Skip to content

TrailView ground plan GP3 follow-ups: open decisions and review items #3132

Closed as duplicate

Description

@timohueser

GP3 (#3112) landed in ground-plan with PR #3131 (merge 61bfb3b). This issue holds the items that are still open after that PR.

Owner decisions (pictures on the morning page and in #3112 comment 6645427679):

  1. sr2: a junction guest that is wider than every other arm leaves a square-end tooth. The options are (a) taper the guest to the widest other arm (prototyped as gp3/v3/tools/guest-taper-try1.patch, together with a check that a point round's arc turns less than half a turn), (b) a land hull for overlapping rounds, (c) accept the corner and add a check for sharp convex corners.
  2. Class A needles: footways overlap the host's edge by 0.16–0.3 m and leave land needles. The options are (a) fold, (b) pave land holes narrower than the least inner radius (this acts as a closing), (c) count them as GP2 data faults and let GP4's sidewalk candidates take these footways.

Open items from the review (#3131 comment 6646966480; the #3112 body has the scenarios):

  1. A shared seam's far point can test lines that lie out to about 114 m from the window, outside the read box. Reading that far made rural cells up to ×2 slower, so it was taken out.
  2. Outline rounds depend on the order of the union's rings and on the asymmetry of the near filter. None of the four rules tried kept the fixture counts.
    Both items must be fixed before GP7 depends on the seams and the outline rounds.

Smaller items:

The leg split (question 4 of the round; GP3R #3135, request 5, gives it to this issue):

Mode: change #3096
Requirements: none


Generated by Claude Code

Activity

  1. timohueser commented on Nov 5, 2026

    @timohueser
    OwnerAuthor

    Status of the follow-ups (PR #3133, into ground-plan, not merged)

    Item State
    Open item 2: outline rounds and ring order (finding 3) Done in 7360404. The places are local (the corner within extent_m / 2 about each vertex), in the order of their turn and position; a place's round depends on the places within extent_m only (a place yields to the round of a sharper place within extent_m / 2 that is first). A round's feet lie within extent_m / 2, so every cell decides the places within 15 m of its window. Carriageways to lines_m = 65.0 m feed the paved outline; reach row faces outline. Two-cell test added.
    Open item 1: the far point's off-line test (finding 2, the rest) Done in f00740f. Ways read to ways_m = 114.1 m, reach row faces off-line test; per-16-segment carriageway boxes make the test cheap (item 2 alone ×1.015). Two-cell test added.
    Taper end on a bend (ck, n1) Open. Three tries, none kept: a relative narrowing (the carriageway gets too narrow where the widening falls off), a chord at the end (2 cm jogs, short features 162 to 163), a rate ladder (short features 160 but 2 new slivers). Pictures: /media/timo/OSM1/tmp/ground-plan/gp3/followups/pics/item3/, patch tools/taper-ladder-try.patch.
    Step where a zero-width shoulder starts (a1, a2) Open, not tried.
    New open point The outline places read the rounds of the points within input_m only; a point past it whose round runs through a place out to 35 m is not seen (not in the fixture).

    Fixture at the head: crashes 0, partition 0, check 2 0 (480 borders), kinks 15, short features 162, fragments 3, slivers 0, jogs 0, bent borders 96 (base: 163 short features, 2 fragments, 98 bent borders). Time ×1.087 against 61bfb3b at load 7 to 9 (item 1 about 6 %).


    Generated by Claude Code

  2. timohueser commented on Nov 5, 2026

    @timohueser
    OwnerAuthor

    Status after the review of PR #3133 (head 796b8bf, not merged)

    The reply to the review is #3133 (comment).

    Item State
    Order-free outline rounds: a test that the base fails Done (9c20ea5). It fails on 61bfb3b, it passes on the head.
    The outline read one radius short Done (f325e04). len + 2r; lines_m 70 m, row faces outline 140 m.
    Reads by segment boxes, one place definition, comments Done (36b6613, 777aedc, 9a21b77). The kink check now finds 18 kinks (3 corners that the old grouping hid). The plan does not change.
    Open item 2, rest: the fans in the paved outline Open, an owner decision. The outline rounds read the fans and circles of the points within input_m only. No rule fits under 191 m and keeps what the owner sees (rounding the points far enough needs 228 m). Options A to D with their cost are in the reply.
    Known limit: a chain of three places Recorded here. If g is first, h lies on g's round, i lies on h's round only and j on i's circle, then i and j are both made, and their circles can be near-equal (two arcs). A deeper chain moves the case to four places and adds 10 m of reach; a full cascade has no bounded reach. Not seen in the fixture (kinks 18, short features 162).
    Timing of the head Pending: the town-gate pair on a quiet machine.

    Fixture at the head: crashes 0, partition 0, check 2 0 (480 borders), kinks 18, short features 162, fragments 3, slivers 0, jogs 0, bent borders 96.


