Three.js: Invalid Internal Format gl.DEPTH_STENCIL

[WestLangley]

Safari WebGL2 context only

WebGL: INVALID_ENUM: renderbufferStorage: invalid internalformat

in WebGLTextures.js

_gl.renderbufferStorage( _gl.RENDERBUFFER, _gl.DEPTH_STENCIL, renderTarget.width, renderTarget.height );

Disabling the stencil buffer for shadows removes the symptoms, but I expect that is not the proper fix.

/ping @aardgoose
/ping @gkjohnson

https://threejs.org/examples/webgl_shadowmap.html (EDIT: With #19927, this example is no longer an issue.)

https://threejs.org/examples/webgl_clipping_stencil.html (EDIT: This is now working with Safari 14.)

Three.js version

• [ x ] Dev
• [ x ] r118

Browser

• [ x ] Safari - 13.1.2

OS

• [ x ] macOS
• [ x ] iOS

[WestLangley]

It may be this is NOT a browser issue. Other browsers may be converting the specified internal format to something acceptable.

[gkjohnson]

I can't test with safari at the moment but presumably this happens with any example that uses a render target and doesn't disable the stencil buffer?

I'm less familiar with the difference in browser implementations of webgl but according to the spec DEPTH_STENCIL should be a valid value for WebGL2 which makes it sound like a browser issue unless there's something more to it.

When using a WebGL 2 context, the following values are available __additionally__:
...

Maybe it's worth trying some other depth buffer types new to WebGL2? gl.DEPTH_COMPONENT24, gl.DEPTH_COMPONENT32F, gl.DEPTH24_STENCIL8, or gl.DEPTH32F_STENCIL8.

[Mugen87]

I've tested https://threejs.org/examples/webgl_shadowmap.html on macOS 10.15.6 with Safari 13.1.2 and it runs as expected. Same for https://threejs.org/examples/webgl_clipping_stencil.html.

Used an iMac from 2014 with AMD GPU.

[mrdoob]

@WestLangley I guess this is using Safari's WebGL2 experimental setting?

[WestLangley]

I guess this is using Safari's WebGL2 experimental setting?

Correct.

Safari 13.1.2 and it runs as expected.

With a WebGL2 context?

[mrdoob]

I don't know if this is a case of WebGL2 in Safari being more strict or being incomplete.
FWIW, Safari 14 has a different WebGL 2 implementation (ANGLE based).

[WestLangley]

I don't know if this is a case of WebGL2 in Safari being more strict or being incomplete.

well-said. :-)

[Mugen87]

With a WebGL2 context?

Checked this now and it uses WebGL 1.

[WestLangley]

@Mugen87 Please enable experimental WebGL2 in the Develop method. This PR is about Safari WebGL2 context only.

[Mugen87]

Okay, now I can reproduce. I forgot it's necessary to enable a flag^^.

[Mugen87]

How about revisiting this topic with Safari 14? I mean if the implementation will be a new one, does it makes sense to invest now time in this issue?

[gkjohnson]

I just tested this in Safari 14 and the stencil clip planes example seems to work, now.

[mrdoob]

@WestLangley Is there any issue left?

[WestLangley]

I do not think so. Closing.