fix(resample): NaN handling in adaptive sampling + L1 validity on same-size replacement

[jeandet]

Stacked on #34 (← #33 ← #32) — merge those first; this PR's diff then collapses to its own commit (575f666).

• optimizedLineData derived an interval's closing point as intervalFirst + intervalCount - 1, but NaN samples are skipped WITHOUT counting: with interior NaNs that index lands on a NaN, injecting a spurious (NaN, NaN) line break — phantom gaps in dense gappy data (the norm in space physics). The interval now tracks the index of the last accepted sample explicitly.

• The multi-column variant had two related NaN holes: an interval reset landing on a NaN sample seeded minVal/maxVal with NaN, poisoning std::min/std::max for the rest of the interval (envelope silently dropped); and a single-sample NaN interval appended an explicit break point where the adaptive contract (matching the single-column variant) is to skip NaN values — key gaps are detected separately.

  Reproducer: adaptiveSamplingInteriorNaNNoSpuriousBreaks (gap-free keys, every other value NaN → output contained NaN break points before).

• buildL1Cache reused an existing L1 when only sourceSize matched; the previously dead cachedKeyRange member now participates in the validity check (keyRange is O(1) on sorted sources), so replacing the data with an equally-sized batch can never reuse a stale binning.

Part of the 2026-06-09 audit (H4/M7). Last of the stack: #32 → #33 → #34 → this.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR tightens correctness and safety around resampling/pipeline caching, with a focus on NaN-handling in adaptive sampling and preventing stale cache reuse when data is replaced (even at the same size). It also adds/updates regression tests that exercise the previously failing edge cases (NaN holes, async teardown, scheduler shutdown, histogram key restrictions).

Changes:

• Fix adaptive line sampling to skip NaNs without introducing phantom line breaks; harden multi-column min/max tracking against NaN-seeded intervals.
• Strengthen L1 cache validity checks (key-range participates) and improve sorted-key range performance (keyRangeSorted).
• Improve async pipeline / scheduler teardown safety and add regression tests for deletion-with-jobs-queued and destructor queue dropping.

Reviewed changes

Copilot reviewed 25 out of 25 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/auto/test-waterfall/test-waterfall.h	Adds new waterfall regression test declarations for snapshot semantics and concurrency safety.
tests/auto/test-waterfall/test-waterfall.cpp	Adds tests for parameter changes applying immediately and for concurrent parameter/data changes during resampling.
tests/auto/test-scatter-rhi/test-scatter-rhi.cpp	Fixes a dangling-span test setup by moving owned vectors into setData.
tests/auto/test-pipeline/test-pipeline.h	Adds new pipeline regression test declarations (scheduler dtor, colormap2 teardown, histogram2d cache invalidation).
tests/auto/test-pipeline/test-pipeline.cpp	Adds regression tests for scheduler teardown behavior, colormap2 queued jobs after delete, and histogram2d restricted-range behavior + cache invalidation.
tests/auto/test-multi-datasource/test-multi-datasource.h	Adds adaptive-sampling NaN regression test declaration.
tests/auto/test-multi-datasource/test-multi-datasource.cpp	Adds regression test ensuring no spurious NaN breakpoints are emitted for interior NaNs.
tests/auto/test-datasource2d/test-datasource2d.cpp	Drains the async pipeline to avoid stack-use-after-return in a test that uses non-owning views.
tests/auto/test-datasource/test-datasource.h	Adds windowed valueRange equivalence test declaration.
tests/auto/test-datasource/test-datasource.cpp	Adds brute-force equivalence test for key-restricted valueRange vs binary-search windowed implementation.
src/plottables/plottable-waterfall.h	Refactors waterfall adapter into an immutable snapshot; routes parameter/data changes through adapter rebuild.
src/plottables/plottable-waterfall.cpp	Implements immutable adapter + rebuild flow and removes draw-time mutation.
src/plottables/plottable-histogram2d.h	Adds memoization for finite bounds (selectTest/keyRange) and cache invalidation hooks.
src/plottables/plottable-histogram2d.cpp	Implements bounds caching and adds linear restricted-range scan for unsorted histogram keys.
src/plottables/plottable-colormap2.h	Changes gap threshold API to rebake the transform and avoid capturing object memory in background jobs.
src/plottables/plottable-colormap2.cpp	Moves transform installation into a helper and rebakes it on threshold changes; ensures re-resample when needed.
src/datasource/soa-multi-datasource.h	Uses keyRangeSorted (sorted-key contract) for O(1)/O(log n) key range.
src/datasource/row-major-multi-datasource.h	Uses keyRangeSorted (sorted-key contract) for O(1)/O(log n) key range.
src/datasource/pipeline-scheduler.h	Adds shutdown flag to prevent queued work from running during destruction.
src/datasource/pipeline-scheduler.cpp	Drops queued work on destruction and rejects submissions during shutdown before joining the pool.
src/datasource/graph-resampler.h	Strengthens L1 cache reuse checks by incorporating cached key range (not just size).
src/datasource/async-pipeline.h	Replaces atomic destroyed-flag with a mutex-guarded destroy guard shared with jobs to close TOCTOU.
src/datasource/async-pipeline.cpp	Implements guarded destroyed-check + removes redundant destroyed-check in deliverResult.
src/datasource/algorithms.h	Adds binary-search windowing to valueRange with key restriction; fixes adaptive sampling closing-point selection with NaN skipping; hardens multi-column NaN min/max handling.
src/datasource/algorithms-2d.h	Uses keyRangeSorted for 2D xRange to avoid O(N) scans under the sorted contract.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[jeandet]

Already addressed — same finding as on #33, fixed by cfa000b ("clearing the source uninstalls the adapter too", reproducer clearingSourceEmptiesTheGraph), now merged on main. This PR is rebased on top, so the flagged code is no longer part of its diff.