[deno] Calling `CommandEncoder.finish` multiple times should raise a validation error

[andyleiserson]

Many wgpu-core functions return both an object ID and an error when they fail. For typical functions that construct a new object, the returned ID when there is an error refers to a new, invalid object, which is fine. But in the case of CommandEncoder.finish, there is a problem, because wgpu-core does not have distinct object types for command encoders and command buffers. The returned ID (for a command buffer) is just the ID of the encoder being finished. This means that calling CommandEncoder.finish multiple times creates multiple command buffer objects with the same ID, which can cause problems if those objects are not dropped correctly.

According to the spec, calling finish() multiple times for the same encoder should result in a validation error. With the wgpu API, finish() consumes the encoder so it is not possible to call multiple times. In deno, as of #7808, calling finish() multiple times raises a TypeError immediately in order to avoid returning a duplicate command buffer. For Firefox, see the first comment.

[andyleiserson]

Firefox avoids problems here because it gives the command buffer a reference to the command encoder, and does nothing when dropping command buffers. So although it is possible to create multiple command buffers for the same encoder, there is always only one call to wgpu-core to release the encoder/buffer object, after the encoder and all duplicative command buffers are dropped. (However, using the same object for encoders and buffers does cause a different problem in Firefox.)