[hjmjohnson]

Supersedes #6992 by @jdotc1, keeping that fix and its authorship. Narrow-band filters no longer read past the per-work-unit region list when the band has fewer nodes than work units, and stop iterating once the band is empty. Closes #6823.

Behavior change for maintainers to accept: when the band becomes empty, NarrowBandImageFilterBase::GenerateData stops: GetElapsedIterations() stays at the last count, and no IterationEvent is sent for that pass. At main the same input read an empty vector out of bounds, so no working behavior is lost.

Changes relative to #6992

• Squashed the two STYLE commits into the BUG commit; Jeff remains its author.
• Rewrote the BUG commit message with the what and why, including the SplitBand return-size change and the empty-band stop.
• Converted itkNarrowBandSmallBandTest to GoogleTest (itkNarrowBandSmallBandGTest), adding ITKGoogleTest to the module's TEST_DEPENDS.
• Named the image size, initial value, band row, first band column, and iteration limit; replaced the nested ternary with ExpectedElapsedIterations(); renamed Configure(count, shrink) to SetBand(numberOfNodes, emptyBandAfterFirstIteration) and RequestSplitRegion to GetSplitRegionOfWorkUnit.
• New commit: an IsoContourDistanceImageFilter test with a one-node band over eight work units. That filter also indexes SplitBand's result by work-unit id and was fixed silently by the padding.

Local verification

macOS arm64, AppleClang, RelWithDebInfo with AddressSanitizer; modules NarrowBand and DistanceMap plus their IO.

Test	main headers	this PR
NarrowBand.SplitBandReturnsOneContiguousRegionPerWorkUnit	fails: region count	pass
NarrowBandImageFilterBase.UpdatesBandWithFewerNodesThanWorkUnits	ASan SEGV	pass
IsoContourDistanceImageFilter.NarrowBandWithFewerNodesThanWorkUnits	ASan heap-buffer-overflow on a worker thread	pass

All 34 NarrowBand and DistanceMap tests pass at the tip, and pre-commit run --all-files passes.

AI assistance

Claude Code performed the review of #6992, the GoogleTest conversion, the IsoContour test, and the AddressSanitizer runs; I reviewed and directed each change.

[hjmjohnson]

@jdotc1 Some improvements on your proposed PR. Feel free to review this, or update your PR.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge; no actionable issues were found.

Summary

Fixes narrow bands with fewer nodes than work units.

• Every worker gets a region when a narrow band is smaller than the worker count.
• Narrow-band filters skip empty worker ranges and stop when no nodes remain.

No actionable issues found. Tests were not run during this review.

Acknowledged by hjmjohnson:

• An empty-band pass intentionally leaves GetElapsedIterations() unchanged and sends no IterationEvent.
• Raising the work-unit count with ManualReinitialization can still overrun m_TouchedForThread; this predates the PR and is deferred.

_{Reviews (1) · Last reviewed commit: "ENH: Test IsoContourDistanceImageFilter ..." · Reviewed by Greptile}