fix(FxSystemTrayView): use LoadImage for tray icons to ensure proper sizing and alpha

[akai07]

Summary

Fixes system tray icon rendering in Windows dark theme mode by replacing LoadIcon with LoadImage using explicit small-icon dimensions.

Problem

LoadIcon loads icons at the default system icon size (typically 32x32), which then gets scaled down to the system tray small-icon size (16x16 on most displays). This scaling can cause:

• Loss of alpha channel information in dark theme mode
• Visible artifacts around icon edges
• Icons appearing as white blocks or with incorrect colors on dark taskbars

Fix

Replaced all LoadIcon(hInst, L"IDI_LOGO_*") calls with a new LoadTrayIcon helper that uses LoadImage with explicit dimensions from GetSystemMetrics(SM_CXSMICON) and GetSystemMetrics(SM_CYSMICON). This ensures:

• Icons are loaded at the exact size needed for the notification area
• No lossy scaling that degrades alpha transparency
• Correct rendering on both light and dark Windows themes

Files Changed

• fxsound/Source/GUI/FxSystemTrayView.cpp — LoadIcon → LoadImage in both setStatus() and addIcon()

Related Issues

• Fixes dark theme tray icon rendering issues on Windows 10/11

[bvijay74]

@claude review this pull request

[claude]

Claude finished @bvijay74's task in 4m 27s —— View job

---

Reviewing PR #534

• Gather context (diff, related files, CLAUDE.md)
• Review FxSystemTrayView.cpp changes for the dark-theme tray icon fix
• Review the other changed files (sndDevices*, FxController.cpp) for scope/consistency
• Post review feedback

Summary

The core idea (LoadIcon → LoadImage with explicit SM_CXSMICON/SM_CYSMICON) is the right, well-documented fix for blurry/washed-out tray icons on Windows, and the refactor into a LoadTrayIcon helper is clean. However, there's one real regression introduced by the change, plus a scope observation.

🔴 Icon handle leak introduced by switching to LoadImage (High)

fxsound/Source/GUI/FxSystemTrayView.cpp:69-74

static HICON LoadTrayIcon(HINSTANCE hInst, LPCWSTR iconName)
{
    return (HICON)LoadImage(hInst, iconName, IMAGE_ICON,
                            GetSystemMetrics(SM_CXSMICON), GetSystemMetrics(SM_CYSMICON),
                            LR_DEFAULTCOLOR);
}

LoadIcon returns a shared, system-cached handle that doesn't need to be freed. LoadImage (without LR_SHARED) returns a private handle that the caller owns and must release with DestroyIcon once Shell_NotifyIcon has consumed it. Neither setStatus() (line 76) nor addIcon() (line 176) ever calls DestroyIcon on nid.hIcon.

setStatus() is invoked frequently during normal use — e.g. FxController.cpp:1358 and :1371 call it on every audio processing on/off toggle — so this leaks one GDI/USER icon handle per call. Over a long-running tray session with repeated play/pause, this can approach the per-process USER object quota (10,000 by default), eventually causing icon updates (or other UI operations) to silently fail.

Fix: either cache the previously-created HICON (e.g. as a member of FxSystemTrayView) and DestroyIcon() it before/after assigning the new one, or load the 4 icon variants once (e.g. in the constructor) and reuse the handles instead of reloading on every setStatus()/addIcon() call.

Fix this →

🟡 PR bundles unrelated changes (Scope)

The PR title/description only describe the tray icon LoadImage fix, but the branch also contains two unrelated commits that make up ~105 of the 120 added lines:

• a09f1fe — FxController.cpp: prevents sleep during playback (SetThreadExecutionState) and re-inits audio on system resume (WM_POWERBROADCAST/PBT_APMRESUMEAUTOMATIC)
• 9488820 — sndDevices*: filters capture-device events out of OnDeviceStateChanged and adds Bluetooth-style device reconnection handling

These look like reasonable fixes on their own, but bundling them with the tray icon fix makes the PR harder to review/revert independently, and the description doesn't mention them at all. Consider either updating the PR description to cover all three fixes, or splitting them into separate PRs.

🟢 Minor notes (non-blocking)

• FxController.cpp:1358 — SetThreadExecutionState(ES_CONTINUOUS | ES_SYSTEM_REQUIRED | ES_DISPLAY_REQUIRED) also forces the display to stay on while audio is processing. For a background audio app this may be more aggressive than needed (most audio players only set ES_SYSTEM_REQUIRED so the screen can still turn off during playback) — worth confirming this is intentional.
• audiopassthru/src/sndDevices/sndDevicesDeviceCallbacks.cpp:203-224 — OnDeviceStateChanged now creates a new IMMDeviceEnumerator via CoCreateInstance on every callback just to check GetDataFlow. Not a correctness issue, but a bit wasteful given this callback can fire often; caching the enumerator on cast_handle would avoid the repeated COM allocation.

I verified the hasReconnectedDevice/reconnectedDeviceGuid fields are zero-initialized via sndDevicesReInit before first use (called from sndDevicesInit), and that the new reconnection-selection block in sndDevicesImplementDeviceRules.cpp:234-268 correctly sits after the "user manually selected an active device" check (line 188-193), so it won't override an explicit user selection — no bug there.
· Branch: fix/dark-theme-tray-icon → develop/windows

[bvijay74]

@akai07 I got the PR reviewed.
Can you please address the review comments.