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 parts P7: the bake checks each placed part #3684
The ladder of P1 selects a part by rules and takes the first that fits. Nothing proves that the selected part works at its place: in the prototype the wall heights had a cap, so no check could fail. The epic wants the bake to build each place and to test real properties, and to take the next simpler part when a test fails.
Scope
For each placement, the bake builds the place with obc-trailview-parts and the ground plan crate, and tests: no ground and no object in the opening; no wall higher than the limit of its group (wall_max_m in parts.json); the part closed against the ground (no gap between the ground and the part's walls).
A failed check takes the next rung of the ladder. The free tube end is checked too; a failure there is counted and fails the gate of the bake. The floor clashes of P2 become a check.
A count over all tunnel ends of test box B replaces the count of 1,014 test entrances, with a gate of 0 for ground or an object in an opening.
The metric floor_clashes of the step's parts metrics counts each way on top once: a way whose window has a vertex where the shell of the crown lies over its underside. floor_clash_places lists the ten worst with their metres. The design of P2 counted per vertex; this is a deviation of P2.
Test box B: 45 clashes. The ten worst: 47.05029,8.31071 footway +3.52; 46.90218,8.62485 footway +3.39; 47.05440,8.29648 cycleway +3.28; 46.99359,8.30798 motorway +2.51; 47.03478,8.30284 steps +2.09; 47.09892,8.51729 footway +1.92; 47.07399,8.28581 service +1.46; 47.05051,8.31027 footway +1.25; 47.05040,8.30969 footway +1.04; 46.99800,8.31097 motorway +0.82.
Most are short footway and path underpasses whose floor rises steeply to an opening: the opening keeps its height, so the floor cannot go down in time.
In the two worst, the way "on top" lies at the height of the floor (its underside is about 0.4 m under the floor): a way that crosses a tunnel way near its end with no shared node. The check must expect such a crossing.
A way that has a vertex of a tunnel way at the crossing joins the tube and is not on top (also a deviation of P2).
Also in the metrics: low_stretches (132 on box B), low_stretches_at_least_height (33) and lowest_stretches.
P2 (#3679, #3737) after its review: the clashes for the checks
This replaces my earlier comment here. The step's parts metrics now give two counts that the checks of P7 must cover.
floor_clashes (and the ten worst in floor_clash_places): a way that lies over the final floor by more than the at-grade band, where the shell of the crown lies over its underside, once per way. A way on top is the stretch of the tube within its reach (its outer half-width + half + shell_m), up to the openings grown by its outer half-width, so a way across the mouth counts. Test box B: 252, of which 234 lie within 12 m of an opening. In 130 the way lies less than 2 m over the floor: a way beside or across a portal whose band reaches over the tube's shell. The count also holds the ways that the road solver made at grade with a tunnel that it lifted to them (a hump of 7 m under Lucerne, 47.0484,8.2988).
What the metres mean. For a way beside the tube, the metres of a clash are the crown's top over the way's underside, not the depth of the cut at the side of the tube (tube 320, rail: a road 5.5 m off the axis and 0.66 m over the floor gives +6.84 m). Do not read them as cut depths.
tube_clashes (and tube_clash_places): a tube over a tube. P2 does not lower a floor for another tube; it counts where the other tube lies over the floor by more than the band and the shell of the crown lies over the other tube's floor less slab_m. Test box B: 112. In 82 the other tube lies less than 2 m over the floor: two tubes side by side, as in the pairs that overlap (TrailView parts P6: the pair of tubes, the opening in rock and the free tube end #3683). In 30 it lies 2 m or more over it, for example 46.97687,8.30962 (motorway link) and 46.97789,8.31106 (narrow gauge). The gate of P7 must cover these tube clashes too, not only floor_clashes.
The bake's option parts holds groups and bake only (the crate's Ladder). The builder takes design, looks and inside from Rules::embedded() of obc-trailview-parts. When the bake builds the parts for its checks, a change of design changes the result of the checks: then the option must hold design too, or the change must bake the layer again in another way.
Two arch ends of twin tubes that are not a pair each get wings from the ground alone. Their wings can cross the other head wall: axes 15 m apart, frame edges at 7.1 m, wings up to 8 m. The checks of the bake should find a wall of one part inside the footprint of another.
From P4 on, the bake writes the cut of each built end (the tag cut): where the ground of the ground plan lies inside a box of the part. The device's mesh leaves a hole only there.
This is correct only where the bake's ground and the drawn ground agree, at the edge of the cut, within the boxes' margin to the faces of the part. That margin is 0.11 m (specs/trailview-features.md, Tunnel parts), not half the wall. P7 must check that per opening:
Build the cell's mesh with the CPU heights of the ground plan, the ends of BUILT open, as the device builds it.
