On macOS, SDL_SetWindowKeyboardGrab behaves inconsistently with Windows (largely ignored on macOS; cannot intercept system shortcuts)

[pipiwoaini]

On Windows, calling SDL_SetWindowKeyboardGrab(window, SDL_TRUE) allows an application to intercept/suppress certain system-level key combinations (implemented via a low-level keyboard hook). On macOS, the same code does not prevent system shortcuts from triggering (e.g., ⌘Tab app switching, ⌘Space Spotlight, Ctrl+Arrow for Spaces/Mission Control). This results in inconsistent behavior across platforms and significantly impacts use cases such as remote control, cloud/game streaming, emulators, and kiosk mode.

[icculus]

I'm not aware of a way to prevent this on macOS. I don't think you can turn off Command-Tab in any circumstance. Maybe I'm wrong; if you can think of any app that does, I can take a look at what method it uses.

[pipiwoaini]

test.m.zip

I tested this code on macOS 26 and it works, provided you grant Accessibility and Input Monitoring permissions. If you have any questions, please contact me.

[pipiwoaini]

Proposal: add a hint that, when used with SDL_SetWindowKeyboardGrab, enables “intercept as many key events as possible” to prevent the OS from handling system shortcuts—especially important for games.

[maia-s]

UTM can grab system keys like command-tab and command-space. As far as I can see in my settings it doesn't have accessibility or input monitoring permissions and it still works. It doesn't grab touchpad gestures or custom global hotkeys though.

[icculus]

I'll look at what UTM is doing, but I have a suspicion this is a function of Apple Virtualization and probably doesn't work when it falls back to QEMU. Just a wild guess though.

Accessibility is an option, but we'd have to add a hint to request SDL try otherwise-forbidden methods, and apps would have to have code to convince the user to approve permissions. This isn't a dealbreaker, because for certain apps like Remote Desktop, it could be totally reasonable.

This is all likely to slip from 3.4.0 though.

[pipiwoaini]

A “fully loaded” UTM-level implementation isn’t necessary. My test code already covers 90% of scenarios; the effort to bridge the remaining 90% to 100% should rightly be borne by developers of software like UTM.

[icculus]

Yeah, @pipiwoaini's solution is dirt-simple, but will definitely have to default to disabled, since it'll need the app to have UI about getting Accessibility permissions, etc. We'll let apps set a hint to request SDL use an event tap to filter these key combos out.

Current plan: I'm going to try to get this in for 3.4.0.

[icculus]

New current plan: I'm going to bump this to 3.6.0. Getting down to the wire, don't want to make big changes and additions at this point.

[pipiwoaini]

Waiting for 3.6.0 is a bit too far off. I’ll try to update the code and run some tests when I get a chance, and if everything looks good I’ll submit the changes to move this merge forward

[slouken]

We're going to bump this for now, but if you have a PR in the next week or so we'll consider it for 3.4.