[OpenJDK 2D-Dev] [13] RFR 8214579: JFrame does not paint content in XVFB / X11vnc environment
Dmitry Markov
dmitry.markov at oracle.com
Tue Jan 29 17:36:52 UTC 2019
Hello Ichiroh,
Definitely the problem occurrence depends on depth, (i.e. pixel size) settings. Actually such behaviour looks quite expected because the current implementation of XRSurfaceData.getSurfaceType() always returns INT (32-bit) surface type which most likely won't work properly for the configuration where pixel size is 16. I think we can use X11SurfaceData.getSurfaceType() as a fallback option inside XRender if the depth is 16 or less. I’ll try to implement such approach.
Regarding the usage of XVNC4 with depth 32. I didn’t use such configuration. Based on your description I suspect the problem you observed, (i.e. infinity loop) and 8214109 are different. 8214109 is reproducible on XVNC4 with depth 16. The detailed steps to replicate the issue are provided in the bug report. Possible fix and related discussion are located at http://mail.openjdk.java.net/pipermail/awt-dev/2019-January/014949.html <http://mail.openjdk.java.net/pipermail/awt-dev/2019-January/014949.html>
I believe it will make sense to file a new bug for infinity loop issue you observed.
Thanks,
Dmitry
> On 29 Jan 2019, at 10:05, Ichiroh Takiguchi <takiguc at linux.vnet.ibm.com> wrote:
>
> Hello Dmitry.
>
> Thank you for your explanation.
>
> I could recreate your issue with XVFB/X11VNC on Ubuntu 18.04.
> I tested sun.java2d.xrender system property also.
> Since this Xserver had many visuals, I tested depth 32 visual by using FORCEDEFVIS environment variable.
> It worked fine, so it seems it may be depth 16 related issue as you said.
>
> When I used vnc4server package on Ubuntu 18.04 with depth 32,
> window did not come up.
> It seemed infinity loop was happened by following codes:
>
> src/java.desktop/unix/native/common/awt/X11Color.c
> (It seemed "i" was "-1" on line 447)
> ======
> 447 awt_data->awtImage->clrdata.bScale = 0;
> 448 while (i < 0x80) {
> 449 awt_data->awtImage->clrdata.bScale++;
> 450 i <<= 1;
> 451 }
> ======
>
> In case of XVFB/X11VNC:
> visual:
> visual id: 0x41
> class: TrueColor
> depth: 32 planes
> available colormap entries: 256 per subfield <<<====
> red, green, blue masks: 0xff0000, 0xff00, 0xff
> significant bits in color specification: 8 bits
>
> In case of vnc4server:
> visual:
> visual id: 0x24
> class: TrueColor
> depth: 32 planes
> available colormap entries: 2048 per subfield <<<====
> red, green, blue masks: 0x7ff, 0x3ff800, 0xffc00000
> significant bits in color specification: 8 bits
>
> It may be another issue, you handled 8214109 [1].
> Is it same kind of color related issue ?
>
> [1] https://bugs.openjdk.java.net/browse/JDK-8214109
>
> Thanks,
> Ichiroh Takiguchi
>
> On 2019-01-26 00:14, Dmitry Markov wrote:
>> Hi Ichiroh,
>> That’s right, the problem is related to XRender. It does not happen
>> if XRender pipeline is disabled.
>> As far as I know the problem is not specific for Ubuntu 18.10. I think
>> it takes place on other Linux platforms as well. Currently I can
>> reproduce the problem only in XVFB/X11VNC environment, (i.e. you
>> should run X11VNC over XVFB and depth parameter for VNC server must be
>> set to 16). Alternatively you can use docker image (link is located at
>> bug description) which contains necessary configuration to reproduce
>> the problem.
>> Thanks,
>> Dmitry
>>> On 25 Jan 2019, at 13:33, Ichiroh Takiguchi <takiguc at linux.vnet.ibm.com> wrote:
>>> Hello.
>>> I think if it's related xrender, could you try -Dsun.java2d.xrender=false ?
>>> I could not recreate your issue, so I could not test it by myself.
>>> It only happens on Ubuntu 18.10 ?
>>> Please give me more detail configuration.
>>> Thanks,
>>> Ichiroh Takiguchi
>>> On 2019-01-25 20:04, Dmitry Markov wrote:
>>>> Hi Sergey,
>>>> Thank you for the suggestion. I tested the patch proposed in that
>>>> email thread. Unfortunately the problem is still reproducible on the
>>>> build with the fix for JDK-8212677.
