Marlin renderer patches for jdk8u integration

[Laurent Bourgès]

Hi,
Does anybody reproduce this problem ?

Any feedback on this awt issue ? How can it be related to any java2d AA
renderer ?

Could we start the review, please ?

Happy new year,
Laurent

Le ven. 22 nov. 2019 à 21:14, Laurent Bourgès <bourges.laurent at gmail.com> a
écrit :

> Hi David,
>
> I tested on my windows 10 machine but the given MarlinTest looks the same
> either with Pisces or with latest Marlin 0.9.4 (github).
> I enabled tracing with -Dsun.java2d.trace=log
> -Dsun.java2d.renderer.trace=true but it shows no call to the Marlin
> renderer, only:
>
> OpenJDK Runtime Environment (AdoptOpenJDK)(build 1.8.0_232-b09)
> OpenJDK 64-Bit Server VM (AdoptOpenJDK)(build 25.232-b09, mixed mode)
>
> sun.java2d.pisces.PiscesRenderingEngine.getMinimumAAPenSize()
> D3DFillRect
> D3DFillRect
>
> echo "Original"
> java -Dsun.java2d.trace=log -Dsun.java2d.renderer.trace=true -cp .
> MarlinTest 11
>
> echo "Marlin 0.9.4"
> REM -Dsun.java2d.dpiaware=false
> java -Dsun.java2d.trace=log -Dsun.java2d.renderer.trace=true
> -Xbootclasspath/a:./marlin-0.9.4.2-Unsafe.jar
> -Xbootclasspath/p:./:marlin-0.9.4.2-Unsafe-sun-java2d.jar
> -Dsun.java2d.renderer=org.marlin.pisces.DMarlinRenderingEngine -cp .
> MarlinTest 11
>
> It seems Choice widgets are adjusted to the available window height, so it
> fills the space accordingly.
> According to me, it is not related to the Marlin renderer by this patch.
>
> PS: If anybody else has another opinion or test case, it would be great.
>
> Regards,
> Laurent
>
> Le jeu. 14 nov. 2019 à 13:54, Laurent Bourgès <bourges.laurent at gmail.com>
> a écrit :
>
>> Hi David,
>>
>> Thanks for testing the Marlin renderer patches, it looks quite good.
>>
>> Sorry for my late response, I expected to test myself on a windows
>> machine, but still not.
>>
>> I wonder why the 1px difference on windows awt widgets has not been yet
>> reported.
>> Phil or Sergey, are you aware of such issue ? How are awt widgets
>> rendered ? How can the java2d renderer interfere in such way ?
>> Is it related to fractional scaling factors ?
>>
>> I will try using awt/2d debug settings to get more details...
>>
>> Laurent
>>
>> Le ven. 8 nov. 2019 à 02:32, David Alvarez <alvdavi at amazon.com> a écrit :
>>
>>> Hi,
>>>
>>> We’ve run the TCK with Marlin enabled, certification passes. So far so
>>> good.
>>>
>>> Then, in some of our additional testing we noticed small differences in
>>> the rendering of some components.
>>>
>>> The simplest example we’ve been able to find is the java.awt.Choice in
>>> Windows 10. When using a JDK with the patches applied, this component
>>> will be rendered with one extra height pixel.
>>>
>>> I've uploaded some images for comparison. Here[1] is how the component
>>> is displayed using 8u232, and here[2], how it looks when using a version
>>> of 8u232 that contains your patches. The difference is noticeable when
>>> we put the components side by side[3], more so if we zoom in[4].
>>>
>>> This can also cause a one-pixel overlap when these components are
>>> rendered on a vertical box layout[5].
>>>
>>> The behavior persisted even when changing the rendering engine back to
>>> Pisces, so it seems it is caused by changes outside Marlin, most likely
>>> on AAShapePipeline.
>>>
>>> The new behavior is consistent with JDK11, but it still worries me a
>>> little. Any idea why this is happening? If you want, I can provide you
>>> with a sample program and compared images.
>>>
>>> For reference, I've also updated the code used during the images[6]
>>>
>>> --
>>> David Alvarez
>>>
>>> [1] http://cr.openjdk.java.net/~alvdavi/emails/marlin/vanilla.png
>>> [2] http://cr.openjdk.java.net/~alvdavi/emails/marlin/marlin.png
>>> [3] http://cr.openjdk.java.net/~alvdavi/emails/marlin/comparison.png
>>> [4] http://cr.openjdk.java.net/~alvdavi/emails/marlin/detail.png
>>> [5] http://cr.openjdk.java.net/~alvdavi/emails/marlin/multiple.png
>>> [6] http://cr.openjdk.java.net/~alvdavi/emails/marlin/MarlinTest.java
>>>
>>>
>>>
>>> On 2019-11-05 02:40, Laurent Bourgès wrote:
>>> > Hi Paul
>>> >
>>> > Le lun. 4 nov. 2019 à 20:30, Hohensee, Paul <hohensee at amazon.com
>>> > <mailto:hohensee at amazon.com>> a écrit :
>>> >
>>> >     Also, we (Amazon) would like the Marlin renderer enabled by
>>> default,
>>> >     just as it is for Zulu. It's been well-tested starting in 9, plus I
>>> >     doubt it'll get much, if any, use in u242 if we e.g. disable it for
>>> >     u242 and then enable it for u252.
>>> >
>>> >     Please put the patches on cr.openjdk.java.net
>>> >     <http://cr.openjdk.java.net> so we can go through a normal review
>>> >     process.
>>> >
>>> >
>>> > I uploaded all patch files (m01 to m21):
>>> > http://cr.openjdk.java.net/~lbourges/marlin-jdk8/marlin-jdk8.0/
>>> >
>>> > Laurent
>>>
>>>