    Generated by Claude Code

  3. timohueser commented on Nov 5, 2026

    @timohueser
    OwnerAuthor

    Status after PR #3133 (merged, 84b1e4b):

    • Done: order-free outline rounds; the far point's off-line test (reads to 114 m with segment boxes); the outline read of len + 2r; segment-box way selection; one definition of a place, which reveals 3 more kinks: 18.
    • Recorded as known limits (owner: (A)): a round within 15 m of a border can depend on fans of points 46–75 m out (check 2 is the guard); near-equal circles in a chain of three places.
    • Cost: about +4 % per town cell (median per-cell ratio 1.037, low-load A B B A pair, ground-plan: GP3 follow-ups: order-free outline rounds, far ways for the off-line test #3133 comment 6652550899).

    Owner decisions:

    • sr2: (a). A guest wider than every other arm tapers to the widest other arm's width at the point, with the half-turn check on point-round arcs (prototype patch gp3/v3/tools/guest-taper-try1.patch). After it, junction hosts get the taper too.
    • Class A needles: (c). They are GP2 data faults until GP4's sidewalk candidates take these footways. No fold and no paved holes.
    • Band ownership: (a). One band candidate per way and layer, decided by key and global order like the carriageways. The step then lies on the losing way's carriageway edge.

    Still open: the taper end on a bend (ck, n1), and the a1/a2 shoulder step.


    Generated by Claude Code

  4. timohueser commented on Nov 5, 2026

    @timohueser
    OwnerAuthor

    Owner decision update:

    • Band ownership: parked. Three forms gave faults: per-way candidates made 5 slivers, 179 short features and +11 % town time; owners as labels on the union faces left 18 % of band patches mixed (PR ground-plan: a junction guest wider than every other arm tapers into them (GP3 follow-ups) #3134 body, gp3/bands/INDEX.md). It affects only a few height steps (about 5 of the 23 inside faces). It returns with GP4's sidewalk bands, which rebuild the bands anyway. One idea not yet tried: a point's region owns the bands up to its seams.
    • S taper: still wanted (owner decision), not built yet. The best of four tries (gp3/bands/tools/s-taper-try4.patch, tiles taper/s4-a.png) removes the corners at the taper ends but adds 77 short features and 1 sliver. They come from a hollow near the joint and from edges that wave on bends. Retry with a form that does not follow the varying own offset on bends. Until then, the straight taper stays.
    • PR ground-plan: a junction guest wider than every other arm tapers into them (GP3 follow-ups) #3134 now holds only the guest taper (sr2 (a)) and is under review.

    Generated by Claude Code

  5. timohueser commented on Nov 5, 2026

    @timohueser
    OwnerAuthor

    PR #3134 is merged into ground-plan: the guest taper (sr2 (a)) with the half-turn check, after the adversarial review (comment 6654394639) and green CI. Timing ×1.006 (median per-cell ratio 1.000).

    Open from the review:

    • Two guests both wider than the host (medium, owner decision, not caused by the PR). The taper aims at the widest other arm, guests included, so both square ends still stand out past the host's far edge. The options are (i) taper into the widest host instead (this costs a waist at an X junction split at the point), or (ii) record it as a known limit. Until the owner decides, it stays recorded as a known limit.
    • No test covers the half-turn check (it only acts on bent or tapered edges such as sr2b). Add one.
    • The short arc between the feet for an edge fit might make the check unnecessary and keep the class corner radius at sr2b (rule 3). Try it on the fixture.
    • corner_round repeats facing's arc (DRY).
    • List the case distinctions the taper adds in the TrailView ground plan GP3: faces of ways and junctions: carriageways, tight bends, junction areas, mouths, rail beds #3112 body (owner rule 1).

    Generated by Claude Code

  6. timohueser commented on Nov 6, 2026

    @timohueser
    OwnerAuthor

    Owner question at v12 (unequal fillet radii of the road from the right). Cause: a junction's arms end at the next graph point (faces/mod.rs:863-874 stops, faces/points.rs:826, :258). A driveway joins the secondary 3.3 m west on the north side only, but it ends the arm's run on both edges. The SW round (3 m requested: walk bit, min(3, 6)) must have its foot inside the run (round_at, points.rs:432-433), so the ladder shrinks it to 0.82 m; the SE round keeps 5 m. A throwaway build without run stops gives SW 3.00 m. General fix: end an arm's run at the next point only on the side where that point has an arm. Pictures: /media/timo/OSM1/tmp/ground-plan/diag/owner-q/ (q2-*).


    Generated by Claude Code

  7. timohueser commented on Nov 9, 2026

    @timohueser
    OwnerAuthor

    Cut (owner let the orchestrator decide, 2026-11-09): the S taper. Four tries failed the fixture counts. The straight taper stays as the end shape. The other items of this issue stay open: the a1/a2 step, the run stops at v12, the two wide guests.


    Generated by Claude Code

  8. timohueser commented on Nov 20, 2026

    @timohueser
    OwnerAuthor

    Closed in the backlog cleanup: the open items, together with those of #3166, are in #3743. Fixed: #3133, #3134, #3165. The S taper was cut by the owner on 11-09. #3743 starts with a picture pass on develop, because the junctions changed since (#3521, #3570).


    Generated by Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions