Repository navigation
Native neurite graph row navigation fails with an empty-array reduction #548
Description
Observed in the completed uncoached P001 run on immutable ordinary target08 (source0332a696, OpenHCS0.8.7). This is a specific native navigation defect, separate from known graph ROI source publication/rehydration in #404/#394 and unit-frame work #540/#541; check those active owners before implementation.
Frozen evidence: /home/ts/wt/openhcs-issue-batch-20260929/next-ownp00195-after08-20261003/P001_OWN95/author-workspace/output. final/REPORT.md and final/freeze.json retain candidate02, execution355a5003-a190-4f30-be64-06d334db933d. Original runtime/mcp.stdin requests354 and357 call openhcs_navigate_viewer_window on the retained neurite_morphology Shapes route with data_index119, visible/selected true. The first supplies semantic axes; the second omits them. runtime/mcp.stdout around164444 and164490 preserves viewer_window_state_failed/viewer_window_navigation_failed with zero-size array to reduction operation minimum which has no identity. The graph had1526 rendered feature rows; the failure response instead reports empty layers/row count and no selected identity. The author attempted calibrated FITC context correction without successful selected-row linkage; all original outcomes remain retained. No failed call has been replayed.
Diagnosis owner: Planck, coordinated by parent. First trace the original feature-row/geometry/frame owner and its complete consumers; determine wrong-plane request versus geometry-boundary failure before changing code. Do not add a second viewer state store, filename identity, arbitrary fallback or generic empty-array suppression. No competing #394 or #541 implementation.
Acceptance: on a fresh controlled source-bearing graph fixture through the installed MCP/native viewer path, choosing a persisted feature row preserves its row/edge/subject identity and native coordinates, navigates to the compatible source plane, and supports matched raw/result/combined review. Incompatible or missing geometry should fail with the owning structured error and accurate unchanged viewer state, not a bare NumPy reduction or fabricated empty-view state. Preserve original frozen P001 artifacts; no biological success is implied by the technical fix.
Activity
trissim commentedon Oct 3, 2026
Working source fix visible draft #549, headf1b9607f0; 24 real native-Qt source controls PASS, including exact empty-box mechanism reproduction and logical membership/Points/Shapes/source-frame regressions. Source mechanism: existing controller assigned all logically linked rows irrespective of native slice; native multi-shape interaction_box uses displayed_vertices and calls create_box(empty). Correction navigates queued source selection first, then derives each linked visual selection from native _indices_view through the existing selection authority. Canonical subject groups remain unchanged; requested source row cannot be silently dropped.
Exact frozen row119 stack was not retained, so the reduction site for that original run remains inferred—not retroactively proven. #541 units/#394 graph writer and codec untouched. Singer released these methods and owns #522 restore consumer migration; parent owns future ordinary package/public receiving. No frozen replay/live RPC/provider/native/catalog launch.
Merge pull request #549 from OpenHCSDev/fix/off-slice-linked-selectio…
trissim commentedon Oct 3, 2026
PR549 is merged and remote main is verified. Parent reviewed the complete selection-authority/controller change and the recorded 24 passing native Qt controls (real off-slice Shapes reduction, mixed Points/Shapes, frame-first navigation, unchanged logical subject membership). No hosted CI wait. Current frozen target08 scientific environments were not altered. Planck is building the separate ordinary-package public MCP receiving case on existing display88. That remaining native/public acceptance is why this issue is still open; source integration is no longer pending. The exact original row119 reduction site remains inferred because its old log had no stack. No frozen failed operation is replayed.
trissim commentedon Oct 3, 2026
Merged #549 now has actual installed public MCP/native receiving, not only24 source Qt checks: source-bearing Shapes archives on different Z planes preserve logical subject7 identity, select only compatible displayed polygons [0,1], clear off-slice native selection, and navigate the requested source row before group projection. Unrelated subject8 selects only [2]. Nine matched native raw/result/combined PNGs personally inspected; exact typed close and all original frozen packet hashes PASS. Complete technical evidence linked in latest #549 comment and engineering548/public88-attempt01 output/PUBLIC-RECEIVING.rst. Original P001 graph row119 was not replayed/reconstructed; this verifies its shared native selection mechanism using the reviewed controlled fixture, not scientific analysis or separate graph publication/source-frame ownership.
trissim commentedon Oct 5, 2026
Merged PR #549 projects logical selections onto the displayed native slice and navigates the requested source row before group projection. The latest issue/PR comments record actual installed public MCP/native Shapes selection receiving, unchanged subject identity and nine matched captures. Original scientific row119 was not replayed; its exact historical stack remains unknown. All five reviewed merge commits are ancestors of current main (GitHub compare reports behind_by=0). The existing PR body now contains Closes #548. Closing this covered focused issue; no new runtime or scientific operation was performed.