Sunlight artifacts, and a double door's open leaf

[nicosandller]

Closes #206, closes #219.

Two separate bugs in the light geometry, one commit each.

#206 — the hard diagonal edge on a sun patch

A gradient's gradientTransform scales its semi-axes. The beam's falloff was being handed the gap's whole width for the across one, so the ellipse ran nearly twice as wide as the polygon carrying it. At the flank the light was still at about half strength and the polygon's straight edge cut it off there — the hard boundary circled in the issue.

Measured on a plan like the report's, a 160-wide window:

	value
ellipse semi-axis across	142.8
polygon flank at	75.2
so the cut lands at	0.53 of the ellipse radius, still ~47% opaque

Halving it puts the flanks where the constant's own comment always said they were: the light dies just inside the gap's own edges, so the patch has soft sides as well as a soft tip and nothing but the falloff bounds it.

It survived because the existing "runs its outline past the falloff" test only checked the two far corners. It now checks all four, which is the invariant that matters: the polygon may only clip the ellipse where the ellipse has already reached zero.

#219 — light through a double door with one leaf open

wallsLightPassesThrough cut a gap of the right size in the wrong place — length * fraction, always centred. For a double door with a sensor per leaf that is off by a quarter of the opening: with the left leaf open the card lit the middle, which is half of the leaf still shut and half of the one that is open.

The callback may now answer with a [start, end] span instead of a bare width. A number still centres, so every existing call — sunlight included — renders exactly as it did.

On a 40-wide double door centred at 500, first leaf open:

	wall pieces
before (centred)	0..490, 510..1000
after (the open leaf's half)	0..480, 500..1000

openingClearSpan works out where, and by construction agrees with openingClearFraction about how much — there's a test holding the two together, since a span wider or narrower than the fraction would leak light or lose it.

Only the hinged double moves. A slider's centred gap is just as approximate, but those want more than one interval per opening, which this function's shape still cannot say, and moving them would change every existing plan's lighting for a case nobody has reported.

Writing the placement turned up one bug the tests caught: a wall drawn right-to-left runs backwards under the same doorway, so a span taken as given has its ends cross over and the gap is dropped entirely — a door that stops passing light because of the direction its wall was drawn. Projected and ordered now, with a test.

The other half of #219

The reporter also asked for glazed to work for a lamp and not only the sun. That already landed in #215, so this PR leaves it alone — a glazed door passes a lamp's pool on main today. Worth saying in the issue when it closes.

Verification

• 6394 tests pass, tsc --noEmit clean. Both fixes fail their new tests when the source change is stashed — I checked, with the exact numbers above.
• Driven in a browser: sun beams at several bearings, hard flanks gone and the rounded tip kept; a lamp lighting the next room through a double door with the left leaf, the right leaf, and both open, the pool lining up with the drawn arc each time.

Sunlight still centres its own beam within the opening, so a partly open double door in direct sun keeps today's approximation. Moving that means moving the beam and the wall gap together, and the beam ids are position-stable per opening for a reason — a separate change, not a drive-by.

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

The fixes are narrowly scoped, maintain backward-compatible behavior for existing callers, and are protected by new regression tests covering both reported issues and key edge cases.

Pull request overview

This PR fixes two rendering/geometry bugs in the lighting system: (1) a sunlight falloff ellipse that was sized using the beam’s full width instead of its semi-axis width (causing hard diagonal clipping artifacts), and (2) wall “gaps” for light transmission that were always centered, which misplaces light passing through a hinged double door when only one leaf is open.

Changes:

• Correct sunlight falloff sizing by using half the beam width when computing the ellipse “across” semi-axis, preventing polygon edges from clipping non-zero falloff (#206).
• Extend wallsLightPassesThrough to accept either a centered width (number) or an explicit [start, end] span, enabling correct placement for per-leaf double doors (#219).
• Add targeted tests and documentation updates to lock in the corrected invariants and behavior.

File summaries

File	Description
src/render.ts	Adds span-based opening placement (openingClearSpan, glowClearSpan), updates wall-gap cutting to respect spans, and fixes sunlight falloff semi-axis scaling.
src/render.test.ts	Adds unit tests for span placement invariants and wall-gap cutting behavior for double doors and reversed wall direction.
src/render.opening.test.ts	Strengthens sunlight regression tests to ensure the beam polygon never clips non-zero ellipse falloff and validates semi-axis sizing.
src/floorplan-card.ts	Switches lamp/wall clearing logic from fraction-only to span-aware (glowClearSpan) for correct double-door behavior.
src/editor.ts	Mirrors the card’s span-aware wall clearing logic in the editor preview.
README.md	Documents that lamp pools follow per-leaf behavior for hinged doubles (gap placement matches the open leaf).

Review details

• Files reviewed: 6/6 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.