Files
MarkusRe bdfdc1fda1 solidcore: a plane meets a torus in its exact section in either order (#14246 batch A13c) (#15987)
* claim: #14246 A13c

Co-Authored-By: Claude Code <noreply@anthropic.com>

* Tests: a plane meets a torus in its exact circles on Kernel::default() (WR-721)

a_plane_meets_a_torus_in_its_exact_circles_in_either_order now runs on
Kernel::default() without setting plane_first_torus_section_exact, and
its half that pinned upstream's marched sections (1.36e-6 and 3.46e-6
off the circles) goes. On main it fails: the flag is off by default and
a box-first boolean marches the plane/torus section.

Co-Authored-By: Claude Code <noreply@anthropic.com>

* solidcore: a plane meets a torus in its exact section in either order (#14246 batch A13c)

Retires the UpstreamFixes flag plane_first_torus_section_exact (WR-721).
The fix becomes the only path; the flag, its preset line and its
fixes.json entry go.

- surface::plane::p2::global_surface_intersect_into dispatches a torus
  to Torus::intersect_with_plane_owned_into, the orders and UV curves
  swapped as for a surface of revolution, where upstream has no torus
  case and marches the section.
- SurfOfRevolution's plane section through the axis rotates a circle
  generator as the circle it is, so a torus's meridian keeps its kind.

Co-Authored-By: Claude Code <noreply@anthropic.com>

* Goldens: the plane/torus section is exact in either order (WR-721)

35 rows of 5 units move with plane_first_torus_section_exact retired;
only rows outside the goldens tolerance are taken.

- surface.torus (5), surface.surf_of_revolution_p1 (13) and _p2 (5):
  a plane section through the axis is a circle (ellipse kind) instead
  of a B-spline with the same NURB; surface.torus also writes the
  circle's frame and radii (gen0177: 126 -> 150 values).
- topology.face_p3 (3): touching_torus_plane_circle_10_* (plane first)
  now equal the _01_ rows (torus first).
- boolean.api_brep (9): the #383 tangent-torus boxes and the horn torus
  box; volumes and face areas move within the measuring accuracy; seed
  4's intersection has 9 edges and 6 vertices instead of 10 and 7, and
  seed 86's difference lists its two pieces in the other order.

The WR-721 addendum records the retirement and the moved rows.

Co-Authored-By: Claude Code <noreply@anthropic.com>

* WR-721 addendum: the meridian change covers any ellipse-kind generator

Co-Authored-By: Claude Code <noreply@anthropic.com>

* Findings: the 35 goldens rows WR-721's retirement moves, old and new

Co-Authored-By: Claude Code <noreply@anthropic.com>

---------

Co-authored-by: Claude Code <noreply@anthropic.com>
2026-11-13 12:46:32 +00:00
..

Findings

One file per finding: Rust-vs-C++ disagreements, upstream bugs, nondeterminism and other facts later agents need. Every PR adds its own new file, so parallel PRs never conflict. Never append to a shared file. Upstream C++ defects (UB, crashes, hangs, nondeterminism, wrong results, wrong docs) also get their own file under plan/upstream-issues/ for the upstream maintainers.

  • Name: <unit id or general>--<short-slug>.md, lowercase, e.g. topology.brep_p3--turn-to-nurbs-uv-trim-notify.md. Pick a slug nobody else will.
  • Content: start with # <unit>: <title> (<YYYY-MM-DD>), then the unit, inputs, both values, suspected cause and the PR.
  • Records are not edited later (like ADRs). To update, resolve or supersede a finding, add an addendum: a new file <same name without .md>--addendum-<YYYY-MM-DD>-<slug>.md whose first line after the title is Updates: <path>. Fixing a typo in your own PR's file before it merges is fine.
  • References from code, crates/difftest/known_mismatches/<unit>.toml or unit notes give the file path, e.g. plan/findings/topology.brep_p3--turn-to-nurbs-uv-trim-notify.md.
  • Older references to plan/findings.md: <heading>: grep the heading under this directory (grep -rlF '<heading>' plan/findings/). The heading is the first line of its file.

History

Until 2026-10-25 findings were sections of one shared file, plan/findings.md. On 2026-10-25 scripts/split_archives.py split it into one file per ## section here (<unit or general>--<slug>.md, unit taken from the heading, -2, -3, ... for repeated names) and deleted it; python3 -I scripts/split_archives.py check proves the split lossless. Each file starts with the section's heading as # <heading>, followed by the section verbatim. Their names do not carry the archive order. The archive's introduction read:

Frozen archive (2026-10-25). Do not add entries here: every new finding is its own file under plan/findings/, so parallel PRs never conflict.

Rust-vs-C++ disagreements, upstream bugs, nondeterminism and other facts later agents need. Add entries under a heading per unit (or "General"); keep the newest at the end of each section.

Upstream C++ defects only (UB, crashes, hangs, nondeterminism, wrong results, wrong docs) are also logged, one entry each, in plan/upstream-issues.md for the upstream maintainers.