From 573ef7a83f477fb72bdb84dadab07c83c85199a2 Mon Sep 17 00:00:00 2001
From: Boris Tyshkevich <btyshkevich@altinity.com>
Date: Tue, 4 Aug 2026 18:06:20 +0200
Subject: [PATCH] test(#599): fix resize-then-measure race in
 inspector-dock-layout and tile-open-workbench specs
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Two e2e specs raced an async settle with no wait, both surfaced by #587's
"surface out-of-scope findings" rule and reproduced again on CI for #601.

- inspector-dock-layout.spec.js:155 read `.inspector-host`'s boundingBox()
  immediately after `page.setViewportSize` — but the displayed width is
  recomputed by app-shell.ts's `reclampInspectorWidth`, a real `window`
  'resize' event LISTENER dispatched asynchronously relative to
  `setViewportSize`'s own resolution. Under `--repeat-each=15 --workers=12`
  this reproduced the exact CI signature ("Expected 320, Received 500");
  wrapping the read in `expect.poll` closes it (confirmed: 0 failures across
  the same stress level, run twice, after the fix).

- tile-open-workbench.spec.js:364 read the committed `window.__dashboard()`
  state immediately after `widen.click()` — but the widen press applies
  OPTIMISTICALLY first (`runCommand`, src/ui/dashboard.ts) while the actual
  persisted commit is a separate, fire-and-forget `app.mutateWorkspace` call.
  The geometry assertions right above it are safe (same optimistic doc, no
  gap); only the persisted-state read raced. Same `expect.poll` fix, mirroring
  the pattern the file's own later "narrow tile" test already uses for the
  identical read.

Audited both files in full for the same shape (state/viewport change
immediately followed by a bare geometry or persisted-state read); every other
instance is either already polled or gated behind a prior polling assertion
whose pass already implies the read is safe, so no other site needed a
change. No production code touched — the optimistic-apply-then-commit
behavior powering both races is working as designed.

Verified: full gate green (types/arch/schemas/examples/unit/build); both
specs run 8x in isolation with 0 failures; the fixed assertions stress-tested
at `--repeat-each=15 --workers=12` (chromium+webkit) with 0 failures across
multiple rounds; full parallel suite (`--project=chromium --project=webkit`)
run 3x, 414 passed/4 skipped every time.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R32bb4VZGPgNo3tB9iVSKF
---
 tests/e2e/inspector-dock-layout.spec.js | 15 +++++++++++----
 tests/e2e/tile-open-workbench.spec.js   | 10 +++++++++-
 2 files changed, 20 insertions(+), 5 deletions(-)

diff --git a/tests/e2e/inspector-dock-layout.spec.js b/tests/e2e/inspector-dock-layout.spec.js
index 6bd3702a..f175c729 100644
--- a/tests/e2e/inspector-dock-layout.spec.js
+++ b/tests/e2e/inspector-dock-layout.spec.js
@@ -153,11 +153,18 @@ test.describe('docked right-inspector dock-aware width (#586 findings 2a/2b)', (
     expect(Math.round(before.width)).toBe(500);
 
     await page.setViewportSize({ width: 900, height: 800 });
+    // #599: `setViewportSize` resolves as soon as the viewport is applied, but
+    // the DISPLAYED width above is recomputed by app-shell.ts's
+    // `reclampInspectorWidth`, run from a real `window` 'resize' event
+    // LISTENER — dispatched asynchronously relative to that resolution, and
+    // more likely to lose the race under parallel load. Poll for the settled
+    // width instead of reading it on the very next microtask.
+    await expect.poll(async () => Math.round((await inspectorHost.boundingBox()).width))
+      // Default sidebarPx (248) + 2 handles (14) reserved, minus CENTRE_MIN_PX
+      // (320): ceiling = 900-262-320 = 318, below the shared 320 floor — clamp
+      // floors it at 320 exactly.
+      .toBe(320);
     const after = await inspectorHost.boundingBox();
-    // Default sidebarPx (248) + 2 handles (14) reserved, minus CENTRE_MIN_PX
-    // (320): ceiling = 900-262-320 = 318, below the shared 320 floor — clamp
-    // floors it at 320 exactly.
-    expect(Math.round(after.width)).toBe(320);
     expect(after.width).toBeLessThan(before.width);
 
     test.info().annotations.push(
diff --git a/tests/e2e/tile-open-workbench.spec.js b/tests/e2e/tile-open-workbench.spec.js
index 98a81c08..3dca7b99 100644
--- a/tests/e2e/tile-open-workbench.spec.js
+++ b/tests/e2e/tile-open-workbench.spec.js
@@ -361,7 +361,15 @@ test('widen doubles a grid tile\'s rendered width, preserves its height, then wr
   // until #564.
   expect(wide.width).toBeGreaterThan(before.width);
   expect(wide.height).toBe(before.height);
-  expect((await page.evaluate(() => window.__dashboard('sales'))).items['t-sales'])
+  // #599: the widen press applies OPTIMISTICALLY first (`runCommand`,
+  // src/ui/dashboard.ts) — the geometry above is already settled by the time
+  // `widen.click()` resolves, but the PERSISTED commit this reads is a
+  // separate, fire-and-forget `app.mutateWorkspace` call that lands later.
+  // Poll rather than reading it on the very next microtask, matching the same
+  // commit-republish race the later "narrow tile" test already guards
+  // against (its own "Read defensively" comment, below).
+  await expect.poll(async () => (await page.evaluate(() => window.__dashboard('sales')))
+    ?.items?.['t-sales'] ?? null)
     .toEqual({ grid: { span: 12, height: 2 } });
 
   // At the maximum the next press wraps to a single column.