Reading Attributes

Reading Attributes

When to Use

Use this skill when the user asks to read an attribute value, fetch mesh data (points, faceVertexCounts, etc.), inspect a render setting, or sample transforms.

Inputs

Resolve inputs in this order: existing repository files and referenced snippets, explicit user request, then broader agent context.

• Target API surface: Python, C/C++, or both.
• Prim paths or prim-list handle, attribute name, expected USD value type, and scalar versus array read mode.
• Target consumer: CPU NumPy, C DLTensor, GPU-aware DLPack consumer, or metadata/schema inspection.
• Whether the caller needs raw storage, shape/dtype discovery, or values copied out for later use.
• Repository source snippets referenced below. Treat these snippets as the API source of truth.

Prerequisites

• Use an ovrtx checkout that contains the referenced examples and docs tests.
• Read the relevant > **Source:** snippet before writing or explaining API usage.
• Use stage-queries first if the user needs to discover prims or attribute schemas before reading values.

Instructions

1. Identify the target language, prim paths or prim-list handle, attribute name, scalar/array mode, memory target, and sync/async requirement.
2. Read the matching source snippet and copy its lifecycle pattern rather than inventing equivalent calls.
3. Validate dtype, shape, semantic, and ownership rules before proposing or editing code.
4. Keep bindings, mapped buffers, operation handles, and C strings alive only for the lifetimes documented by the referenced examples.
5. When changing code, run the narrow docs test or example that owns the snippet whenever practical.

Output Format

• For explanations, cite the relevant API names, source snippets, and caveats.
• For code changes, summarize the files changed, snippets affected, and validation run.

Scripts

This skill has no scripts.

Limitations

• Runtime string arrays, asset arrays, and timecode attributes are not supported.
• Read semantics must be OVRTX_SEMANTIC_NONE; semantic conversion is write-side only.
• The referenced snippets remain the source of truth; update or add tested snippets before documenting new API usage.

Overview

read_attribute returns one value per prim (a scalar read). read_array_attribute returns a variable-length array per prim. Both produce DLPack tensors — NumPy, Warp, PyTorch, and anything else that speaks DLPack can consume them without a copy.

The read sees the stage as-if all prior stream-ordered operations have completed, so a write followed immediately by a read is safe.

For mapping (zero-copy writes into ovrtx's internal buffer), see the mapping-attributes skill. For queueing work instead of blocking, every read_* method has an _async twin that returns an Operation — see the async-operations skill.

USD Type Lookup

Use this table to answer "how do I read USD type X?" For array-valued USD attributes, use read_array_attribute in Python and set binding.binding_desc.attribute_type.is_array = true in C. Python exposes shape dimensions; C uses DLDataType::lanes.

Python

Scalar read

Returns a ManagedDLTensor with shape (N,) for N input prims. Convert with np.from_dlpack() for a zero-copy numpy view.

Source: tests/docs/python/test_attribute_read.py snippet doc-read-attribute-scalar

Scalar read into a pre-allocated destination

Pass dest= with a DLPack-compatible tensor (NumPy array, Warp array, etc.). The read writes directly into dest; the returned tensor aliases the same memory. The dest dtype must match how the runtime stores the attribute.

Source: tests/docs/python/test_attribute_read.py snippet doc-read-attribute-dest-tensor

Array read

Returns a dict[prim_path, ManagedDLTensor]. Iteration order matches the input prim_paths.

Source: tests/docs/python/test_attribute_read.py snippet doc-read-array-attribute

Async read

read_attribute_async returns an Operation; call .wait() then .fetch() to pull the ManagedDLTensor out.

Source: tests/docs/python/test_attribute_read.py snippet doc-read-attribute-async

GPU destination (CUDA)

Allocate the destination on the GPU via any DLPack-compatible allocator (e.g. Warp). Pass the CUDA stream handle so the read is stream-ordered with your GPU work.

Source: tests/docs/python/test_attribute_read.py snippet doc-read-attribute-cuda-dest

Supported authored attribute read snippets

Python raw snippets:

The snippets listed in this table live in tests/docs/python/test_all_attributes.py.

