You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
TrailView ground plan GP9 follow-ups: mouth seams, edge teeth, cut band #3166
Follow-ups of GP9 (#3103, PR #3164, merged into ground-plan). The live viewer draws the new ground near and far. These look faults remain. Each needs a general rule, not a per-place fix.
Seam lines at opened junction mouths (roads/corridor.wgsl:270-274). The open rule acts only past the frame's half-width. The pixels just inside keep the edge fray. Seen at v12 (parking aisle mouths) and v07 (left track inside the fan). Proposed fix: a per-face flag from GP3/GP7-C. A way face's side that touches a junction face opens.
Dark dashes and teeth along road and path edges (j-tt-a, hp, d-hairpin, conn-road-into-*, v04). These are not the step slots. Their cause is the face normals: the paved function normal at edge corners, or drawn_normal.
Light band with a toothed edge on cut slopes (main-road-eye, mountain-eye). This is GP8's model_ground. The shaped/cuts terms end at MASK_M and at the moved threshold on mask texels. One option: GP9 takes the cut/fill look from the corner label.
A dark wall-height step slab at conn-road-into-path.
v07's new fan is plain gravel without ruts, with a hard seam to the left track.
Optional, cost: a terrain-only restamp runs 5 lanes per corner but uses one (plan/corners.wgsl:237-249). It shows no cost on the GTX. Fix it only if the phone shows a cost.
Pictures: /media/timo/OSM1/tmp/ground-plan/gp9/fixes/INDEX.md, /media/timo/OSM1/tmp/ground-plan/gp9/live/INDEX.md. Reviews: PR #3164 comments 6675661949 and 6676387055.
Hairpins are a separate item (leg split, PR #3165, which waits for an owner decision on band ownership).
Added by the review of #3167: a hard, jagged light/dark line at the edge of a fill (eye/main-road-eye, right side). Cause: geometry, not look. The heights drop to the terrain over a narrow strip at the slope band's edge (moved goes from over 0.15 m to 0), and the mesh draws that drop with a sawtooth outline. This is GP6/GP7-C. The old dark look hid it. Pictures: /media/timo/OSM1/tmp/ground-plan/gp9/teeth/eye/main-road-eye.png, /media/timo/OSM1/tmp/ground-plan/gp9/review-teeth/pics/.
Items 2 and 3 are fixed in ground-plan with PR #3167, after the review and its fixes. The ground terms take the nearest way from the mesh corners near the eye and from the mask far away, with one rule in both: a path shapes the ground only within PATH_SHAPED_M. The edge dashes, v17's ridge lines and the light band on cuts are gone. Cost: about +0.1 ms a frame. The j-tt-a teeth are a sun shadow (develop has them too). Open: items 1, 4, 5, 6 and the fill-edge step (comment above).
Item 1 is in draft PR #3168 (into ground-plan). Deviation: the item proposed a per-face flag from GP3/GP7-C. The fix does not use one. A piece's pixel near its frame's edge reads the records' paved outline from the paint texture, which the far look reads too. The pixel opens its edge where the paved area goes on past the frame's edge. There is no format change.
The fill-edge line (comment above) has a diagnosis in the PR, but no fix. It is GP6's toe round, not GP7-C. The round lifts the fill about 0.24 m above the terrain, and the lift collapses in 0.5 m where the room to the slope band's outer edge caps the toe width. A fix needs an owner decision on the rule. Pictures: /media/timo/OSM1/tmp/ground-plan/gp9/mouths/INDEX.md.
Item 1 is fixed in ground-plan with PR #3168 (review: merge, no must-fix). A piece pixel within 0.6 m of its frame's edge opens its edge by how far the paved area goes past it, read from GP8's paint. Edge lines now end at mouths; the seam lines at v12, v11 and v07 are gone. Old path unchanged; cost about +0.01 ms. Open: items 4, 5, 6 and the fill-edge crease (GP6 toe round; owner decision, options on the morning page).
Owner question at cut-region (two paths on develop, one in the old render): no way is missing. The old GP9 first-step export drew a trail inside a junction in its material line's frame, so the second path's tread read as grass. The live viewer draws both paths (plan/corners.wgsl:239). Still wrong live: a uniform stone gore at the region's west end with a thin shard line, and a smeared patch with a notch at the east end (top view). Cause not traced (GP3 region/round or GP11 junction look). Pictures: /media/timo/OSM1/tmp/ground-plan/diag/owner-q/ (q1-*).
The road hairpin wall (v17, i07) is the best solution so far and the road looks clean, but the wall needs work: its top has steps, and the stone pattern repeats visibly. It must look as integrated into the terrain as possible. To do after ground-plan GP3 follow-up: split a way into legs where its parts touch at heights a step apart #3165 merges: a smooth wall top and non-repeating stone mapping on step slots.
Owner feedback on the leg split preview (#3165), 2026-11-06: at the v17 island the split draws two walls (one at each leg's edge) with a grass shelf between them; it must be ONE solid wall whose top meets the upper road. Sent to #3165 as a must-fix before merge.
Owner decision (2026-11-06): merge the leg split #3165 with one solid wall between legs of one way (i07, the junction corner at 46.82475, 8.40130) and no walls on paths. The v17 island lies between two DIFFERENT ways (they meet at a junction above it) and still shows two walls; a follow-up PR extends the rule "the higher side owns the ground between; one wall at the lower edge" to pairs of different ways of wall classes. Also open: the u-loop dimple (46.8246, 8.4042).
Owner feedback on conn-track-off-road-eye (ground plan, live viewer; also present on ground-plan 649f50e): a short stub of stone wall stands at the road edge just before the track mouth and ends abruptly. The owner: it "looks weird and should probably not be there".
