Scene test failure in Chrome with WebGL validation (ANGLE)

[bagnell]

[mramato]

I'm actually getting 6 failures in Chrome with validation on (and several failures in Firefox as well). They almost all have the same error RuntimeError: WebGL Error: INVALID_OPERATION: drawElements(4, 6, 5123, 0);

[pjcozzi]

I did not get these when doing the 1.11 release today.

[bagnell]

In the console I also see these errors:
WebGL: INVALID_OPERATION: drawElements: no ELEMENT_ARRAY_BUFFER bound
GL ERROR :GL_INVALID_VALUE : glDrawElements: out of memory

[bagnell]

I'm actually getting 6 failures in Chrome with validation on...

All of the tests pass for me if the debugShowGlobeDepth test doesn't run.

[mramato]

I'm still getting these failures as of the 1.13 release.

[pjcozzi]

Still the same in 1.15.

[pjcozzi]

All tests pass on Mac with WebGL validation so this is Windows/ANGLE related.

[mramato]

I can confirm that disabling ANGLE on windows makes the errors go away, we should write up an issue for them when we get a chance (after confirming our tests aren't doing anything odd themselves).

[lilleyse]

Getting these failures for the 1.16 release as well.

[mramato]

I can no longer produce the above failures; but I am seeing a new set of failures related to Scene/Sunacross all browsers. It's all the same error across multiple tests:

Scene/Sun draws in 3D
DeveloperError: Framebuffer is not complete. Incomplete attachment: at least one attachment point with a renderbuffer or texture attached has its attached object no longer in existence or has an attached image with a width or height of zero, or the color attachment point has a non-color-renderable image attached, or the depth attachment point has a non-depth-renderable image attached, or the stencil attachment point has a non-stencil-renderable image attached. Color-renderable formats include GL_RGBA4, GL_RGB5_A1, and GL_RGB565. GL_DEPTH_COMPONENT16 is the only depth-renderable format. GL_STENCIL_INDEX8 is the only stencil-renderable format.

Can someone confirm? Should we look at this before 1.20? A DeveloperError in our own code seems bad.