It's already documented that to unmap a buffer it has to have been mapped.
Vulkan was the only backend that was returning an OOM on missing `Buffer.block` but `Buffer.map_buffer` already returns an error in this case.
The Vec only ever contained 0 or 1 command buffers.
We now acquire an encoder on every submit for pending writes but that's not a problem since those are coming from a pool anyway.
The code was written with the incorrect assumption that if no lifetime is specified in a method definition, then all lifetimes are elided to the lifetime of self. In fact only lifetimes in the returned value are elided to the lifetime of self, and other parameters get their own lifetimes.
Kudos to @teoxoy for catching the issue!
- Improve logging in `StatelessTracker::remove_abandoned` to show the
outcome of the call.
- Add similar logging to `BufferTracker` and `TextureTracker`.
- Let `device_create_buffer`'s log only the new buffer's label, id,
and whether it's mapped at creation. It used to show the whole
descriptor, which is too much detail.
- Have `queue_submit` log the submission id, and have `device_poll`
log what it was waiting for, and what actually got done.
- Have `Device::drop` log the destruction of the raw device when it
actually happens, so it's properly ordered with respect to logging
from other parts of the device, like `Device::command_allocator`.
Fix two major synchronization issues in `wgpu_val::vulkan`:
- Properly order queue command buffer submissions. Due to Mesa bugs, two semaphores are required even though the Vulkan spec says that only one should be necessary.
- Properly manage surface texture acquisition and presentation:
- Acquiring a surface texture can return while the presentation engine is still displaying the texture. Applications must wait for a semaphore to be signaled before using the acquired texture.
- Presenting a surface texture requires a semaphore to ensure that drawing is complete before presentation occurs.
Co-authored-by: Jim Blandy <jimb@red-bean.com>