Aardvark.Rendering.Vulkan 5.6.10

• • Aardvark.Base.FSharp (>= 5.3.27 && < 5.4.0)
  • Aardvark.Base.Incremental (>= 5.3.27 && < 5.4.0)
  • Aardvark.Base.Tensors (>= 5.3.27 && < 5.4.0)
  • Aardvark.Rendering (= 5.6.10)
  • FShade (>= 5.7.16 && < 5.8.0)
  • FSharp.Core (>= 8.0.300)
  • GLSLangSharp (>= 0.5.0 && < 0.6.0)

- Fixed adaptive texture-attachment size metadata to track the selected mip level and texture changes rather than reporting the base-level extent.
- Fixed geometry unions to concatenate only live per-vertex attribute prefixes, preventing trailing padding from becoming another operand's vertex data.
- Fixed struct-uniform property getters to use value-type addresses without boxing, while preserving reference getter dispatch and field access.
- Enforced overflow-safe logical bounds for buffer-range transfers and vector slices, including empty and reversed views.
- Prevent signed-integer overflow from bypassing shared texture slice, mip-level, and window bounds validation ([#143](https://github.com/aardvark-platform/aardvark.rendering/issues/143)).
- Fixed BC1/BC2/BC3 encoding to preserve color variation when the initial endpoint-estimation direction lies in the covariance matrix's nullspace.
- Fixed BC2/BC3 decoding for ascending or equal RGB endpoints by always using the four-color palette.
- Fixed BC volume layout and DDS mip payloads (#171)
- Fixed adaptive texture and texture-array replacement failures to forget successfully disposed handles, preventing their reuse or repeated disposal during retries and release.
- Fixed line and triangle strip conversion to preserve source geometries and handle empty or underfilled strips.
- Fixed orthographic picking to use parallel unit rays from pixel positions on the near plane rather than the camera location.
- Fixed `BufferView.getCount` and automatic draw counts to honor byte strides, count only complete elements, and return zero for exhausted storage.
- Fixed `Task.bind` to complete when its mapping function throws synchronously or returns a null task.
- Made `ManagedBuffer` range filling overflow-safe.
- Fixed `Trie` preorder links for prefix and sibling updates, while preserving comparer and insertion ordering. Replaced, removed, and cleared `Trie` values are now detached cleanly from their trie-owned links.
- Fixed indirect draw array/list/sequence range validation to reject integer overflow while preserving exact-end empty ranges and array reuse.
- Fixed array/span buffer creation to release newly allocated buffers when initial upload fails.
- Fixed `Arr` uniform writes to truncate surplus values, zero-fill missing target storage, and remain within shader-array bounds.
- Fixed phi/theta sphere tessellation to keep latitude rings between the explicit poles, eliminating degenerate triangles and zero-length wire segments without changing output counts.
- Fixed `ChunkedMemoryManager` reallocation of freed handles to adopt replacement backing storage while preserving the caller's block identity.
- Made `ChunkedMemoryManager` frees idempotent under contention, preventing duplicate waiters from retiring backing chunks that still contain live allocations.
- Fixed `MemoryManager` and `ChunkedMemoryManager` shrinking to update the surviving block size before releasing its tail, preventing overlapping reported live ranges.
- Made array-backed ray-tracing geometry subranges overflow-safe and restricted the all-remaining sentinel to -1; `Box3d` bounding-box subranges now convert only selected elements into compact storage.
- Amortized `ASet.compact` storage resizing without retaining removed keys (#173)
- [Sg] Build line and triangle attributes in one source pass without per-primitive temporary arrays (#177)
- [Sg] Separate weak buffer-cache representations and atomically publish concurrent misses (#180)
- [Sg] Correct tapered-cylinder and cone side normals while reducing normal-buffer allocation (#186)
- [Sg] Fixed automatic picking to honor `DrawCallInfo` ranges and indexed `BaseVertex` offsets.
- [Sg] Made render-object bounding-box caching caller-scope aware, fixing reused render-object sets beneath distinct transforms while retaining weak, allocation-free cache hits.
- [Text] Make ShapeList.concat linear while preserving layout and enumeration semantics (#175)
- [Application] Fix camera orbiting around nonzero world-space centers and avoid an intermediate view (#42)
- [Application] Cancel held keyboard and mouse input on focus loss without generating click gestures or preserving double-click history.
- [GL] Explicitly disable GL_EXT_debug_printf