Their complaint is that the resulting image is empty - most likely
because the error gets passed through and clears the image buffer. It
appears that the sequence of events is:
We send SINGLEFRAMEDONE.
- the image is displayable
We send STATICIMAGEDONE.
- filter throws NPE
- image is probably still displayable
We send ERROR
- if it gets passed through to the toolkit image consumer then we'll
clear the image buffer
- image is no longer displayable
It's hard to say for sure without having an instance of their filter
in-house for testing, but previously we weren't sending the STATICDONE
notice and the program was working just fine. Since the NPE comes from
their code when we send it, it shouldn't have affected any down-stream
consumers, so we should be OK with just continuing on and the image
should still be displayable just as if we hadn't sent the STATICDONE
notice in the first place.
Now, *during debugging*, they were thwarted by the fact that we ate the
exception, true, but the original issue is that the image wasn't
displayable at all. We might want to do both:
- not send ERROR for STATICDONE when we used to not send STATICDONE at all
- show the exception so that someone realizes that there is a problem,
even though we've made it not hurt their functionality.
Does that make sense?
...jim
On 5/25/2016 1:36 PM, Sergey Bylokhov wrote:
On 25.05.16 23:33, Jim Graham wrote:
How about option 3 -
NPE before imageComplete sends an ERROR as it does now.
NPE during imageComplete is ignored, both for backwards compatibility
and because it is more of a "hint" than an operation that requires
action from the consumer.
I guess the case is that when we ignore this NPE(w/o any notification)
the users complains that the bug is in jdk.
...jim
On 5/25/2016 1:31 AM, Ajit Ghaisas wrote:
Hi
Bug :
https://bugs.openjdk.java.net/browse/JDK-8139192
Custom ImageFilters return blank images in Java 8(.45) while working
in 7
Root cause :
private method produce() in OffScreenImageSource.java consumes a
NullPointerException that originates from a custom ImageConsumer (a 3^rd
party image library class - com.jhlabs.image.WaterFilter)
Analysis:
1. How the behavior changed between JDK7 and JK8 :
A call to imageComplete(ImageConsumer.SINGLEFRAMEDONE) was added in
addition to imageComplete(ImageConsumer. STATICIMAGEDONE) as a fix for
JDK-7143612.
2. What debugging revealed:
A NullPointerException is being thrown from the library during the call
to imageComplete(ImageConsumer.STATICIMAGEDONE)
3. It looks like the fix of JDK-7143612 is valid and successive calls to
imageComplete() are allowed.
4. The 3rd party library behavior appears incorrect (if it can not
handle subsequent calls to imageComplete(), it should de-register
itself).
The 3rd-party library code should be fixed.
Possible modifications in JDK :
Currently, OffScreenImageSource::produce() consumes the
NullPointerException - this is incorrect and results in silent failure
of this particular image filter.
We need to let the image filter library know that exception has occurred
in its code and not in JDK. We have two options -
Option 1 : Rethrow the NullPointerException --- It is the clearest way
to let 3^rd party library know that their code is erroneous with latest
JDK. This will change the 3^rd party image filter behavior that
generates blank image.
Option 2 : Add a stack trace where the exception is being consumed ---
Adding stack trace provides more information regarding failure of 3^rd
party image filter with retaining the current behavior that generates
blank image.
I have implemented both the options:
Option 1: http://cr.openjdk.java.net/~aghaisas/8139192/webrev.00/
Option 2: http://cr.openjdk.java.net/~aghaisas/8139192/webrev.01/
Request you to review both the webrevs and provide your preference.
I will add a test to the selected webrev.
Regards,
Ajit
