Small fix in Extensions3DCache.cpp prevents crashes in WPEWebProcess

[Gavriliuk]

backtrace.txt

[zdobersek]

@magomez should weigh in. GLContext::sharingContext() shouldn't be returning null.

[magomez]

Indeed. GLContext::sharingContext() should never return null. The problem has to be in the context creation. @Gavriliuk, where are you seeing this crash? RPi? which page?

[Gavriliuk]

Hi people,

On Mon, Dec 12, 2016 at 11:12 AM, Miguel Gómez ***@***.***> wrote: Indeed. GLContext::sharingContext() should never return null. The problem has to be in the context creation. @Gavriliuk <https://github.com/Gavriliuk>, where are you seeing this crash? RPi? which page?

I do not remember in details but maybe it was from WPELauncher with http://youtube/tv P.S. Abbreviation "RPi" is not understandable ( http://www.abbreviations.com/RPi)

[Gavriliuk]

@magomez <https://github.com/magomez> there was on Arris XG1v3 (look at https://github.com/Metrological/WebKitForWayland/files/645483/backtrace.txt)

[magomez]

RPi refers to the Raspberry pi, the board we use for reference ;)

I see the backtrace but, as a I mentioned, GLContext::sharingContext() cannot return null, that's why it's not currently checked in that part of the code. The problem has to be happening in the context creation, which is probably failing, and that's where the null comes from.
But if the context creation fails, nothing can be rendered.

I don't have access to your hardware, so I cannot investigate it, but if the bug is reproducible and you are interested, you should start by checking the return value of GLContextEGL::createPbufferContext() in GLContextEGL.cpp.

[psivasubramanian]

@magomez I'm seeing this issue on Dragonboard410c and this pull fixes it.

[magomez]

@psivasubramanian, as I explained in my last comment, this pull request may workaround the problem, but it's not a proper fix. GLContext::sharingContext() cannot return null because if there's no sharingContext that call will create one. If it's returning null it means that the call to create the sharingContext is failing, and that has to be checked. It's probably some issue with the EGL layer. You can start the debugging checking for EGL errors in GLContextEGL::createPbufferContext() in GLContextEGL.cpp. I cannot help you more without having access to your hardware and being able to debug there.

BTW, this pull request has an important side effect: Extensions3DCache will try to get the available extensions from a null context so, despite the crash is avoided, the checked extensions will be set to false no matter whether they are available or not. That's why it's not a valid fix.

[psivasubramanian]

@magomez Ok, otherwise we can have it under device flag or as device patch?

[zdobersek]

No, you need the sharing context for WebGL, accelerated 2D canvas and accelerated video playback.

Please investigate why the sharing context creation is failing. A call to GLContextWPE::createContext() should occur with the isSharing argument set to true, and apparently that call fails to create a valid context.

[psivasubramanian]

Ok.

…

On 10 January 2017 at 19:15, Žan Doberšek ***@***.***> wrote: No, you need the sharing context for WebGL, accelerated 2D canvas and accelerated video playback. Please investigate why the sharing context creation is failing. A call to GLContextWPE::createContext() should occur with the isSharing argument set to true, and apparently that call fails to create a valid context. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#142 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ARaeXk4UnooupYGz4mtMmtfh5u3djMf_ks5rQ4tqgaJpZM4LKUJN> .

-- Regards, Sivasubramanian