Test the corners and the edge middles of the hole against the band of ±0.11 m round the bake's ground there.
Count a miss, and let a miss fail the gate of the bake.
The bake already builds each built end's boxes (end_boxes), and outside(box, point) gives a point's depth in them. The cut pass reads the ground of the ground plan over the boxes. The check of the 0.11 m band can reuse that grid.
A local census of the crate (not committed) tested two things on test box B: whether each point of the room lies in a box, and whether each point of a box lies inside the shell as the faces draw it. After fix round 1 no sample fails. TrailView parts: the room folds in a tight bend, and a head wall under a low deck is lower than its opening #3762 holds the faults of P3's geometry (a fold in a tight bend, a head wall under a low deck); P7 can count such ends.
The band of 0.11 m is inward. The boxes lie at least 0.11 m inside the part's inside faces. Toward the outside faces the margin can be less: about 3 cm to 6 cm at the shoulders of a wide arch (specs/trailview-features.md now says so). A check of the drawn ground against the bake's ground uses the inward band.
The device's ground near an opening is the stamped ground. With the ground plan, a built tunnel line now stamps nothing from each opening's way end to the start of its part's cover (its road into its portal dug the hill in the part). Other stamps still act there and the bake does not read them: water beds, the trough under a deck, a gallery's fill.
The sampled cut misses the plan's steps. The bake reads the ground each 0.5 m and interpolates. Where the ground plan's ground steps (the road's end at the frame line, the road's walls and edges in the opening), the ring lies up to a sample step from the step, and a strip of ground with its wall stays in the opening. At the frame line the ring lies 1 cm to 4 cm in by construction (a point on the plane where two boxes meet counts as outside). Pictures with this fault: parts/p4b/pics/ (road2-eye, bend-mouth, lowered-eye). A check that builds the mesh and tests its triangles and step slots against the room finds these.
Open from P4 (b) #3763, for the checks: at the road tunnel's mouth in box B (p4b/pics/boxb/road-mouth-new.png on timosPC), a dark tip of about 15 px stands where the road arch meets the roof of the unbuilt footway's box (today's tunnel) in the same tube; it is a ground plan piece (gone with debug=lod), and neither the cut's walls nor its rule for steep ground take it (from inside: p4b/pics/diag/sh-in1.png).
From the delta review of #3763 (issuecomment-7059073771), left for later:
D2: the tail's gate on Structures.holds is per cell, not per box: where one box's structure does not draw yet, the ground stays in every room of that cell (as on develop, short-lived). Fix if it shows: a held flag per box in the cell's box buffer.
D3: in the fade band at the finest level 18 a strip vertex can move more than √2 grid steps (snap and morph), so a far strip triangle in the plain draw can keep ground in a room; the drawn point is tested, so nothing drops wrongly.
P5 (#3682): faults for the checks, from the underpass pictures
Pictures on the owner's PC: parts/p5/pics/ (INDEX.md). Box B, features revision 35 with box built.
A wall of the approach in the room. At the rail underpass under the junction at 47.05705,8.39532 (tunnel-end at world 561910920 377413528, face 9.64 m in a deep cut), the ground plan's cut wall of the railway stands in the opening's room at the right jamb. The box test drops its pixels in the boxes, so from the other opening it shows as a grey slab, dark shards and sky (rail-near, railfar-near). Develop's new roads show the same wall as masonry in that opening. The check "no object in the opening" must also test the ground plan's walls, not only its ground.
A gap at the threshold. Just past the front of the head wall, a thin band shows the sky from inside (path-inside, close-up diag/path-thr-in-new.png); develop's new roads show the same gap as two pale corners at today's portal (diag/path-thr-in-plan.png), and develop's arches as a thin line (P4 road-inside). Where the way in front lies some centimetres under the floor, the head wall's face under the sill shows as a pale step (path2-near).
