<Swing Dev> [OpenJDK 2D-Dev]  Review Request JDK - 8164811 : [hidpi]Tests fail with OpenGL Rendering
pankaj.b.bansal at oracle.com
Mon Nov 6 12:23:30 UTC 2017
<< Adding new API to BufferedImage was definitely NOT the way to approach this.
Yes, I understand now. I have to add a API in TranslucantPainter.
<<Why are none of the failing tests referenced with an updated reg-test @bug tag to indicate this fix corrects them ?
I have updated the webrev for this.
<<You say it is observable with GDI .. the bug report refers only to OpenGL. So is it both ?
Yes, the issue is reproducible with both GDI and OpenGL. The D3D one was fixed in http://hg.openjdk.java.net/jdk9/client/jdk/rev/14918637b76e.
<<So does it reproduce with just -Dsun.java2d.d3d=false
Yes the issue is reproducible by setting uiScale > 1.0 and -Dsun.java2d.d3d=false. The issue is also reproducible by just setting uiScale without -Dsun.java2d.d3d=false, as it runs the GDI because D3D is disabled on Intel cards as mentioned in D3DBadHardware.h.
I tested this on a machine which supports D3D pipeline. It fails with uiScale > 1.0 with -Dsun.java2d.d3d=false. With just scale being set, it passes as D3D is used as default pipeline on this machine.
From: Philip Race
Sent: Saturday, November 4, 2017 1:23 AM
To: Pankaj Bansal
Cc: Sergey Bylokhov; Prem Balakrishnan; swing-dev at openjdk.java.net; 2d-dev at openjdk.java.net
Subject: Re: [OpenJDK 2D-Dev]  Review Request JDK - 8164811 : [hidpi]Tests fail with OpenGL Rendering
Adding new API to BufferedImage was definitely NOT the way to approach this.
Adding new public API we need to support forever to fix an internal implementation bug on one platform in a very atypical scenario is the wrong trade-off.
And anyway as Sergey said BI should have no dependency or relationship on a BI
TranslucentWindowPainter already had the GraphicsConfiguration anyway so it does not seem necessary and you seem to be more along the right lines with this fix but I'd like to try it ..
Couple of questions :
Why are none of the failing tests referenced with an updated reg-test @bug tag to indicate this fix corrects them ?
You say it is observable with GDI .. the bug report refers only to OpenGL.
So is it both ?
If so you don't need to enable the flaky OGL pipeline to verify this .. but then I am left puzzled since Intel cards don't use D3D so I'd expect more reports.
So does it reproduce with just
On 11/3/17, 6:44 AM, Pankaj Bansal wrote:
> Hi Sergey,
> I have created the BufferedImage of correct size considering the window dimension and graphics context scale.
> I changed BufferedImage just to store graphicsConfig, so that it can be used by BufImgSurfaceManager while creating BufImgSurfaceData. The scale in BufImgSurfaceData is used to create config while creating the SunGraphics2D through getDefaultTransform function. Now when the TranslucentWindowPainter create scaled BufferedImage and then creates Graphics2D using that, the Graphics2D has transform scale as (1,1) as the BufImgSurfaceData has scale (1,1). Then this Graphics2D is used to update the Window in updateWindow in TranslucentWindowPainter. As the transform is identity, none of the Swing Components which are added to the window are scaled.
> So I modified the BufferedImage to store graphicsConfig which can then be used by BugImgSurfaceManager to create BugImgSurfaceData with correct scale values and hence correct transform value in SunGraphics2D.
> If this is not the correct way, I have one more way of doing this by transforming the SunGraphics2D object with transform of peer. So the SunGraphics2D will have correct transform and objects are properly scaled. I have updated the webrev for the same. Please have a look. Also if this is also not the correct way, can you please suggest a way I can do this?
> Pankaj Bansal
> -----Original Message-----
> From: Sergey Bylokhov
> Sent: Friday, November 3, 2017 5:34 AM
> To: Pankaj Bansal; Prem Kumar Balakrishnan
> (prem.balakrishnan at oracle.com); swing-dev at openjdk.java.net;
> 2d-dev at openjdk.java.net
> Subject: Re:  Review Request JDK - 8164811 : [hidpi]Tests fail
> with OpenGL Rendering
> Hi, Pankaj.
> The BufferedImage class is a raster in the memory which should be unrelated to any grpahics config on the system. So this class should not be changed, but instead you should create a correct BI-as a back buffer which takes into account the size of the window and the scale of the GC.
> On 31/10/2017 04:59, Pankaj Bansal wrote:
>> Hi All,
>> Please review the fix for JDK 10.
>> The tests given in the bug were failing when run with OpenGL and GDI.
>> This bug is related to
>> which states that that HIDPI does not work with swing components when
>> Translucent window is used. Because of which all the tests in the bug
>> were failing.
>> The TranslucentWindowPainter class was creating BufferedImage for
>> OpenGL (when /forceOpt/ is false) and GDI pipeline, but it is not
>> considering the device HiDPI scale. There is no way to create a
>> scaled BufferedImage because of which the scale value in
>> BufImgSurfaceData is always 1. Made changes to store graphics config
>> in Buffered image, so that the BufImgSurfaceManager can create BufImgSurfaceData with scale set properly.
>> This fix also fixes https://bugs.openjdk.java.net/browse/JDK-8189257
>> Pankaj Bansal
> Best regards, Sergey.
More information about the swing-dev