>>>> Thanks,
>>>> Dmitry
>>>>> On 24 Jan 2019, at 16:05, Sergey Bylokhov <sergey.bylokhov at oracle.com> wrote:
>>>>> Hi, Dmitry.
>>>>> Can you please check your test on top of this fix?
>>>>> https://mail.openjdk.java.net/pipermail/awt-dev/2019-January/014940.html
>>>>> The description of the bug looks similar to this one.
>>>>> On 23/01/2019 03:23, Dmitry Markov wrote:
>>>>>> Hi Phil,
>>>>>> I have found that the problem occurrence depends on depth setting of VNC server. XVFB/X11VNC configuration (where the problem takes place) uses 16 bpp and at the same time I cannot reproduce the issue on X11VNC with 32 bpp. In other words the issue depends on pixel size used by the configuration: it takes place if the pixel size is 16 and does not happen if the pixel size is 24. I am not an expert in XRender but the following seems correct: current implementation of XRSurfaceData.getSurfaceType() always returns INT (32-bit) surface type which might not work properly for the configuration where pixel size is 16.
>>>>>> Also the problem is not reproducible on XVNC4 (default depth value is 16) because XRender pipeline cannot be enabled there for some reasons. That may explain why the issue is not observed on other XVNC configurations. The root cause of XRender pipeline failure for XVNC4 is currently unclear but I think it is out of scope for this bug.
>>>>>> Thanks,
>>>>>> Dmitry
>>>>>>> On 8 Jan 2019, at 18:44, Phil Race <philip.race at oracle.com <mailto:philip.race at oracle.com>> wrote:
>>>>>>> I don't really understand why this only affects XVFB/X11VNC ?
>>>>>>> The bug evaluation is vague in explaining the root cause.
>>>>>>> What are they doing that is different ?
>>>>>>> Is there an unexpected alpha channel ?
>>>>>>> If so,
>>>>>>> - are we then selecting a loop which is supplying a zero value alpha
>>>>>>> channel instead of an opaque one ?
>>>>>>> - why is it only for X11VNC ?
>>>>>>> - Why was this not seen on Solaris ? Most if not all testing there uses Xvnc.
>>>>>>> -phil.
>>>>>>> On 1/8/19 2:24 AM, Dmitry Markov wrote:
>>>>>>>> Hi Sergey,
>>>>>>>> We started using XRSurfaceType (surface type from XRSurfaceData) after integration of JDK-8204931 [1]. Before that fix getSurfaceType() was not overridden in XRGraphicsConfig and surface type from X11GraphicsConfig/X11SurfaceData, (i.e. X11SurfaceType) was used for XRWindowSurfaceData and XRPixmapSurfaceData. If I got it right JDK-8204931was intended for fixing problems with XRPixmapSurfaceData and it solved them by introducing surface type with PixelConverter in XRSurfaceData and overriding getSurfaceType() in XRGraphicsConfig. These changes are correct for XRPixmapSurfaceData but they accidentally broke XRWindowSurfaceData and caused this problem.
>>>>>>>> In proposed fix I restored the previous behaviour for XRWindowSurfaceData, (i.e. use surface type from X11SurfaceData instead there one from XRSurfaceData).
>>>>>>>> Thanks,
>>>>>>>> Dmitry
>>>>>>>> [1] - https://bugs.openjdk.java.net/browse/JDK-8204931
>>>>>>>>> On 7 Jan 2019, at 23:14, Sergey Bylokhov <Sergey.Bylokhov at oracle.com <mailto:Sergey.Bylokhov at oracle.com>> wrote:
>>>>>>>>> Hi, Dmitry.
>>>>>>>>> On 03/01/2019 10:29, Dmitry Markov wrote:
>>>>>>>>>> Fix:
>>>>>>>>>> It is necessary to use X11SurfaceType instead of XRSurfaceType inside createData() for XRWindowSurfaceData
>>>>>>>>> Can you please provide some more details why it is necessary? From the
>>>>>>>>> first point of view the XRSurfaceType should be used for XRWindowSurfaceData,
>>>>>>>>> because all this code is implementation of the XRender pipeline.
>>>>>>>>> --
>>>>>>>>> Best regards, Sergey.
>>>>> --
>>>>> Best regards, Sergey.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.java.net/pipermail/2d-dev/attachments/20190129/94add7c0/attachment.html>
More information about the 2d-dev
mailing list