Previews, as seen when this build job started (37303fa):
WebGPU _{webgpu.idl | Explainer | Correspondence Reference}
WGSL _{grammar.js | wgsl.lalr.txt}

D3D11 does have limits on the viewport size, as noted in the D3D11_VIEWPORT struct docs. Specifically it says the x and y coords must lie between D3D11_VIEWPORT_BOUNDS_MIN and D3D11_VIEWPORT_BOUNDS_MAX.

However, the D3D12 spec has no such limits mentioned in the D3D12_VIEWPORT struct docs, and I can't find any other docs that mention a limit in D3D12. As such, I'm inclined to assume that there is no limit, similar to Metal, but I would love confirmation from someone at Microsoft on that point. @RafaelCintron?

It's probably a safe guess that D3D12 has the same limits as D3D11, and in any case, it does have these constants, so presumably they're correct:
https://github.com/microsoft/DirectX-Headers/blob/06d1185aa7ba650f0f5ae25b9cb39b8a85573bbe/include/directx/d3d12.h#L1374-L1376

We should note these in the computation of maxTextureDimension2D on D3D12:

Thanks for locating that! I tried searching for D3D12_VIEWPORT_BOUNDS_MAX directly on Google and it simply didn't show up. 🙄.

Added related values to the correspondence doc.

Notes from today's call:

• Group feels like this is OK to accept.
• It would be good to do testing across more hardware to see if there are any common driver bugs in this area.
  • If so, polyfilling the behavior should be investigated, which may be complicated.
• There may be some considerations for compat mode, though it's worth noting that this does NOT introduce the ability to specify negative viewport widths or heights.
• It is worth investigating how viewports interact with foveated/variable rate shading. It was pointed out that the common pattern in Metal may require a significantly larger viewport than the render target in some cases. If that's the case, however, this change should still strictly be an improvement, with additional loosening being considered in the future.

@toji here is the reference to the Metal VRR map documentation https://developer.apple.com/documentation/metal/mtlrasterizationratemap?language=objc#Configuring-the-Rate-Map and https://developer.apple.com/documentation/metal/mtlrasterizationratemap/physicalcoordinates(screencoordinates:layer:)?language=objc

This is probably a separate topic, but at the framework level the viewport in Metal can be arbitrary sized based on the VRR map. Hiding this fact at the WebGPU level is probably wrong too since then the viewport region of size W x H would map to N x M pixels where W*H is not necessarily N*M, but as discussed this is overall an improvement over the existing spec. But perhaps this will not be sufficient in the future.

I commented in the WG meeting I thought there were GL issues but I'm pretty sure I was mistaken. There are no GL issues

width >= 0 and width <= maxTextureDimenion2D
height >= 0 and height <= maxTextureDimenion2D

No restrictions on x and y listed.

To get into the weeds. It says you can query the max width,height from GL_MAX_VIEWPORT_DIMS but it requires the values returned to be effectively >= what's returned by querying GL_MAX_TEXTURE_SIZE

Great, I think that means we can merge this! @toji would you mind addressing the two open comments for the correspondence doc first though?

Correspondence changes made, thanks! PTAL, I think this should be ready to land now.