAssignment(Value
value,
40 ValueContainer destination)
41 throws InvalidTypeException,
ClassNotLoadedException {
42 if (value == null) {
43 /*
44 * TO DO: Centralize JNI signature knowledge
45 */
46 if (destination.signature().length() == 1) {
47 throw new InvalidTypeException("Can't
set a primitive type to null");
48 }
49 return null; // no further checking or
conversion necessary
50 } else {
51 return
((ValueImpl)value).prepareForAssignmentTo(destination);
52 }
53 }
This behavior is consistent with the code existing in
open/src/jdk.jdi/share/classes/com/sun/tools/jdi/ObjectReferenceImpl.java,
open/src/jdk.jdi/share/classes/com/sun/tools/jdi/ClassTypeImpl.java,
and
open/src/jdk.jdi/share/classes/com/sun/tools/jdi/ArrayReferenceImpl.java
279 } catch (ClassNotLoadedException e) {
280 /*
281 * Since we got this exception,
282 * the field type must be a reference
type. The value
283 * we're trying to set is null, but if the
field's
284 * class has not yet been loaded through
the enclosing
285 * class loader, then setting to null is
essentially a
286 * no-op, and we should allow it without
an exception.
287 */
288 if (value != null) {
289 throw e;
290 }
291 }
292 }
The fix corrects the tests to construct a mock value object
with a required type and instead of null pass this mock value as an
argument in the method invocation in the target VM.
Bug: https://bugs.openjdk.java.net/browse/JDK-4698670
Webrev: http://javaweb.us.oracle.com/~datitov/4698670/webrev.01
Best regards,
Daniil
