dx12: Adopt BAL approch #2266
dx12: Adopt BAL approch #2266
Conversation
dx12: Update formatting
|
Hmm, I'm not sure it's worth it for us to split the backends into HAL/BAL sections. It would introduce longer dependency chains for us and clients. Couldn't the backend crates just provide both? Assuming HAL implementation is mostly just wrapping around BAL primitives. |
src/bal/dx12/src/copy.rs
Outdated
| } | ||
|
|
||
| #[derive(Clone)] | ||
| pub struct Copy { |
There was a problem hiding this comment.
Naming conflict with std::prelude::Copy is undesirable.
src/bal/dx12/src/native/com.rs
Outdated
| } | ||
|
|
||
| impl<T: Interface> WeakPtr<T> { | ||
| pub fn as_unknown(&self) -> &IUnknown { |
There was a problem hiding this comment.
This function must be marked unsafe.
Otherwise this "safe" expression WeakPtr::from_raw(1usize as *mut u32).as_unknown() will trigger UB.
src/bal/dx12/src/native/com.rs
Outdated
| } | ||
|
|
||
| // Cast creates a new WeakPtr requiring explicit destroy call. | ||
| pub fn cast<U>(&self) -> D3DResult<WeakPtr<U>> |
src/bal/dx12/src/native/com.rs
Outdated
|
|
||
| // Destroying one instance of the WeakPtr will invalidate all | ||
| // copies and clones. | ||
| pub fn destroy(&self) { |
| impl<T> Deref for WeakPtr<T> { | ||
| type Target = T; | ||
| fn deref(&self) -> &T { | ||
| debug_assert!(!self.is_null()); |
There was a problem hiding this comment.
This type is pub so debug_assert doesn't seem enough.
|
The Moving out common pieces like blitting will done in a separate PR. |
PoC for switching towards the BAL concept suggested in #2249
/srcstructure:backend/dx12moved tohal/dx12hal/dx12has a new dependency onbal/dx12Source code from HAL will be gradually refactored into the corresponding BAL backend in order to slim down the HAL implementation and move 'unsafe' portions into BAL (ie zero cost wrappers over d3d12 structs)
Upcoming PR should be a rustfmt as the backend code doesnt seem to be formatted at all (?)