chore: upgrade OpenXR-SDK 1.1.54 and fix build issues #205
Conversation
I'm not sure what you're referring to here. I don't see any references to |
Update the generator to handle new types and requirements introduced in OpenXR SDK 1.1.54. Generator changes: - Derive `Default` for enums with a `0` value (required by `SpatialBufferEXT`). - Relax `has_non_default` check for enums; added workaround for `XrNegotiateLoaderInfo` to avoid unwanted implementations. - Fix identifier normalization for `XrSpatialMarkerArucoDict` and `XrSpatialMarkerAprilTagDict` (prevents invalid starting digits). - Exclude `XrRaycastHitResultANDROID` re-exports in high-level generation to avoid conflicts. - Temporarily exclude `XrSpatialAnchorsCreateInfoBaseHeaderML` and `XrFutureCompletionBaseHeaderEXT` from polymorphic builders due to complexity changes. Manual adjustments: - Update `sys/lib.rs` and `openxr/lib.rs` with necessary use statements and basetype workarounds. - Refactor `SpaceUserIdFB` and `FutureEXT` to use the `wrapper!` macro.
MungbeanOuO
force-pushed
the
feature/dependency-update
branch
from
961bc88 to
a7bcba2
Compare
|
Thanks for the clarification! To explain my original confusion: I noticed that For this PR, I'll proceed with:
Since I'm not yet sure how strictly the generated code relies on the EXT variant (given the XML situation), I'd prefer to land this safely first. I will investigate the proper way to handle deprecation and unification in a future iteration. |
This commit addresses several warnings and compilation errors across the crate: 1. Generator: - Modified `generate_reader` to conditionally apply `#[allow(dead_code)]` to struct fields when no getters are generated. - This fixes the "field 0 never read" warning observed in empty event structs like `InteractionRenderModelsChangedEXT`. 2. Refactor: - Changed `UuidEXT` from a direct re-export to a type alias of `Uuid`. - Note: The `#[deprecated]` attribute is currently commented out to prevent internal warnings, pending a future cleanup. 3. Fixes: - Added missing `unsafe` blocks in `opengles.rs` required by recent edition/compiler updates. - Added hardcoded `AIBinder` definition in `platform.rs` to fix build errors.
MungbeanOuO
changed the title
| } else if meta.has_pointer || meta.has_array && name != "XrUuid" { | ||
| quote! { #[derive(Copy, Clone, Debug)] } | ||
| } else if meta.has_non_default { | ||
| } else if meta.has_non_default || name == "XrNegotiateLoaderInfo" { |
There was a problem hiding this comment.
Why is this a special case? Document.
| .iter() | ||
| .cloned() | ||
| .filter(|&x| { | ||
| x != "XrRaycastHitResultANDROID" && x != "XrTrackableMarkerDatabaseEntryANDROID" |
| let has_zero = e | ||
| .values | ||
| .iter() | ||
| .any(|v| matches!(v.value, ConstantValue::Literal(0))); |
There was a problem hiding this comment.
This logic appears twice. Should we compute it once up front and store the result in Enum?
| }); | ||
|
|
||
| let field_attr = if readers.clone().next().is_none() { | ||
| quote! { #[allow(dead_code)] } |
| MAX_COLOCATION_DISCOVERY_BUFFER_SIZE_META, MAX_VIRTUAL_KEYBOARD_COMMIT_TEXT_SIZE_META, Path, | ||
| SystemId, Time, Uuid, Version, | ||
| }; | ||
| // #[deprecated(note = "Use `Uuid` instead")] |
There was a problem hiding this comment.
Adding commented-out code doesn't serve any useful purpose. Replace with an explanatory TODO or omit entirely.
| SystemId, Time, Uuid, Version, | ||
| }; | ||
| // #[deprecated(note = "Use `Uuid` instead")] | ||
| pub type UuidEXT = Uuid; |
There was a problem hiding this comment.
What is the benefit to introducing a new type alias here rather than reexporting the one we already generate regardless?
2 participants
Notes regarding Generator Workarounds:
Polymorphic Builders Exclusion: I've temporarily excluded XrFutureCompletionBaseHeaderEXT. It seems the SDK update introduced a need for manual handle processing here, which is out of scope for this PR.
Android Structs: Excluded XrRaycastHitResultANDROID from simple_structs generation due to re-export conflicts.
Default Trait: Added logic to derive Default for enums (needed for SpatialBufferEXT), but had to patch XrNegotiateLoaderInfo manually to prevent it from cascading incorrectly.
Questions:
Uuid Usage: I noticed Uuid and UuidEXT are used separately in lib.rs. Is this the intended behavior or should they be unified?
Warnings: There is currently a dead_code warning on InteractionRenderModelsChangedEXT (field 0 never read). I plan to address this in a follow-up commit, but heads-up if you see it in CI.