Roll out out-of-range coordinate validation to workers that bypass WorkerClient.process()

[arjunrajlab]

Background

#165 added out-of-range batch coordinate validation to WorkerClient.process(), so a Batch XY/Z/Time range naming coordinates the dataset does not have now fails with an actionable sendError instead of a bare KeyError from coordinatesToFrameIndex (self.map[channel][T][Z][XY]).

That covers only the six workers on the WorkerClient.process() path: cellposesam, cellpose, condensatenet, laplacian_of_gaussian, random_squares, sample_interface.

The workers that parse batch ranges and/or call coordinatesToFrameIndex() directly were never routed through that guard and still crash with a bare KeyError. This issue tracks extending the same protection to them.

This list originated in the (now closed) #143, whose branch carried it as todo/out-of-range-coordinate-validation-rollout.md. Moving it here so it survives the branch. Line references below were re-derived against master at 3f8f2b5, not copied from the branch.

Workers that need individual fixes

Each parses batch ranges with process_range_list / get_batch_information and then iterates coordinatesToFrameIndex() / getRegion() itself. Each should validate against IndexRange (then sendError and stop) before its processing loop.

Worker	Batch parse	Frame lookup
workers/annotations/cellori_segmentation/entrypoint.py	L86-88	L112, L117
workers/annotations/sam2_automatic_mask_generator/entrypoint.py	L86-88	via get_images_for_all_channels
workers/annotations/sam2_refine/entrypoint.py	L197-199	via get_images_for_all_channels
workers/annotations/sam2_propagate/entrypoint.py	L288-290	via get_images_for_all_channels
workers/annotations/sam2_video/entrypoint.py	L213-215	via get_images_for_all_channels
workers/annotations/sam_fewshot_segmentation/entrypoint.py	L294-296	via get_images_for_all_channels
workers/annotations/sam2_fewshot_segmentation/entrypoint.py	L273-275	via get_images_for_all_channels
workers/annotations/cellpose_train/entrypoint.py	—	L257, L263

Shared chokepoint — fixing it protects most of the SAM callers above:

• annotation_utilities/annotation_utilities/annotation_tools.py → get_images_for_all_channels(), coordinatesToFrameIndex() at L453.

Separate but adjacent exposure:

• workers/annotations/histogram_matching/entrypoint.py L138 — the reference-image lookup uses user-entered reference_XY / reference_Z / reference_Time number fields, which are not range-checked. fix: handle frames that omit index keys for size-one dimensions #159/fix(channelCheckboxes): report malformed values instead of crashing #162 validated this worker's channel selection (split_channel_selection, L122-133) but left the reference XY/Z/Time coordinates unguarded, so an out-of-range reference position still raises KeyError here.

Already validate — reference patterns, no change needed

• workers/annotations/registration/entrypoint.py — validates reference Z/Time against IndexRange; intersects parsed batch ranges with the valid ranges.
• workers/annotations/crop/entrypoint.py — intersects all three batch ranges with the dataset ranges.
• workers/properties/blobs/blob_intensity_worker/entrypoint.py and blob_annulus_intensity_worker — build range_z = range(0, IndexZ), filter Z planes, and sendWarning on out-of-range.

Open design question

The two policies in the tree disagree, and the rollout should settle it per worker rather than by accident:

• Strict error — what WorkerClient (Validate WorkerClient batch coordinates before image access #165) and crop/registration do: any out-of-range coordinate aborts the run.
• Filter and warn — what the blob intensity property workers do: drop the invalid coordinates, sendWarning, process the rest.

Strict is the right default for batch annotation workers (a partially-run batch that reports success is the failure mode #159 was chasing), but the property workers' filter-and-warn is deliberate and should probably stay.

Lower-risk, not in scope unless convenient

Preview/interface paths that load a single tile — vulnerable only if a user hand-edits batch params: random_point, random_point_annotation_M1, annulus_generator_M1, gaussian_blur, rolling_ball, deepcell, stardist, laplacian_of_gaussian (preview path).

Property workers that iterate annotation-sourced locations rather than user batch ranges are lower risk, since the locations come from existing annotations.

Related

• Validate WorkerClient batch coordinates before image access #165 — the WorkerClient fix this extends
• Report actionable error for out-of-range batch coordinates #143 — closed as superseded; this list came from its branch
• coordinatesToFrameIndex raises a bare KeyError for out-of-range coordinates NimbusImage#1185 — defense-in-depth at the lowest level, making coordinatesToFrameIndex itself raise a descriptive exception