feat: remember app window size and position on restart

[fuleinist]

Summary

• Add window state persistence (width, height, x, y, maximized) to the config
• Restore window position and size on app startup
• Save window state on window close

Test Plan

• Resize and move the window, close and reopen the app - window should restore to the same position and size
• Maximize the window, close and reopen - window should start maximized
• Unmaximize from a maximized state - window should restore to previous size
• First launch with no saved state - should use default window size

Closes #534

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

The previous SUGGESTION (Wayland detection test assumed WAYLAND_DISPLAY was unset on the host) is resolved in commit ffc52f6: is_wayland() now delegates to is_wayland_with_env(), which takes an injectable env lookup, and the test exercises both the present and absent cases via closures — deterministic regardless of host session. No new issues found in the changed code.

Files Reviewed (1 file)

• src-tauri/src/config.rs

Previous Review Summaries (2 snapshots, latest commit 2c38c54)

Current summary above is authoritative. Previous snapshots are kept for context only.

Previous review (commit 2c38c54)

Status: 1 Issue Found | Recommendation: Address before merge

Overview

Severity	Count
CRITICAL	0
WARNING	0
SUGGESTION	1

Issue Details (click to expand)

SUGGESTION

File	Line	Issue
src-tauri/src/config.rs	1240	is_wayland() test assumes WAYLAND_DISPLAY is unset instead of controlling it, so it fails on Wayland dev/CI hosts

Files Reviewed (1 file)

• src-tauri/src/config.rs - 1 issue

Fix these issues in Kilo Cloud

Previous review (commit e65192a)

Status: No Issues Found | Recommendation: Merge

Files Reviewed (2 files)

• src-tauri/src/config.rs
• src-tauri/src/lib.rs

---

_{Reviewed by glm-5.2 · Input: 22.9K · Output: 2.1K · Cached: 143.1K}

[debba]

Hey @fuleinist, thanks for the PR! I tested this locally on my machine (Arch, GNOME on Wayland, 2560x1440) and found a couple of issues, one of which is pretty subtle.

The window grows on every restart under Wayland. I planted 900x600 in the config and instrumented the app to report its real geometry. The restore itself actually works (the webview content ends up exactly 900x600), but inner_size(), which is what gets saved on close, reports 952x699 because on Wayland GTK includes the client side decoration margins in that value. So every close saves content + ~52x99 px, and every launch applies that as the new content size. The window inflates by roughly 50x100 px per session, silently, no warning in the logs. Running the same build with GDK_BACKEND=x11 round trips exactly (900x600 in, 900x600 out), so this is Wayland only. It's the same tao/GTK quirk that's been reported upstream against tauri-plugin-window-state.

Closing while maximized poisons the restore size. This one is cross platform. On close the handler saves inner_size() unconditionally, so if you close a maximized window it stores the full screen geometry as the "normal" size. After restart, unmaximizing gives you a window the size of the whole screen instead of the previous size, which is exactly what test plan item 3 says should work. The usual fix is to only update width/height/x/y when the window is not maximized, and always update just the maximized flag.

Also worth noting: on Wayland the position gets saved as 0,0 (GTK can't query it there, and set_position is a no-op since the compositor decides placement). Not something this PR can fix, but if the same config is later loaded in an X11 session the window jumps to the top left corner. A sanity clamp against the current monitor would also help for the unplugged monitor case on X11/Windows.

On the plus side: the new config tests all pass (20/20 in config::tests), save on close fires reliably, the merge logic in save_config correctly keeps frontend saves from wiping the window fields, and first launch with no saved state falls back to the default size as expected.

So, requesting changes for now. I think the minimum here is: persist geometry only when not maximized, and on Wayland skip position persistence and either compensate for the decoration delta or skip size persistence too (checking WAYLAND_DISPLAY is enough to detect it). Happy to re-test once that's in, I have the setup ready.

[fuleinist]

Fixes applied:

1. Maximized window fix: Only persist size/position when not maximized. When closing maximized, only the maximized flag is saved so unmaximizing restores the previous size correctly.

2. Wayland fix: Added helper that checks . On Wayland, position (compositor-controlled) and size (GTK includes CSD margins in causing ~50×100px inflation per session) are skipped. Only the maximized flag is persisted on Wayland.

Built successfully (0 errors, 1 warning - pre-existing unused import). Test added for . Ready for re-test on your Wayland setup.

[kilo-code-bot]

SUGGESTION: Test assumes WAYLAND_DISPLAY is unset rather than controlling it

is_wayland() reads the process environment, but this test never removes WAYLAND_DISPLAY — it only assumes it isn't set. On a Wayland Linux dev machine (or any CI runner where WAYLAND_DISPLAY is present) this assertion fails, despite the test name promising "without wayland_display". Consider removing the var for the test's duration (or refactoring is_wayland to take an injectable env source) so the test is isolated from the host session rather than dependent on where it runs.

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[fuleinist]

Addressed in ffc52f6 — refactored is_wayland() to delegate to is_wayland_with_env(), which takes an injectable env lookup. The test now exercises both the present and absent cases via closures, so it is deterministic regardless of whether WAYLAND_DISPLAY is set on the host running it.

[fuleinist]

@debba All review feedback is now addressed:

1. Wayland size inflation + position — skip size/position persistence on Wayland (commit 2c38c54)
2. Maximized-close poisoning — only persist geometry when not maximized; maximized flag saved separately (commit 2c38c54)
3. start_maximized setting regression — startup maximize now checks both window_maximized and start_maximized (kilo-bot CRITICAL, fixed same day)
4. is_wayland test host-dependence — refactored to an injectable env lookup; test now deterministic regardless of host session (commit ffc52f6)

Happy to have this re-tested with your setup when convenient.