What if the game application creates PbufferSurface instead of WindowSurface?

[OliverCho1]

As far as I know, PATrace supports the on-screen surface for capturing.
However, some kinds of apps create surfaces by "eglCreatePbufferSurface", I mean if it creates an off-screen surface, is there any way I can capture GLES calls of the application?

Please let me know the currently available support and whether you have any further plans.

Thanks.

[per-mathisen-arm]

It should work exactly the same way, but I don't think we've tried tracing and replaying apps without any surface window. There might be bugs. Please let us know if you encounter any.

[OliverCho1]

I'm trying debugging on "Space STG 3".
https://play.google.com/store/apps/details?id=org.ovh.SpaceSTG3&hl=en_US

It creates the surface by calling eglCreatePbufferSurface.
And there is no EGLSurface in the SurfMap, so I can see the below error log message.

"egltrace.cpp,1131: Error: EGLSurface 0xb400007249045ec0 does not exist!"

I guess if the EGLSurface does not exist, PATrace cannot capture it.

If you find any, please let me know.

[per-mathisen-arm]

Looks like we're not registering created pbuffers during tracing. I guess we simply haven't come across apps using them yet ourselves. This is a completely untested patch that might fix it, feel free to try it and please let us know if it works:

index 7284846b..77297f6c 100644
--- a/patrace/src/tracer/trace.py
+++ b/patrace/src/tracer/trace.py
@@ -423,7 +423,7 @@ class Tracer:
 
         self.invokeFunction(invoke_func)
 
-        if func.name == 'eglCreateWindowSurface':
+        if func.name in ['eglCreateWindowSurface', 'eglCreatePbufferSurface']:
             print('    EGLint x = 0, y = 0, width = 0, height = 0;')
             print('    if (_eglQuerySurface(dpy, _result, EGL_WIDTH, (EGLint*) &width) == EGL_FALSE)')
             print('        DBG_LOG("Failed to query surface width\\n");')
@@ -896,7 +896,7 @@ class Tracer:
             print('    gTraceOut->mpBinAndMeta->writeHeader(true);')
         if func.name == 'eglMakeCurrent':
             print('    after_eglMakeCurrent(dpy, draw, ctx);')
-        if func.name == 'eglCreateWindowSurface':
+        if func.name in ['eglCreateWindowSurface', 'eglCreatePbufferSurface']:
             print('    after_eglCreateWindowSurface(dpy, config, _result, x, y, width, height);')
         if func.name == 'eglDestroySurface':
             print('    after_eglDestroySurface(surface);')

[OliverCho1]

I tested your patch, and I don't see the error logs I mentioned before.
("egltrace.cpp,1131: Error: EGLSurface 0xb400007249045ec0 does not exist!")

Even the EGLSurface is created, it's still not working.

[Suspicious Logs]

1. egltrace.cpp,333: Width: 1, Height: 1
   -> It seems "_eglQuerySurface" gets wrong value in "patrace_eglCreateWindowSurface"(egltrace_auto.cpp)
2. egltrace_auto.cpp,46511: eglDestroyContext called (implies flush to disk)
   -> It's not appeared in normal working app capturing.
3. retracer.cpp,1684: Never rendered anything.
   -> I guess it means there is no data for replay.

So..., please check in detail whether something was missed again.

[per-mathisen-arm]

Do you think you could upload the trace file somewhere? Then I could take a quick look at it.

[OliverCho1]

I uploaded the pat file to the below link.

https://drive.google.com/file/d/13ZoDqpWeiJ7j1t9Y3Z3Ds-tIvy3AvzM-/view?usp=share_link
-> "org.ovh.SpaceSTG3.1.zip"

[per-mathisen-arm]

That trace file did not really contain anything useful, certain no gameplay. If the tracing created multiple trace files, then perhaps one of the other trace files contained the actual gameplay.

[OliverCho1]

Unfortunately, even though I tried multiple times to get multiple tracing files, it did not work at all.

So, you don't have any plans for supporting pBufferSurface cases now, right?

But, if you want to debug this kind of case further, I'm willing to help you.

[per-mathisen-arm]

I've made an internal ticket for us to look into this. No idea when we'll have the time to do so, though.