[Hona]

Closes #58247
Related: #56970, #59229, #57732

Visual

<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 240 72" width="240" height="72" fill="none">
  <rect x="12" y="14" width="96" height="52" stroke="currentColor" stroke-opacity="0.35" />
  <rect x="124" y="6" width="104" height="60" stroke="currentColor" stroke-opacity="0.35" />
  <rect x="128" y="10" width="96" height="52" fill="currentColor" fill-opacity="0.14" stroke="currentColor" stroke-opacity="0.75">
    <animate attributeName="x" values="34;34;128;128" keyTimes="0;0.35;0.55;1" dur="3.2s" repeatCount="indefinite" />
    <animate attributeName="y" values="26;26;10;10" keyTimes="0;0.35;0.55;1" dur="3.2s" repeatCount="indefinite" />
    <animate attributeName="width" values="52;52;96;96" keyTimes="0;0.35;0.55;1" dur="3.2s" repeatCount="indefinite" />
    <animate attributeName="height" values="30;30;52;52" keyTimes="0;0.35;0.55;1" dur="3.2s" repeatCount="indefinite" />
    <animate attributeName="stroke-dasharray" values="3 2;3 2;0;0" keyTimes="0;0.35;0.55;1" calcMode="discrete" dur="3.2s" repeatCount="indefinite" />
  </rect>
  <rect x="128" y="10" width="96" height="5" fill="currentColor" fill-opacity="0.3">
    <animate attributeName="fill-opacity" values="0;0;0.3;0.3" keyTimes="0;0.35;0.55;1" dur="3.2s" repeatCount="indefinite" />
  </rect>
</svg>

Before	After

Summary

• Maximized state kept (desktop: maximized state lost when saved bounds exceed the display #58247): resolveWindowState reset the whole saved state when the normal bounds did not fit a display, isMaximized included. On a 1366×768 panel (1388×781 DIP) the 1280×800 default never fits, so the window opened windowed on every launch. Every fallback now keeps isMaximized and isFullScreen.
• Maximized on the saved display: normal bounds are only recorded while the window is restored. A window maximized on another display kept x/y on the first one, and maximize() used that display. While the saved displayBounds is still connected, the normal bounds now move into it, including displays at negative coordinates.
• Primary fallback: a window whose display is gone opens at the primary display's origin instead of (0, 0), which is not always on the primary display on Linux.
• One decode at the boundary: readWindowState decodes the file with Schema.fromJsonString, so resolveWindowState takes a typed state.
• Opens maximized on Windows: a window saved as maximized is created hidden. Electron's maximize() shows a hidden window with SW_SHOWMAXIMIZED, so it never paints at its normal bounds first. Other platforms keep showing first, because macOS does not reliably enter fullscreen while hidden.
• Embed corners: embeds refresh on move and display-metrics-changed, because their corner masks are rendered for the display scale.

flowchart LR
  File["window-state-*.json"] --> Decode["readWindowState: Schema.fromJsonString"]
  Decode --> Resolve["resolveWindowState"]
  Resolve -->|"maximized, saved display connected"| Anchor["move normal bounds into displayBounds"]
  Resolve -->|"fits a display"| Keep["saved state"]
  Resolve -->|"display gone"| Primary["primary origin, flags kept"]
  Anchor --> Create["BrowserWindow show: false (win32, maximized)"]
  Keep --> Create
  Primary --> Create
  Create --> Max["maximize(): SW_SHOWMAXIMIZED"]

Loading

// packages/desktop/src/main/windows/window-state.ts
const target =
  saved.isMaximized || saved.isFullScreen ? all.find((bounds) => sameBounds(bounds, savedDisplay)) : undefined

if (target && !within(saved, target)) {
  const width = Math.min(saved.width, target.width)
  const height = Math.min(saved.height, target.height)

  return {
    ...saved,
    x: target.x + Math.floor((target.width - width) / 2),
    y: target.y + Math.floor((target.height - height) / 2),
    width,
    height,
  }
}

Not in this PR

• [BUG] Desktop (Windows): app intermittently becomes an invisible input-eating overlay after display/DPI changes on multi-monitor setups #59229: after a display change, the window keeps its correct bounds but stops rendering while it still takes clicks. OpenCode only creates a standard frameless window, so the cause is being reproduced against plain Electron and Chrome, to be fixed upstream.
• Desktop 2.1.1: cursor confined to stale 1280x720 ClipCursor rect on multi-monitor while sessions process #57732: the browser pane and the app session already deny pointer lock, so the reported cursor clip has another source.