No Python raw snippets are provided for token or asset reads yet. read_attribute returns raw numeric token IDs for token payloads and raw (token, 0) pairs for scalar asset payloads, and there is no public Python wrapper for resolving those payload IDs back to strings today. query_prims resolves prim paths and attribute names internally, but token-valued attribute payloads remain raw IDs.

Local and world-space extents

extent is the authored local-space extent. _worldExtent is populated as the transformed world-space extent.

Source: tests/docs/python/test_all_attributes.py snippet doc-extent-world-extent

C

Scalar read

Source: tests/docs/c/test_attribute_read.cpp snippet doc-read-attribute-scalar-c

Array read

Set binding.binding_desc.attribute_type.is_array = true — ovrtx_make_binding_desc defaults it to false.

Source: tests/docs/c/test_attribute_read.cpp snippet doc-read-array-attribute-c

Supported authored attribute read snippets in C

C raw snippets:

The snippets listed in this table live in tests/docs/c/test_all_attributes.cpp.

C token and asset snippets include path-dictionary resolution because the C API exposes ovrtx_get_path_dictionary().

Local and world-space extents in C

Source: tests/docs/c/test_all_attributes.cpp snippet doc-extent-world-extent-c

Key Types / Functions

C result layout (ovrtx_read_output_t):

• Scalar reads: buffer_count == 1, single tensor with shape [prim_count]. Multi-component values use dtype.lanes (for example, a 4x4 double matrix is shape=[prim_count], dtype={kDLFloat, 64, 16}).
• Array reads: buffer_count == prim_count, one tensor per prim (variable length). Multi-component array elements also use dtype.lanes (for example, a point3f[] with 10 points is shape=[10], dtype={kDLFloat, 32, 3}).
• When a caller-supplied read_dest tensor was passed in: buffer_count == 0 (data landed in your tensor).
• Python reshapes lane-based C attribute tensors into NumPy-style trailing dimensions before exposing them through ManagedDLTensor.

Troubleshooting

• Semantics are write-only. ovrtx_read_attribute rejects any semantic other than OVRTX_SEMANTIC_NONE with "Semantic conversion is not supported for read_attribute". Always use NONE on the read binding and ask for the raw storage dtype (e.g. {kDLFloat, 64, 16} for a 4x4 matrix).
• Generic authored USD attributes require opt-in population. Root-layer customLayerData.populateAllAuthoredAttributes = true asks the runtime to populate authored attributes beyond the normal schema set. Use it only when needed: populating everything can dramatically increase memory usage on assets with many unused properties. See loading-usd for the layer metadata tradeoff.
• Schema-owned attributes fix the element type. omni:rtx:rtpt:maxBounces is stored as uint32 even if you wrote it as int32. When allocating a dest tensor, match the runtime's dtype (np.uint32) — not what you wrote.
• PrimMode.EXISTING_ONLY (default) silently skips prims that don't exist on the stage. Use MUST_EXIST if you want an error instead. CREATE_NEW is not supported for reads.
• Release in C, rely on GC in Python. ovrtx_release_read_result must be called after ovrtx_fetch_read_result. In Python ManagedDLTensor releases automatically when the tensor (and every DLPack consumer of it) is dropped.
• Array reads need is_array=true on the binding. The ovrtx_make_binding_desc helper defaults is_array=false; set it explicitly after the call for array attributes.
• C binding descriptors borrow path storage. ovrtx_make_binding_desc stores the ovx_string_t* prim path array you pass; it does not copy it. Keep the ovx_string_t objects and array alive until the read has been enqueued, waited, fetched, and released. For persistent bindings, keep them alive until ovrtx_create_attribute_binding finishes.
• Unsupported authored types are tested as absent. tests/docs/data/all-attributes.usda deliberately authors string[], asset[], custom relationships, timecode, and timecode[]; the all-attributes tests assert they are not populated by the current runtime.

Related skills

• stage-queries — discover prims and their attribute schemas before reading.
• writing-attributes — write values that can then be read back.
• mapping-attributes — zero-copy mapping (no fetch step).
• async-operations — polling, timeouts, the two-phase Operation/PendingFetch lifecycle.

References

• Use the > **Source:** directives in this skill to locate tested snippets before reusing API patterns.
• Keep related skills, docs, and snippets synchronized when changing the workflow.