Well, I disagree. My point is that AWTEvents for internal components
should not be delivered to listeners that can be installed via public
API. If JDK developers need to trace those events, they could use some
logging facilities. Application developers do not need to handle those
events.
--
best regards,
Anthony
On 2/6/2014 4:24 PM, Sergey Bylokhov wrote:
On 06.02.2014 15:41, Anthony Petrov wrote:
Indeed, it wasn't an external developer who filed the original bug.
Thanks for the clarification.
Still, I don't see any reason to post events for components that
belong to our peers' implementations and shouldn't be accessible via
public API.
The reason is that we easily can check if someone flood the system. And
it is not contradict the specification.
I'd like to ask you again: does the original fix really break anyone?
Or is the only reason to back it out is the fact that we don't like
the implementation of the filtering mechanism (and I admit, I don't
like it too)? If it's the latter, I'd suggest you to actually _rework_
that fix as the synopsis of 8032187 suggests, rather than simply
remove it. Or leave it as is and keep the 8032187 bug open for now.
My point is that the previous fix is incorrect, because there was no bug
in jdk and the test should be updated instead.
--
best regards,
Anthony
On 2/6/2014 2:45 AM, Sergey Bylokhov wrote:
On 06.02.2014 1:26, Anthony Petrov wrote:
[ adding back the awt-dev@ mailing list ]
Hi Sergey,
This all boils down to whether we want to deliver events related to
our internal components to the user code or not. The fact that the
MACOSX_PORT-424 was filed by an external developer suggests that this
behavior may be unwanted. Also, common sense suggests that such events
are useless for external consumers, and hence sending them just
doesn't make any sense.
No, it is not an external developer. It was filed because some test,
which was provided by Apple, failed.
Could you please clarify again exactly what the problem is with the
original fix for this issue? I do not quite like it myself (e.g. when
I see "enableEvents(0xFFFFFFFF)" - this just looks fishy, and the
setting/unsetting the listener looks like a hack, I agree), but I
don't see any serious consequences that that fix could cause. Could
you please elaborate on that?
That's because this fix try to eliminate events during peers creation
only. This was a hack and the goal was to fix one particular test.
If we see that there's some serious flows with the original solution,
let's get rid of it. However, we'll need to file a new bug to deal
with all the unwanted events, and implement it some other way then.
Why they are unwanted? AWT fires AWTEvents, and toolkit have an ability
to catch all events, which are dispatched system-wide without any
exceptions. As far as I understand toolkit listeners is kind of debug
mechanism which can help to track all events.
--
best regards,
Anthony
On 2/5/2014 5:54 PM, Sergey Bylokhov wrote:
On 05.02.2014 16:34, Anthony Petrov wrote:
I agree that the listener should receive all the events for compound
controls. They make sense because all the parts of compound controls
are accessible via our public API, so the app could just assign
listeners to specific sub-components if it needed to.
Not necessary, for example for our text component in xawt. The user
should filter out unnecessary events anyway, because they can
belongs to
a different application or some internal components(like
SharedOwnerFrame).
However, this isn't true for our internal LWAWT components hierarchy.
This hierarchy is hidden from users (they can't access the underlying
Swing components via public API, and moreover, they shouldn't be able
to do so even if they want to). It is an implementation detail of the
LWAWT toolkit. And therefore, I don't see a reason to post events
related to this hidden hierarchy to a listener that can be installed
using the public API.
Can we find another way to filter the events out?
Only if we will add additional checks to the places, where
Toolkit.enabledOnToolkit is used.
--
best regards,
Anthony
On 2/5/2014 1:26 AM, Sergey Bylokhov wrote:
On 04.02.2014 23:06, Anthony Petrov wrote:
Hi Sergey,
From the bug report:
After discussion, it was decided to accept MACOSX_PORT-424 as
not a
defect
Are you saying that user code will start receiving hierarchy events
related to the internal hierarchy of our LWAWT peers? Why would
anyone
want to process these events and why would we want to post them to
user code? Is there any way to filter them out?
Documentation of Toolkit.addAWTEventListener() states:
* Adds an AWTEventListener to receive all AWTEvents dispatched
* system-wide that conform to the given
<code>eventMask</code>.
....
* Note: event listener use is not recommended for normal
* application use, but are intended solely to support special
* purpose facilities including support for accessibility,
* event record/playback, and diagnostic tracing.
And components intentionally cannot filter them out. All our non
trivial
compound components posts lots of such events.
While I agree that using reflection is a bad idea, but is there any
other real problem with the original fix for MACOSX_PORT-424? I.e.
does that fix break anything? If not, can we simply replace
usages of
reflection with AWTAccessor-like mechanism?
The problem not in reflection but in replacing global list of
listeners.
--
best regards,
Anthony
On 2/4/2014 3:52 PM, Sergey Bylokhov wrote:
Hello.
Please review the fix for jdk 9.
- Initial fix for MACOSX_PORT-424 was reverted back.
- delegate.addNotify(),because it was called from
delegateContainer.addNotify();
- Testcase was updated to filter out events not from the Frame:
84 if (e.getSource() instanceof Frame) {
85 counter++;
86 notify();
87 }
Bug: https://bugs.openjdk.java.net/browse/JDK-8032187
Webrev can be found at:
http://cr.openjdk.java.net/~serb/8032187/webrev.00