---

Added by the opencode bot (AI-generated).

Fixes #58247

Related: #56970, #59229, #57732

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.

[opencodejr]

I couldn't verify this: it needs the Windows desktop app on mixed-DPI multi-monitor hardware. What I ran: the desktop window tests and gui-plugins browser tests pass, and tsc -b passes for packages/desktop. I also ran the saved state from #58247 through resolveWindowState on both checkouts. On the base it loses isMaximized; with this PR it keeps the flag and re-anchors the bounds onto the saved display.

Public API: no SDK, server, config or CLI change. There is one behaviour change for plugin code that uses Electron sessions directly: every session now denies all permissions by default (Electron's default grants them), passing null to setPermission*Handler now means deny-all, and pointer and keyboard lock are always denied.

Concerns, most important first:

• #57732: the session hardening may not change the behaviour this PR closes it for (inline).
• macOS: the deferred reveal applies to all platforms, which risks breaking fullscreen restore on macOS (inline).
• Tests: the wireDisplayRecovery test is large for what it checks (inline).

For #59229 and #57732, could you post before/after evidence from a Windows multi-monitor setup? For example a GetClipCursor log, and a lock/unlock or sleep/wake run with the window maximized on the negative-Y monitor.

Fixes #58247
Related: #45171, #56970

[Hona]

Pushed b15489a, which addresses the three inline comments. The PR body is updated.

• Desktop 2.1.1: cursor confined to stale 1280x720 ClipCursor rect on multi-monitor while sessions process #57732: Dropped from this PR. Closes is now Related. The inline reply has the reasons.
• desktop: maximized state lost when saved bounds exceed the display #58247: Thanks for running the saved state through both checkouts. I added Closes #58247.
• [BUG] Desktop (Windows): app intermittently becomes an invisible input-eating overlay after display/DPI changes on multi-monitor setups #59229 evidence: I don't have a mixed-DPI multi-monitor setup. On one monitor, I forced the maximized HWND to the logical rect with SetWindowPos(..., SWP_NOSENDCHANGING). setBounds(workArea) restored it, and isMaximized() stayed true. A sleep/wake or lock/unlock run on hardware like the reporter's is still needed before merge.

[opencodejr]

The new commit addresses all three earlier points. The session permission hardening is gone and #57732 is no longer closed. The deferred reveal now applies only to a Windows window saved as maximized, so macOS restores fullscreen as before. The wireDisplayRecovery test is now one short test.

I still couldn't verify #59229 itself, because it needs the Windows desktop app on mixed-DPI monitors. The desktop window tests pass and tsc -b passes. Public API is unchanged. I have no new concerns.

[Hona]

Pushed 2e6ed66. This PR now carries only OpenCode's own fixes.

• Removed display.ts: the post-wake bounds reset and forced repaint worked around an upstream rendering problem instead of fixing it. Closes #59229 is now Related; the cause is being reproduced against plain Electron and Chrome, to be fixed upstream.
• Saved state: readWindowState now decodes the file with Schema.fromJsonString, and the tests cover the desktop: maximized state lost when saved bounds exceed the display #58247 display size (1388×781 DIP). The new cases fail on dev.

[opencodejr]

With this PR, the #58247 saved state still comes back maximized on its saved display. On the base it opens windowed at the default size. I checked this by running resolveWindowState with the issue's saved state on both checkouts. The desktop window tests pass. Public API is unchanged.

One blocker: tsc -b in packages/desktop now fails in window-state.ts and passes on the base (inline).

• packages/desktop/src/main/windows/window-state.ts:152: This breaks the type check. bun run typecheck:tsc in packages/desktop now fails with TS18048: 'state.width' is possibly 'undefined' at line 156, and the same for state.height at line 158. It passes on the base. hasBounds now takes Partial<WindowState>, and Number.isInteger doesn't narrow its argument, so state.width > 0 no longer type-checks. A typeof state.width === "number" check (or an isInteger type guard like the one the previous commit had) would fix it.

[opencodejr]

The type-check failure in window-state.ts is fixed: tsc -b in packages/desktop now passes. I re-ran the #58247 saved state through resolveWindowState: with this PR it is still restored maximized on its saved display, and on the base it opens windowed at the default size. The desktop window tests pass. Public API is unchanged. I have no new concerns.