fix: 🐛 pcb layers tab

docs/features/wx-dom-port/visual-notes.md

@@ -235,3 +235,34 @@ trees fault identically; binary timing decides whether the race is lost,
 so any rebuild can flip these specs. They were red before this
 consolidation (the "31 passed" gates) and stay red after it; greening
 them is the async feature's exit criterion.
+
+## Layers tab goes blank after a tab round-trip (2026-06-15, pcbjam#8)
+
+Switching the pcbnew appearance notebook Layers → Objects/Nets → Layers left
+the Layers page blank: every layer row was present in the DOM but fully
+clip-pathed away, and the layers `wxScrolledWindow` viewport had collapsed to
+zero height (`154x0`).
+
+Root cause (two layers):
+  30. KiCad's `APPEARANCE_CONTROLS::OnNotebookPageChanged` calls
+      `m_panelLayers->Fit()` on every page change, which shrinks the page to
+      its content min size and collapses the proportion-1 scrolled child's
+      viewport to height 0. `wxBookCtrlBase::DoSetSelection()` fires
+      `wxEVT_NOTEBOOK_PAGE_CHANGED` LAST, so this happens after the page was
+      sized/shown.
+  31. `wxWindowWasm::DoSetSize()` only emits `wxEVT_SIZE` (which drives
+      auto-Layout) when the size actually changes. Resizing the page back to
+      the page area is therefore a no-op that never re-runs the sizer, so the
+      collapsed scrolled child sticks and `UpdateDomGeometry` clip-paths every
+      row out of the empty ancestor viewport.
+
+Fix: `wxNotebook::OnDomEvent` now calls `WasmRelayoutSelectedPage()` after
+`SetSelection()` returns (all PAGE_CHANGED handlers done) — it re-asserts the
+page to the full page area, forces `page->Layout()`, and re-projects the DOM
+geometry. wxwidgets-layer only; KiCad untouched.
+
+Test hardening: `appearance.spec.ts`'s `rowLabelTops()` previously returned
+`getBoundingClientRect().top`, which is `0` (not null) for a clip-pathed row,
+so the round-trip assertion passed even when the panel was blank. It now
+hit-tests the row centre via `elementsFromPoint` (clip-path affects
+hit-testing), so a clipped-away row reads as null and fails the assertion.

tests/kicad/appearance.spec.ts

@@ -67,8 +67,10 @@ async function selectTab(page: Page, label: string): Promise<void> {
     await page.waitForTimeout(400);
 }
 
-// DOM port only: viewport rects of row labels inside the appearance pane
+// DOM port only: viewport tops of row labels inside the appearance pane
 // (spans are real elements there; the canvas port draws them as pixels).
+// Returns the layout-box top regardless of clipping — used to track row
+// MOVEMENT under scrolling, where rows legitimately clip at the pane edges.
 async function rowLabelTops(page: Page, labels: string[]): Promise<Record<string, number | null>> {
     return page.evaluate((wanted: string[]) => {
         const out: Record<string, number | null> = {};
@@ -83,6 +85,39 @@ async function rowLabelTops(page: Page, labels: string[]): Promise<Record<string
     }, labels);
 }
 
+// Whether each row label is genuinely PAINTED (not merely present in the DOM).
+// A bare getBoundingClientRect is not enough: when the layers scrolled window
+// collapses on a tab round-trip the rows keep their layout box but are fully
+// removed by an ancestor's clip-path. clip-path also affects hit-testing, so we
+// sample points down each row's box and treat it as visible only if the row (or
+// its own content) is actually returned by elementsFromPoint somewhere. This is
+// what makes the "pages came back blank" regression fail the assertion below.
+async function rowsVisible(page: Page, labels: string[]): Promise<Record<string, boolean>> {
+    return page.evaluate((wanted: string[]) => {
+        const out: Record<string, boolean> = {};
+        for (const w of wanted) out[w] = false;
+        if (!window.wxDomControls) return out;
+        const visible = (el: HTMLElement): boolean => {
+            const cs = getComputedStyle(el);
+            if (cs.display === 'none' || cs.visibility === 'hidden') return false;
+            const r = el.getBoundingClientRect();
+            if (r.width < 1 || r.height < 1) return false;
+            const cx = r.left + r.width / 2;
+            const ys = [r.top + 1, r.top + r.height / 2, r.bottom - 1];
+            return ys.some(y => {
+                const hits = document.elementsFromPoint(cx, y);
+                return hits.some(h => h === el || el.contains(h));
+            });
+        };
+        for (const [, el] of window.wxDomControls) {
+            const txt = el.textContent || '';
+            if (el.tagName === 'SPAN' && wanted.includes(txt) && visible(el))
+                out[txt] = true;
+        }
+        return out;
+    }, labels);
+}
+
 test.describe('Appearance panel (Layers/Objects/Nets)', () => {
     test.beforeEach(async ({ page }) => {
         await page.goto('/kicad/pcbnew.html');
@@ -107,11 +142,13 @@ test.describe('Appearance panel (Layers/Objects/Nets)', () => {
         await selectTab(page, 'Layers');
         await page.screenshot({ path: 'test-results/appearance-03-layers-again.png' });
 
-        // Layer rows must survive the tab round-trip (regression:
+        // Layer rows must survive the tab round-trip (regression pcbjam#8:
-        // pages came back blank after switching away and back)
+        // pages came back blank after switching away and back — the rows stayed
-        const tops = await rowLabelTops(page, ['F.Cu', 'B.Cu']);
+        // in the DOM but their scrolled window collapsed and clip-pathed them
-        expect(tops['F.Cu'], 'F.Cu row visible after tab round-trip').not.toBeNull();
+        // all away, so this must assert real paint, not just DOM presence).
-        expect(tops['B.Cu'], 'B.Cu row visible after tab round-trip').not.toBeNull();
+        const vis = await rowsVisible(page, ['F.Cu', 'B.Cu']);
+        expect(vis['F.Cu'], 'F.Cu row visible after tab round-trip').toBe(true);
+        expect(vis['B.Cu'], 'B.Cu row visible after tab round-trip').toBe(true);
     });
 
     test('layer list scrolls with the wheel and clips at the pane', async ({ page }) => {