feat(data)!: field windows measured at run; one Resample, no bound to declare

[vboussot]

BREAKING. A displacement field no longer needs — or accepts — a declared bound: each region's source window is sized from the field values read at run (one read, shared between sizing and sampling), while the plan stays headers-only and prices the field as zero. max_displacement is gone, and so is every legacy Resample spelling (Warp, ResampleToReference, ResampleTransform, ResampleToResolution, ResampleToShape) — Resample is the one name.

• 59fa477 feat(data): size a field's windows from the field itself at run
• 9c94f89 feat(data): drop max_displacement — the field itself is the bound
• 7a809b3 feat(data)!: one Resample, no other spelling
• d0ae5ab fix(impact-reg): one staging root per group, so mixed forms stay readable
• 3a2458c perf(impact-reg): fill Transform.h5 region by region

Published configs migrate with s/ResampleToResolution:/Resample:/ (and likewise for the other names) at the 1.8 bundle bump.

Stacked on feat/konfai-python-api.

[coderabbitai]

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 60.29% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check	⚠️ Warning	The description explains the breaking changes and migration, but it omits the template sections for testing, type selection, checklist, and related issues.	Add the required template headings, record test commands and results, select the change type and breaking-change option, and complete the checklist.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly summarizes the main breaking change and follows the repository's Conventional Commits format.

---

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

[coderabbitai]

Actionable comments posted: 5

🧹 Nitpick comments (2)

apps/impact_reg/tests/unit/test_orchestration.py (1)

248-258: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Consider covering a group whose entries do not share one form.

This test pins re-staging with one form. The case that is not covered is a group staged with two forms — for example an ensemble whose presets emit DVF.mha and DVF.ome.zarr. _stage_group then labels the root from whichever entry comes last. A test asserting the intended behavior for that input would pin the contract the docstring claims.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/impact_reg/tests/unit/test_orchestration.py` around lines 248 - 258, The
test coverage for _stage_group only exercises repeated staging of one form; add
a test that stages a single group containing entries whose sources use different
forms, such as DVF.mha and DVF.ome.zarr. Assert the intended root/spec form
behavior promised by _stage_group’s docstring, including that the result does
not depend incorrectly on whichever entry is processed last.

konfai/data/transform.py (1)

1268-1270: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick win

Release _field_window after whole-volume sampling. Clear it in Resample.__call__ after _sample completes. Keep it between measured_region_source and stream_region; that streamed path shares the field read. The cached field is float32, not float64.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@konfai/data/transform.py` around lines 1268 - 1270, Update Resample.__call__
to clear self._field_window immediately after _sample completes, before
stream_region while preserving it between measured_region_source and
stream_region. Keep the existing shared-read behavior for the streamed path and
retain the field cache’s float32 representation.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@apps/impact_reg/impact_reg_konfai/impact_reg.py`:
- Around line 186-195: Update apps/impact_reg/impact_reg_konfai/impact_reg.py
lines 186-195 in _stage_group to collect entry suffixes into a set, reject
groups containing more than one storage form, and derive the returned format
token from the single collected suffix rather than the loop variable. Update
lines 226-234 so each case uses its own staging base, such as work /
"transform_h5" / case, preventing links for different forms from sharing the
same case directory.
- Around line 243-268: Update the function containing the staging write to wrap
the h5py creation and population block in try/except, unlink staging on failure,
then re-raise the original exception; keep os.replace for successful writes.
Revise its memory-usage documentation to state the one-slab peak applies only to
h5 and OME-Zarr inputs, while compressed MetaImage inputs may retain the full
field in resident memory.
- Around line 708-714: Update the transform handling in the evaluation flow
around _stage_group and _run_transform so .h5 inputs are not staged as
directory-backed Dataset roots while still selecting Dataset.H5File; preserve
.itk.txt through the existing SitkFile path, and add explicit handling that
opens the documented Transform.h5 location correctly before evaluation.

In `@docs/source/usage/large-images.md`:
- Line 86: Update the Resample documentation consistently: in
docs/source/usage/large-images.md at lines 86-86, qualify the adjacent
arbitrary-displacement fallback as unbounded, custom, or otherwise refused
displacement; in docs/source/troubleshooting.md at lines 106-106, add bounded,
recorded, or runtime-measured field-backed Resample to the streamable region
list.

In `@konfai/data/transform.py`:
- Around line 1795-1802: Update plan_note so the bound-less displacement note
does not return early before coverage is computed. Preserve that field note and
append or combine the partial-coverage note when applicable, while retaining the
existing disjoint-case handling and returning both notes for cases covering only
part of the target grid.

---

Nitpick comments:
In `@apps/impact_reg/tests/unit/test_orchestration.py`:
- Around line 248-258: The test coverage for _stage_group only exercises
repeated staging of one form; add a test that stages a single group containing
entries whose sources use different forms, such as DVF.mha and DVF.ome.zarr.
Assert the intended root/spec form behavior promised by _stage_group’s
docstring, including that the result does not depend incorrectly on whichever
entry is processed last.

In `@konfai/data/transform.py`:
- Around line 1268-1270: Update Resample.__call__ to clear self._field_window
immediately after _sample completes, before stream_region while preserving it
between measured_region_source and stream_region. Keep the existing shared-read
behavior for the streamed path and retain the field cache’s float32
representation.

🪄 Autofix

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: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 720638b9-311c-48cd-9510-f7c0a92112e7

📥 Commits

Reviewing files that changed from the base of the PR and between ba53ab5 and ed5054d.

📒 Files selected for processing (33)

• apps/impact_reg/impact_reg_konfai/cli.py
• apps/impact_reg/impact_reg_konfai/impact_reg.py
• apps/impact_reg/impact_reg_konfai/models/convexadam.py
• apps/impact_reg/impact_reg_konfai/models/elastix.py
• apps/impact_reg/impact_reg_konfai/models/elastix_engine.py
• apps/impact_reg/impact_reg_konfai/models/fireants.py
• apps/impact_reg/tests/unit/test_displacement_field_io.py
• apps/impact_reg/tests/unit/test_orchestration.py
• docs/source/concepts/streaming.md
• docs/source/config_guide/prediction.md
• docs/source/config_guide/transform.md
• docs/source/examples/visual-gallery.md
• docs/source/reference/components/transforms.md
• docs/source/troubleshooting.md
• docs/source/usage/large-images.md
• konfai/data/patching.py
• konfai/data/transform.py
• konfai/transformer.py
• tests/integration/test_konfai_streamed_prediction.py
• tests/integration/test_transform_example.py
• tests/unit/test_api.py
• tests/unit/test_case_expansion.py
• tests/unit/test_itk_transforms.py
• tests/unit/test_resample.py
• tests/unit/test_resample_to_reference.py
• tests/unit/test_resample_transform.py
• tests/unit/test_streamed_read_dispatcher.py
• tests/unit/test_streamed_write_dispatcher.py
• tests/unit/test_transform.py
• tests/unit/test_transform_locality_contract.py
• tests/unit/test_transformer_workflow.py
• tests/unit/test_warp.py
• tests/unit/test_write_pyramid_and_field_bound.py

💤 Files with no reviewable changes (1)

• apps/impact_reg/impact_reg_konfai/cli.py