jpmschweitzer/settled-reach
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

chore(meta): pql changelog — T-1157 review transition (write-through landed main-side via FR-4 vault resolution)

Browse Source 
Tyre's PR #212 finding 1: the review-status row existed only in a live
DB, unrecorded in any git-tracked changelog — a rebuild reverted it.
Root cause: pql run from the capture worktree resolved the MAIN vault
(FR-4), so the write-through row landed here, uncommitted. Committing
it main-side makes the transition rebuild-safe; branch and main
changelog rows union at merge.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
jpmschweitzer
2026-07-25 22:22:30 +02:00
co-authored by Claude Fable 5
parent ae45d5e21a
commit deeec94ec9
2 changed files with 18 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.pql/changelog/ticket_history/2026-07.sql
+1
View File
@@ -2604,3 +2604,4 @@ pair (GJ4056b, GJ660Ad) are alternates with maximum relief energy but
airless/thin atmospheres — shield/lava-field morphology, likely NOT
Alpine-classed; use only if the Active pair fails to produce MorphologyZone
13/14 crops. GJ0c/GJ0f-1 lack terrain_reference and cannot open in the Atlas.', NULL, '2026-07-25 19:00:39', '2026-07-25 19:00:39.731', '2026-07-25 19:00:39.731', NULL, '194ab5efa1ff3b059c0296830334f801', 2) ON CONFLICT(hash) DO NOTHING;
INSERT INTO ticket_history (ticket_record_id, field, old_value, new_value, changed_by, changed_at, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('06FRJD6EKPBQE08NA8VHP692W8', 'status', 'in_progress', 'review', NULL, '2026-07-25 20:13:40', '2026-07-25 20:13:40.919', '2026-07-25 20:13:40.919', NULL, '01a486c57de6c9de11129fde26a19a2f', 2) ON CONFLICT(hash) DO NOTHING;
.pql/changelog/tickets/2026-07.sql
+17
View File
@@ -4099,3 +4099,20 @@ pair (GJ4056b, GJ660Ad) are alternates with maximum relief energy but
airless/thin atmospheres — shield/lava-field morphology, likely NOT
Alpine-classed; use only if the Active pair fails to produce MorphologyZone
13/14 crops. GJ0c/GJ0f-1 lack terrain_reference and cannot open in the Atlas.', 'ready', 'low', NULL, 'client', NULL, '2026-07-25 16:14:35.308', '2026-07-25 19:00:39.730', NULL, 'c56c7d2a7a030ee4e959478459ee3d12', 2) ON CONFLICT(record_id) DO UPDATE SET type=excluded.type, parent_record_id=excluded.parent_record_id, title=excluded.title, description=excluded.description, status=excluded.status, priority=excluded.priority, assigned_to=excluded.assigned_to, team=excluded.team, decision_ref=excluded.decision_ref, updated_at=excluded.updated_at, deleted_at=excluded.deleted_at, hash=excluded.hash, canonical_version=excluded.canonical_version WHERE excluded.updated_at > tickets.updated_at OR (excluded.updated_at = tickets.updated_at AND excluded.hash > tickets.hash);
INSERT INTO tickets (record_id, type, parent_record_id, title, description, status, priority, assigned_to, team, decision_ref, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('06FRJD6EKPBQE08NA8VHP692W8', 'task', '06FBPPMZNNEV052DBYYY3A897C', 'Redesign the atlas visual-capture goldens for the zoom ladder (visual_scenarios/visual_capture)', 'T-1153 broke the T-1120/T-1121 golden capture harness: client/tests/visual_scenarios.gd and visual_capture.gd call app._regional_screen._viewer.get_heightmap_texture()/show_body() — both crash against AtlasWindowViewer (regional_screen now wraps the ladder viewer; AtlasViewer is unreachable from nav). Confirmed NOT wired into tests/run-godot or the pre-push hook (no reference in either), so nothing blocks, but the 12 curated atlas_* goldens and the --atlas-matrix capture path are dead until redesigned for the ladder (capture at defined rungs/zooms instead of the fixed heightmap frame; the new canonical orbital frame is the natural replacement for the old whole-body shot). Also decide the fate of the retired AtlasViewer code these harnesses were the last live consumers of — likely deletable once T-1156 (overlay port) settles what carries forward.
Live round 4 addendum: client/tests/test_atlas_window_overlay_draw_smoke.gd (3 scenarios -> now 2 after removing a flaky harness-sanity check: single-window draw + tile-mosaic draw, both asserting real composited pixels via SubViewport render) is a standalone gdUnit file with its own headless-skip guard, added because this redesign hadn''t landed yet. When this ticket lands, migrate its 2 scenarios into the redesigned tests/visual.json harness (folding the standalone file away) for consistency with the rest of the real-rendering test suite — see docs/DEVOPS.md''s ''Real-rendering test exception'' note for the interim rationale.
