Fix reallocation of a depth render target in the SceneGrab class

[querielo]

There's a scenario where a camera has a custom render target without a depth buffer texture, and the SceneGrab class incorrectly assumes that the grabbed depth texture has to be the size of the device. This can cause issues when the color buffer of the render target and a canvas have different sizes.

To fix this issue, I suggest that the code should check the size of the color buffer in the absence of a depth buffer, since they should have the same size. This will ensure that the reallocation of SceneGrab.depthRenderTarget works as expected.

I hope this helps! Let me know if you have any questions.

I confirm I have read the contributing guidelines and signed the Contributor License Agreement.

[mvaligursky]

If the render target does not have depth, we should probably just log error and early out / fail as what is the reason to execute depth grab if there is no depth buffer?

[querielo]

Hm. I like your suggestion.

But it just looks like there are two independent ways to grab depth buffers. "Depth grabpass" simplify usage "getLinearScreenDepth" or something like this. And a user can instantiate RenderTarget the next way:

const renderTarget = new pc.RenderTarget({
    colorBuffer: texture
    depth: true
});

In this case, renderTarget._depthBuffer === undefined

[mvaligursky]

In this case, renderTarget._depthBuffer === undefined

Yep, that's a valid point!

[mvaligursky]

Happy with this, thanks!

[querielo]

Thanks!

I have met another case. A camera with posteffects + Depth grabpass + Render target

engine/src/framework/components/camera/post-effect-queue.js

Line 114 in d753e00

depth: useDepth,