Fix #8775: handle torch.Tensor input in compute_shape_offset

[williams145]

Description

compute_shape_offset in monai/data/utils.py passes spatial_shape directly to np.array(). When spatial_shape is a torch.Tensor, this relies on the non-tuple sequence indexing protocol, which PyTorch removed in version 2.9. The call raises a hard error on PyTorch ≥ 2.9.

The fix is to wrap spatial_shape in tuple() before passing it to np.array(). This routes through __iter__, which is stable across all PyTorch versions, and produces 0-d scalar tensors that NumPy consumes correctly.

Root cause

The direct caller in monai/transforms/spatial/functional.py (line 115) constructs an in_spatial_size as a torch.Tensor and passes it straight to compute_shape_offset. This path has been broken since PyTorch 2.9.

Changes

• monai/data/utils.py — one-character change: np.array(spatial_shape, ...) → np.array(tuple(spatial_shape), ...)
• tests/data/utils/test_compute_shape_offset.py: new unit tests covering torch.Tensor, np.ndarray, and plain list inputs

Fixes #8775

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: bd7b703f-8213-4152-a539-29987abc607a

📥 Commits

Reviewing files that changed from the base of the PR and between 33ebd0a and dfdb87d.

📒 Files selected for processing (1)

• tests/data/utils/test_compute_shape_offset.py

✅ Files skipped from review due to trivial changes (1)

• tests/data/utils/test_compute_shape_offset.py

---

📝 Walkthrough

Walkthrough

Replaces construction of shape in monai/data/utils.py::compute_shape_offset to materialize spatial_shape as a tuple before converting to a NumPy array (now np.array(tuple(spatial_shape), copy=True, dtype=float)). Adds a new test module tests/data/utils/test_compute_shape_offset.py that sets an identity affine and verifies compute_shape_offset accepts spatial_shape as a NumPy array, a Python list, and a PyTorch tensor, checking output length and numeric equality (with np.testing.assert_allclose).

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	Title directly references the fixed issue (#8775) and accurately summarizes the core change: handling torch.Tensor input in compute_shape_offset.
Description check	✅ Passed	PR description clearly explains the bug (PyTorch 2.9 breaking change), root cause, and solution with code examples. All required sections are present.
Linked Issues check	✅ Passed	Code changes directly address issue #8775: wrapping spatial_shape in tuple() before np.array() call prevents PyTorch 2.9+ indexing breakage. Unit tests validate behavior across torch.Tensor, np.ndarray, and list inputs.
Out of Scope Changes check	✅ Passed	All changes are in-scope: monai/data/utils.py contains the one-line fix, and new test file covers the fix. No extraneous modifications detected.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

tests/data/utils/test_compute_shape_offset.py (1)

24-45: Add docstrings for new test definitions.

setUp and the new test_* methods currently lack docstrings.

As per coding guidelines, "Docstrings should be present for all definition which describe each variable, return value, and raised exception in the appropriate section of the Google-style of docstrings."

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@tests/data/utils/test_compute_shape_offset.py` around lines 24 - 45, Add
Google-style docstrings to the test methods setUp, test_numpy_array_input,
test_list_input, test_torch_tensor_input, and
test_identity_affines_preserve_shape describing purpose, arguments (if any),
return (None), and any exceptions; for example, document that setUp initializes
self.affine, that each test calls compute_shape_offset with different shape
input types and asserts expected output length or values, and that no exceptions
are expected. Keep docstrings brief and placed immediately under each function
definition.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@tests/data/utils/test_compute_shape_offset.py`:
- Around line 29-44: Tests in test_compute_shape_offset.py unpack
compute_shape_offset into (out_shape, offset) but never use offset, triggering
an unused variable lint warning; update the test cases (those calling
compute_shape_offset in test_list_input, test_torch_tensor_input,
test_identity_affines_preserve_shape and the earlier scalar input) to either
unpack as (out_shape, _) or explicitly assert offset when relevant so the unused
`offset` is removed — locate the calls to compute_shape_offset and change the
second variable name to `_` (or add a meaningful assert on `offset` where the
test intends to validate it).

---

Nitpick comments:
In `@tests/data/utils/test_compute_shape_offset.py`:
- Around line 24-45: Add Google-style docstrings to the test methods setUp,
test_numpy_array_input, test_list_input, test_torch_tensor_input, and
test_identity_affines_preserve_shape describing purpose, arguments (if any),
return (None), and any exceptions; for example, document that setUp initializes
self.affine, that each test calls compute_shape_offset with different shape
input types and asserts expected output length or values, and that no exceptions
are expected. Keep docstrings brief and placed immediately under each function
definition.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: c0c7abd9-cf23-46e3-9fa1-2b7c49d4ed74

📥 Commits

Reviewing files that changed from the base of the PR and between 8d39519 and 33ebd0a.

📒 Files selected for processing (2)

• monai/data/utils.py
• tests/data/utils/test_compute_shape_offset.py

[williams145]

The quick-py3 (macOS-latest) failure is a pre-existing infrastructure issue, pytype 2024.4.11 fails to install on the macOS runner due to a missing pybind11 dependency, which is unrelated to the changes in this PR. The same failure appears on other open PRs against dev for the same reason.

[ericspod]

Hi @williams145 thanks for this fix, this was already in the works in #8776 however. What I'd suggest is to merge that one first then merge in your extended test set which I think are good to add. We have to merge #8814 before this can be merged however which should address the pybind11 issue.

[williams145]

Hi @williams145 thanks for this fix, this was already in the works in #8776 however. What I'd suggest is to merge that one first then merge in your extended test set which I think are good to add. We have to merge #8814 before this can be merged however which should address the pybind11 issue.

PR #8814 has been merged. The pybind11 blocker should now be resolved, this should mean that CI can be re-triggered when convenient.

[atharvajoshi01]

Converting to tuple before passing to np.array is a good fix for PyTorch >= 2.9 which deprecated non-tuple sequence indexing. Test coverage looks solid — covers numpy, list, tuple, and torch.Tensor inputs.