RE-SCOPED by D-255 (2026-07-24, body-map-viewer workshop): this becomes the capture harness for the STEPPED render architecture — goldens key on (body, step) canvases, not the retired continuous-zoom frames. The T-1152/T-1153 test suites (coverage-walk, continuous-zoom seams, compute_tile_grid mosaic) retire with their code in T-1182''s change; this re-scoped harness is their replacement and the eyeball-check substrate for every step canvas. More needed than ever — keep, don''t cancel.
PR #203 fix-round inventory (2026-07-25): the retirement is now TOTAL — AtlasViewer and its whole cluster (atlas_viewer, atlas_marker_overlay, atlas_descend_geometry, atlas_legend_panel, atlas_generation_proxy, atlas_generation_state) are deleted. visual_scenarios.gd (_setup_atlas_golden_shot, 6 of 12 atlas_GJ* z2_0/z4_0 non-fit scenarios) and visual_capture.gd (_run_atlas_matrix) now call StepCanvasViewer methods that do not exist: get_heightmap_texture(), set_view(zoom, offset), get_view_offset() — the fit-zoom shots early-return before hitting them. Ruled in PR #203: NO shim (would resurrect the retired continuous-zoom API D-255 kills); the redesign here keys captures on (body, rung) via StepCanvasViewer''s real surface (enter/_scroll_rung/rung state + the echoed extent). get_overlay_defs()/set_overlay_visible() duck-type fine and survive. Until this lands, those 6 goldens + atlas_matrix are knowingly dead (manual make targets only, gate-invisible — confirmed again).
Harness techniques inventory addition (from the PR #204 / T-1183 restart-persistence eyeball, 2026-07-25) — proven live-rendering eyeball patterns for shared-desktop sessions:
1. InputSwallower pattern: root-level Node (added first, low process_priority), _input()+_unhandled_input() both call get_viewport().set_input_as_handled() unconditionally — neutralizes real desktop input reaching a `-s` SceneTree driver''s real window (X11/XWayland delivers pointer events to unfocused windows under the cursor). Verify via a per-capture-window intercept counter: N>0 intercepted AND zero resulting view drift is proof; 0 intercepted proves nothing.
2. In-process navigation via viewer methods, not OS input: call the same methods the app''s own gdUnit suite exercises (StepCanvasViewer.enter()/_scroll_rung(); RegionalScreen.enter() atlas-matrix precedent in visual_capture.gd) — identical production code path, no OS injection needed (Wayland/KWin blocks synthetic X11 injection for XWayland clients entirely).
3. View-transform logging at every capture: world_center, view_offset, the drawing Node2D''s canvas position, footprint_px alongside every screenshot — turns "pixels don''t match" into a root-causable mechanism (PR #204: drift matched PAN_SPEED_PX_S*frame_time*held-frames to 0.02px, identifying unfocused edge-scroll bleed, now T-1191).
4. Full-run-restart-on-anomaly, never partial retry: any mid-sequence assertion failure discards the whole run (cold-wipe for a cold-role attempt; fresh-process-only for a revisit-role attempt) — partial retries silently corrupt the state the comparison depends on.
5. Fixed-center revisit mode: record a prior run''s actual derived centers and re-request them literally on the comparison run, bypassing cursor-anchored re-derivation — required when run 2 must land on run 1''s exact cache keys, not merely repeat the gesture.
Harness-discipline gap found 2026-07-25 (Jeroen spotted a live orphan): agent-side kill-on-completion does not cover an agent CRASHING mid-capture (API error killed a session between server spawn and its own cleanup step — the server ran on unnoticed for ~25 min on the shared machine). Inventory addition: (6) the LEAD runs an explicit orphan sweep (ps for godot4 + settled-reach-server) at every cycle checkpoint — commit, merge, and after any agent dies abnormally — not only inside agent protocols; pr-process step 1a''s >5-min gdUnit grep does not match live-capture servers, so the sweep needs the server binary name too.', 'review', 'medium', 'hoshe', 'client', NULL, '2026-07-22 09:42:47.197', '2026-07-25 20:13:40.919', NULL, '14425bb157c10e8cd85ef57d16270b0e', 2) ON CONFLICT(record_id) DO UPDATE SET type=excluded.type, parent_record_id=excluded.parent_record_id, title=excluded.title, description=excluded.description, status=excluded.status, priority=excluded.priority, assigned_to=excluded.assigned_to, team=excluded.team, decision_ref=excluded.decision_ref, updated_at=excluded.updated_at, deleted_at=excluded.deleted_at, hash=excluded.hash, canonical_version=excluded.canonical_version WHERE excluded.updated_at > tickets.updated_at OR (excluded.updated_at = tickets.updated_at AND excluded.hash > tickets.hash);