Skip to content

The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.

7.0 KB
Raw Blame
import { expect, type Page, test } from '@playwright/test';

import { bootAuthenticatedPage, dismissWalkthroughIfPresent } from '../helpers/core-rpc';

// `bootAuthenticatedPage` runs in every `beforeEach` and costs 30-60s against a
// locally-built debug core — the sidebar suite's first test measured 59.1s
// against the config's 60s non-CI budget, and this suite's first two tests blew
// it outright ("Test timeout of 60000ms exceeded while running beforeEach").
// The work is the harness's, not the assertions': raise the ceiling here rather
// than in the shared playwright.config.ts, which is not this worker's to edit.
test.describe.configure({ timeout: 180_000 });

/**
 * Narrow-viewport behaviour of the root shell.
 *
 * Nothing in the repo resizes the window. Every existing Playwright spec runs
 * at the default 1280×720 and every vitest suite runs in jsdom, which has no
 * layout engine at all — `getBoundingClientRect()` returns zeroes there, so a
 * layout that overflows or collapses to nothing is undetectable by design.
 *
 * The assertions here are deliberately about containment rather than pixel
 * values: no horizontal overflow of the document, the primary surface keeps a
 * usable width, and the sidebar either adapts or gets out of the way — never
 * eats the content area.
 */

const sidebar = (page: Page) => page.locator('[data-testid="root-shell-sidebar"]');
const content = (page: Page) => page.locator('[data-testid="root-shell-content"]');

async function documentOverflowsHorizontally(page: Page): Promise<boolean> {
  return page.evaluate(() => {
    const el = document.documentElement;
    // 1px of slack for sub-pixel rounding on fractional DPR.
    return el.scrollWidth > el.clientWidth + 1;
  });
}

test.describe('App shell — narrow viewports', () => {
  test.beforeEach(async ({ page }) => {
    await bootAuthenticatedPage(page, 'pw-app-shell-responsive-user');
    await dismissWalkthroughIfPresent(page);
  });

  for (const [label, width, height] of [
    ['laptop', 1280, 720],
    ['small laptop', 1024, 768],
    ['tablet portrait', 768, 1024],
    ['large phone', 414, 896],
  ] as const) {
    test(`${label} (${width}×${height}): content stays on screen and the page does not overflow`, async ({
      page,
    }) => {
      await page.setViewportSize({ width, height });

      // Let the layout settle rather than sampling mid-transition.
      await expect.poll(() => page.evaluate(() => window.innerWidth)).toBe(width);

      await expect(content(page)).toBeVisible();
      const box = await content(page).boundingBox();
      expect(box).not.toBeNull();

      // The content surface must keep a usable share of the window. The bug
      // this catches is a fixed-width sidebar that does not shrink: the content
      // column gets squeezed toward zero while nothing visibly "breaks".
      expect(box!.width).toBeGreaterThan(width * 0.4);

      // And it must not be pushed off the right edge.
      expect(box!.x + box!.width).toBeLessThanOrEqual(width + 1);

      expect(await documentOverflowsHorizontally(page)).toBe(false);
    });
  }

  test('the sidebar never takes more than half a narrow window (#5907)', async ({ page }) => {
    // This test used to pin the OPPOSITE — it asserted `width > 414 / 2`, as a
    // deliberate characterization of the defect, with a note saying that adding
    // a viewport clamp should be the thing that makes it fail. #5907 added the
    // clamp, so this is that flip.
    //
    // Before: `clamp()` (`RootShellLayout.tsx:38`) constrained the width against
    // SIDEBAR_MIN_WIDTH = 188 and SIDEBAR_MAX_WIDTH = 420 and never against the
    // window, so the sidebar measured 224px of a 414px viewport — 54% — with a
    // hard 188px floor below that. `tauri.conf.json` declares the main window
    // `resizable: true` with no `minWidth`, so that was reachable by dragging.
    //
    // After: `max-w-[50vw]` on the sidebar column (`components/ui/Sidebar.tsx`).
    // A CSS max-width rather than a JS clamp, because the width arrives as an
    // inline style and nothing in the shell listens for `resize` — a JS clamp
    // alone changes nothing at all until some unrelated render happens.
    await page.setViewportSize({ width: 414, height: 896 });
    await expect.poll(() => page.evaluate(() => window.innerWidth)).toBe(414);

    const count = await sidebar(page).count();
    if (count === 0) return; // a shell that hides it outright is also fine

    const box = await sidebar(page).boundingBox();
    if (box === null || box.width === 0) return; // collapsed away entirely

    // 1px of slack for sub-pixel rounding on a fractional device pixel ratio.
    expect(box.width).toBeLessThanOrEqual(414 / 2 + 1);
  });

  test('the clamp is inert at desktop widths (#5907)', async ({ page }) => {
    // The clamp must not become a second, quieter way to shrink the sidebar on
    // a normal window. 50vw exceeds SIDEBAR_MAX_WIDTH (420) above an 840px
    // viewport, so at the 1280x900 default it can never bind — the stored width
    // decides, exactly as before.
    await page.setViewportSize({ width: 1280, height: 900 });
    await expect.poll(() => page.evaluate(() => window.innerWidth)).toBe(1280);

    // No early return here. Returning on a null or zero box let a failed
    // render or a collapsed sidebar pass as success — the very outcome this
    // test exists to notice (#5941, CodeRabbit).
    const box = await sidebar(page).boundingBox();
    expect(box, 'sidebar has no layout box at 1280px — it did not render').not.toBeNull();
    expect(box!.width).toBeGreaterThan(0);

    // Well clear of the 640px the clamp would allow here, and at or above the
    // 188px floor — i.e. the clamp is not what is deciding this width.
    expect(box!.width).toBeGreaterThanOrEqual(188);
    expect(box!.width).toBeLessThan(640);
  });

  test('resizing back to full width restores the layout', async ({ page }) => {
    // A one-way responsive breakpoint — narrow works, but widening leaves the
    // shell stuck in its compact form — is a real and easy regression.
    await page.setViewportSize({ width: 414, height: 896 });
    await expect.poll(() => page.evaluate(() => window.innerWidth)).toBe(414);

    await page.setViewportSize({ width: 1280, height: 720 });
    await expect.poll(() => page.evaluate(() => window.innerWidth)).toBe(1280);

    await expect(content(page)).toBeVisible();
    const box = await content(page).boundingBox();
    expect(box).not.toBeNull();
    expect(box!.width).toBeGreaterThan(1280 * 0.4);
    // Same right-edge containment the per-viewport cases assert. Added after a
    // mutation exposed the gap: forcing the content surface to a fixed 1600px
    // failed tests 1-4 on exactly this line and left THIS test green, because
    // it only checked the width and the document-overflow probe. The width
    // check passes at 1600 (it is a `>` bound) and the overflow probe never
    // fires, so without this line the test had nothing left that could fail.
    expect(box!.x + box!.width).toBeLessThanOrEqual(1280 + 1);
    expect(await documentOverflowsHorizontally(page)).toBe(false);
  });
});