Function instance_flags 

pub fn instance_flags() -> InstanceFlags

The InstanceFlags to build a wgpu instance with: wgpu’s own defaults, read from the environment, minus VALIDATION_INDIRECT_CALL.

That flag is set in release builds too — it comes from InstanceFlags::from_build_config’s non-debug branch, so it is not a debug-only cost. It makes Device::new build a set of compute and render pipelines that validate the arguments of indirect draws. This renderer issues no indirect draws at all — not one draw_indirect, dispatch_indirect or multi_draw_* anywhere in the workspace — so those pipelines check nothing we will ever submit.

That alone would only be wasted startup work. The reason the flag is cleared is that building those pipelines is also a way for device creation to fail, and failing there is not survivable. wgpu_core::device::resource::Device::new creates the hal device, then its empty_bgl — which registers a bind-group layout with the Vulkan backend’s DescriptorAllocator — and only then calls IndirectValidation::new(..)?. A driver that cannot build them takes that ?, and the early return drops the hal device without unregistering empty_bgl, because hal objects are not RAII and need an explicit destroy. Drop for DescriptorAllocator then finds a non-empty bucket and panics — “buckets are not empty, at least one BGL has not been unregistered” — from an ordinary, non-unwinding drop, so its own thread::panicking() guard does not suppress it.

The process therefore dies inside request_device, and neither caller can do anything about it there: a backend search cannot search past a panic, and an offscreen renderer cannot fall back to a software adapter. Reported from the field on an older Windows 10 machine where an app never opened a window, and confirmed there by setting WGPU_VALIDATION_INDIRECT_CALL=0, which let the window open. D3D12 has no DescriptorAllocator and never runs the assertion, which is why forcing WGPU_BACKEND=dx12 looked like a graphics fix when it was really a way of not reaching this code.

WGPU_VALIDATION_INDIRECT_CALL is honoured in both directions, so the flag stays reachable for anyone debugging wgpu itself.