Activity
timohueser commentedon Nov 20, 2026
P2 (#3679, #3737): the floor clashes, as built
floor_clashesof the step'spartsmetrics counts each way on top once: a way whose window has a vertex where the shell of the crown lies over its underside.floor_clash_placeslists the ten worst with their metres. The design of P2 counted per vertex; this is a deviation of P2.low_stretches(132 on box B),low_stretches_at_least_height(33) andlowest_stretches.Generated by Claude Code
timohueser commentedon Nov 20, 2026
P2 (#3679, #3737) after its review: the clashes for the checks
This replaces my earlier comment here. The step's
partsmetrics now give two counts that the checks of P7 must cover.floor_clashes(and the ten worst infloor_clash_places): a way that lies over the final floor by more than the at-grade band, where the shell of the crown lies over its underside, once per way. A way on top is the stretch of the tube within its reach (its outer half-width +half+shell_m), up to the openings grown by its outer half-width, so a way across the mouth counts. Test box B: 252, of which 234 lie within 12 m of an opening. In 130 the way lies less than 2 m over the floor: a way beside or across a portal whose band reaches over the tube's shell. The count also holds the ways that the road solver made at grade with a tunnel that it lifted to them (a hump of 7 m under Lucerne, 47.0484,8.2988).tube_clashes(andtube_clash_places): a tube over a tube. P2 does not lower a floor for another tube; it counts where the other tube lies over the floor by more than the band and the shell of the crown lies over the other tube's floor lessslab_m. Test box B: 112. In 82 the other tube lies less than 2 m over the floor: two tubes side by side, as in the pairs that overlap (TrailView parts P6: the pair of tubes, the opening in rock and the free tube end #3683). In 30 it lies 2 m or more over it, for example 46.97687,8.30962 (motorway link) and 46.97789,8.31106 (narrow gauge). The gate of P7 must cover these tube clashes too, not onlyfloor_clashes.Generated by Claude Code
timohueser commentedon Nov 20, 2026
P3 (#3680, #3754): the rules that the checks read
The bake's option
partsholdsgroupsandbakeonly (the crate'sLadder). The builder takesdesign,looksandinsidefromRules::embedded()ofobc-trailview-parts. When the bake builds the parts for its checks, a change ofdesignchanges the result of the checks: then the option must holddesigntoo, or the change must bake the layer again in another way.Generated by Claude Code
timohueser commentedon Nov 20, 2026
P3 fix round 1 (#3754): a case for the checks
Two
archends of twin tubes that are not apaireach get wings from the ground alone. Their wings can cross the other head wall: axes 15 m apart, frame edges at 7.1 m, wings up to 8 m. The checks of the bake should find a wall of one part inside the footprint of another.Generated by Claude Code
timohueser commentedon Nov 20, 2026
P4 (#3681): a check of the cut
From P4 on, the bake writes the cut of each built end (the tag
cut): where the ground of the ground plan lies inside a box of the part. The device's mesh leaves a hole only there.This is correct only where the bake's ground and the drawn ground agree, at the edge of the cut, within the boxes' margin to the faces of the part. That margin is 0.11 m (
specs/trailview-features.md, Tunnel parts), not half the wall. P7 must check that per opening:BUILTopen, as the device builds it.Generated by Claude Code
timohueser commentedon Nov 20, 2026
P4 (a) (#3761): what the checks can use
end_boxes), andoutside(box, point)gives a point's depth in them. The cut pass reads the ground of the ground plan over the boxes. The check of the 0.11 m band can reuse that grid.Generated by Claude Code
timohueser commentedon Nov 20, 2026
P4 (b) (#3763): what P7 must check
specs/trailview-features.mdnow says so). A check of the drawn ground against the bake's ground uses the inward band.parts/p4b/pics/(road2-eye, bend-mouth, lowered-eye). A check that builds the mesh and tests its triangles and step slots against the room finds these.Generated by Claude Code
timohueser commentedon Nov 21, 2026
Open from P4 (b) #3763, for the checks: at the road tunnel's mouth in box B (
p4b/pics/boxb/road-mouth-new.pngon timosPC), a dark tip of about 15 px stands where the road arch meets the roof of the unbuilt footway's box (today's tunnel) in the same tube; it is a ground plan piece (gone withdebug=lod), and neither the cut's walls nor its rule for steep ground take it (from inside:p4b/pics/diag/sh-in1.png).Generated by Claude Code
timohueser commentedon Nov 21, 2026
From the delta review of #3763 (issuecomment-7059073771), left for later:
Structures.holdsis per cell, not per box: where one box's structure does not draw yet, the ground stays in every room of that cell (as on develop, short-lived). Fix if it shows: a held flag per box in the cell's box buffer.timohueser commentedon Nov 21, 2026
P5 (#3682): faults for the checks, from the underpass pictures
Pictures on the owner's PC:
parts/p5/pics/(INDEX.md). Box B, features revision 35 withboxbuilt.tunnel-endat world 561910920 377413528,face9.64 m in a deep cut), the ground plan's cut wall of the railway stands in the opening's room at the right jamb. The box test drops its pixels in the boxes, so from the other opening it shows as a grey slab, dark shards and sky (rail-near,railfar-near). Develop's new roads show the same wall as masonry in that opening. The check "no object in the opening" must also test the ground plan's walls, not only its ground.path-inside, close-updiag/path-thr-in-new.png); develop's new roads show the same gap as two pale corners at today's portal (diag/path-thr-in-plan.png), and develop's arches as a thin line (P4road-inside). Where the way in front lies some centimetres under the floor, the head wall's face under the sill shows as a pale step (path2-near).Generated by Claude Code