gfx object must call destroy, and should never be used again

[star-e]

trafficstars

Use Case

In webgpu backend, we have to destroy resources and delete object in xxx.destroy() method.

In native engine, v8 finalizer is used to delete native c++ object, while in webgpu, browser finalizer is not fully supported. As a result, we must call destroy to delete webgpu object.

Problem Description

Currently, gfx object might be used after destroy.

const framebuffer = shadowFrameBufferMap.get("xxx");
frameBuffer.destroy();
frameBuffer.initialize(info);

It is forbidden in webgpu backend.

Proposed Solution

Remove all usages of re-initialization of gfx object.

How it works

No response

Alternatives Considered

destroy leak gfx object in webgpu backend.

Additional Information

No response

[minggo]

Does it affect native usage? In native, we just use IntrusivePtr to control the lifecycle.

[minggo]

I agree that should not reuse gfx objects when they are destroyed.

[star-e]

Does it affect native usage? In native, we just use IntrusivePtr to control the lifecycle.

It does not affect native usage. In native engine, v8's finalizer will be called and c++ object is deleted or decRefed. So there is no leak.

In webgpu, browser's finalizer might not be called or is not supported at all. So every webgpu/webasm object must be deleted manually.

[pandamicro]

@star-e I assigned @hana-alice , if it's not appropriate, please re-assign