For the wall-look work after #3165 (smooth top, no visible repeat), add: a road wall must not stand as a short stub. A wall that is too short to read as a retaining wall becomes a bank, and every wall end goes down into the bank smoothly. This must be one rule value for all walls, with no case for mouths.
Closed in the backlog cleanup: items 1–3 and the fill-edge crease are fixed (#3167, #3168, #3169). Items 4–6 and the review leftovers (two walls between different ways, the u-loop dimple, the short wall stub) are in #3743, together with those of #3132.
Activity
timohueser commentedon Nov 6, 2026
Added by the review of #3167: a hard, jagged light/dark line at the edge of a fill (eye/main-road-eye, right side). Cause: geometry, not look. The heights drop to the terrain over a narrow strip at the slope band's edge (moved goes from over 0.15 m to 0), and the mesh draws that drop with a sawtooth outline. This is GP6/GP7-C. The old dark look hid it. Pictures:
/media/timo/OSM1/tmp/ground-plan/gp9/teeth/eye/main-road-eye.png,/media/timo/OSM1/tmp/ground-plan/gp9/review-teeth/pics/.Generated by Claude Code
Merge pull request #3167 from timohueser/claude/gp9-teeth
timohueser commentedon Nov 6, 2026
Items 2 and 3 are fixed in
ground-planwith PR #3167, after the review and its fixes. The ground terms take the nearest way from the mesh corners near the eye and from the mask far away, with one rule in both: a path shapes the ground only withinPATH_SHAPED_M. The edge dashes, v17's ridge lines and the light band on cuts are gone. Cost: about +0.1 ms a frame. The j-tt-a teeth are a sun shadow (develop has them too). Open: items 1, 4, 5, 6 and the fill-edge step (comment above).Generated by Claude Code
timohueser commentedon Nov 6, 2026
Item 1 is in draft PR #3168 (into
ground-plan). Deviation: the item proposed a per-face flag from GP3/GP7-C. The fix does not use one. A piece's pixel near its frame's edge reads the records' paved outline from the paint texture, which the far look reads too. The pixel opens its edge where the paved area goes on past the frame's edge. There is no format change.The fill-edge line (comment above) has a diagnosis in the PR, but no fix. It is GP6's toe round, not GP7-C. The round lifts the fill about 0.24 m above the terrain, and the lift collapses in 0.5 m where the room to the slope band's outer edge caps the toe width. A fix needs an owner decision on the rule. Pictures:
/media/timo/OSM1/tmp/ground-plan/gp9/mouths/INDEX.md.Generated by Claude Code
timohueser commentedon Nov 6, 2026
Item 1 is fixed in
ground-planwith PR #3168 (review: merge, no must-fix). A piece pixel within 0.6 m of its frame's edge opens its edge by how far the paved area goes past it, read from GP8's paint. Edge lines now end at mouths; the seam lines at v12, v11 and v07 are gone. Old path unchanged; cost about +0.01 ms. Open: items 4, 5, 6 and the fill-edge crease (GP6 toe round; owner decision, options on the morning page).Generated by Claude Code
timohueser commentedon Nov 6, 2026
Owner question at cut-region (two paths on develop, one in the old render): no way is missing. The old GP9 first-step export drew a trail inside a junction in its material line's frame, so the second path's tread read as grass. The live viewer draws both paths (
plan/corners.wgsl:239). Still wrong live: a uniform stone gore at the region's west end with a thin shard line, and a smeared patch with a notch at the east end (top view). Cause not traced (GP3 region/round or GP11 junction look). Pictures:/media/timo/OSM1/tmp/ground-plan/diag/owner-q/(q1-*).Generated by Claude Code
timohueser commentedon Nov 6, 2026
The fill-edge crease (comment above) is fixed in
ground-planwith PR #3169 (GP6 toe fade, owner option 1).Generated by Claude Code
timohueser commentedon Nov 6, 2026
Owner feedback on the leg split renders (#3165), 2026-11-06:
Generated by Claude Code
timohueser commentedon Nov 6, 2026
Owner feedback on the leg split preview (#3165), 2026-11-06: at the v17 island the split draws two walls (one at each leg's edge) with a grass shelf between them; it must be ONE solid wall whose top meets the upper road. Sent to #3165 as a must-fix before merge.
Generated by Claude Code
timohueser commentedon Nov 6, 2026
Owner decision (2026-11-06): merge the leg split #3165 with one solid wall between legs of one way (i07, the junction corner at 46.82475, 8.40130) and no walls on paths. The v17 island lies between two DIFFERENT ways (they meet at a junction above it) and still shows two walls; a follow-up PR extends the rule "the higher side owns the ground between; one wall at the lower edge" to pairs of different ways of wall classes. Also open: the u-loop dimple (46.8246, 8.4042).
Generated by Claude Code
timohueser commentedon Nov 6, 2026
Owner feedback on
conn-track-off-road-eye(ground plan, live viewer; also present on ground-plan 649f50e): a short stub of stone wall stands at the road edge just before the track mouth and ends abruptly. The owner: it "looks weird and should probably not be there".For the wall-look work after #3165 (smooth top, no visible repeat), add: a road wall must not stand as a short stub. A wall that is too short to read as a retaining wall becomes a bank, and every wall end goes down into the bank smoothly. This must be one rule value for all walls, with no case for mouths.
Picture:
/media/timo/OSM1/tmp/ground-plan/gp4/m/crops3/conn-track-off-road-eye.png(develop | ground-plan | stop-mouth branch).Generated by Claude Code
timohueser commentedon Nov 20, 2026
Closed in the backlog cleanup: items 1–3 and the fill-edge crease are fixed (#3167, #3168, #3169). Items 4–6 and the review leftovers (two walls between different ways, the u-loop dimple, the short wall stub) are in #3743, together with those of #3132.
Generated by Claude Code