[Torchvision API] Exception for isolated objective operator execution

[mdabek-nvidia]

Category:

Bug fix

Description:

RuntimeError raised when executing a standalone operator, which is not a part of a Compose pipeline.

Additional information:

Affected modules and functionalities:

Key points relevant for the review:

Tests:

• Existing tests apply
• New tests added
  • Python tests
  • GTests
  • Benchmark
  • Other
• N/A

Checklist

Documentation

• Existing documentation applies
• Documentation updated
  • Docstring
  • Doxygen
  • RST
  • Jupyter
  • Other
• N/A

DALI team only

Requirements

• Implements new requirements
• Affects existing requirements
• N/A

REQ IDs: N/A

JIRA TASK: DALI-4644

[greptile-apps]

Greptile Summary

This bug fix improves the error experience when a DALI torchvision Operator is invoked standalone, outside of a Compose pipeline. Previously the operator's __call__ would run internal pipeline logic against a raw tensor and produce a cryptic internal RuntimeError; now it immediately raises a clear, descriptive RuntimeError pointing the user to Compose.

• operator.py: __call__ now raises a descriptive RuntimeError; the original execution logic is moved verbatim into a new private _invoke method.
• compose.py: _pipeline_function updated from op(input_node) to op._invoke(input_node), the sole call-site inside the pipeline graph builder.
• test_tv_compose.py: adds test_error_in_isolated_operators that asserts the new error fires with the expected message glob.

Confidence Score: 5/5

Safe to merge — the change is a straightforward split of __call__ into a hard-failing stub and a private _invoke, with a matching one-line update in compose.py and a covering test.

The three changed files are tightly coupled and the logic move is mechanical: no execution logic was altered, only the call-site routing changed. The new test verifies the error path with a message glob. No regressions to existing behavior are expected.

No files require special attention.

Important Files Changed

Filename	Overview
dali/python/nvidia/dali/experimental/torchvision/v2/operator.py	Splits __call__ into a hard-failing stub (raises RuntimeError) and a new private _invoke method that carries the original execution logic; the change is clean and well-scoped.
dali/python/nvidia/dali/experimental/torchvision/v2/compose.py	One-line update to call op._invoke(input_node) instead of op(input_node); straightforward and necessary follow-up to the operator.py change.
dali/test/python/torchvision/test_tv_compose.py	Adds test_error_in_isolated_operators that verifies the new RuntimeError is raised with the expected message glob; uses assert_raises with a message pattern per DALI convention.

Sequence Diagram

sequenceDiagram
    participant User
    participant Operator
    participant Compose
    participant _pipeline_function

    Note over User,Operator: Direct call (now blocked)
    User->>Operator: op(data_input)
    Operator-->>User: RuntimeError: not directly callable

    Note over User,_pipeline_function: Correct path via Compose
    User->>Compose: compose(data_input)
    Compose->>_pipeline_function: build pipeline graph
    _pipeline_function->>Operator: op._invoke(input_node)
    Operator-->>_pipeline_function: pipeline DataNode
    _pipeline_function-->>Compose: output node
    Compose-->>User: torch.Tensor / PIL.Image

Loading

_{Reviews (4): Last reviewed commit: "Adjust Compose tests" | Re-trigger Greptile}

[mdabek-nvidia]

@greptileai re-review