🔗 Expose the Dxc linker #71
Conversation
| let exports = exports.ok().unwrap().get_result().unwrap(); | ||
| let use_exports = use_exports.ok().unwrap().get_result().unwrap(); |
There was a problem hiding this comment.
Ugh it really is time for me to PR that error handling simplification, this Result<DxcOperationResult, (DxcOperationResult, HRESULT)> is so useless 😬
I think you even need to call get_status() etc :(
There was a problem hiding this comment.
Agreed, but I didn't feel like cleaning it up in this PR. There's a bit of an ambiguity there too because there are multiple possible failure operations making this quite complex. (e.g. link and compile can fail without an actual compiler or linker error)
There was a problem hiding this comment.
Yeah definitely not. I was more so thinking of the other way around: any of these might return an error code (including .compile() or .link()), but DxcOperationResult might still contain an error blob so that there's more context why the HRESULT was false. It just doesn't make sense that there's also get_status() giving you two more HRESULTs on top of the function return.
There was a problem hiding this comment.
I think we can probably replace all of this with a thiserror type enum that just represents all of these error cases, and potentially just unwraps the DxcOperationResult into this new enum.
| let mut linker_error = 0u32; | ||
| let status_hr = unsafe { result.get_status(&mut linker_error) }; | ||
|
|
||
| if !hr.is_err() && !status_hr.is_err() && linker_error == 0 { |
There was a problem hiding this comment.
Like, how many errors do they want us to check, and at which point will DxcOperationResult::get_error_buffer() return sensible help text?
Alright so this needs a bit of work still, but I'm happy to be where we are.
spirv-linkerto link spirv binariesUnfortunately,
dxcdoesn't compileuse-export.hlslproperly when compiled for spirv. The compiler doesn't output any sensible ops for the shader, and as a result, actually linking anything for spirv can't quite be done at the moment.Relevant DXC bug: microsoft/DirectXShaderCompiler#5854