Fix sig 11 crash on macOS when using a Japanese (or possibly other CJK type) keyboard

[john-at-charpa]

Greetings,

I recently attempted to bathe in the nostalgia that is gameflorist/dunedynasty and immediately encountered a signal 11 crash. At first I assumed the problem was dunedynasty, but enough messing about with lldb showed me that it was actually liballegro5 that was crashing. Enough debugging with a -pg build showed me:

frame #2: 0x00000001050cb36c
liballegro.5.2.10.dylib`_al_osx_keyboard_handler(pressed=,
event=0x000060000390c300, dpy=0x0000000125832200) at keybd.m:280:69 [opt]

Digging a bit further I found that this is actually a fairly common problem; if you use a CJK type layout (Chinese, Japanese, or Korean) you can trigger it fairly easily. Discussions with suggested code fixes can be seen in places like:

https://github.com/microsoft/node-native-keymap/blob/main/src/keyboard_mac.mm
microsoft/vscode#23833

This PR fixes the issue with relatively the same method. The spice code must flow.

Cheers

[john-at-charpa]

I should also mention that I did do a fair amount of due diligence when it came to the check for layout_data. I was unable to find a single example where a second null check against layout_data was done. Most everyone either returned (which cannot be done here as it is void) or did nothing at all. It looks like (despite a certain company vanishing API docs) TISCopyCurrentKeyboardLayoutInputSource will always return something, so in theory layout_data will never be null.

Interestingly enough the actual cause of the crash is this (when keyboard_input is logged):

2025-07-08 15:34:53.097 dunedynasty[5365:98712406] The first attempt at keyboard input is <TSMInputSource 0x6000027e8ec0> Input Mode: com.apple.inputmethod.Roman (Parent = com.apple.inputmethod.Kotoeri.RomajiTyping)

2025-07-08 15:34:53.097 dunedynasty[5365:98712406] The second attempt at keyboard input is <TSMInputSource 0x6000027a39c0> KB Layout: ABC (id=252)

You can see in the first instance, where TISCopyCurrentKeyboardInputSource is used, the result is actually the Input Method (in my case, Japanese but a Romaji input system).

Either way, I'm fairly confident that this fix is sufficient.

[SiegeLord]

Wow, how gnarly! I'll give this a test, but otherwise this looks good. Thanks!

[SiegeLord]

I tested it, it seems to work good. One thing I noticed is that it doesn't pop up the IEM composition (i.e. multi-key strokes don't work for some reason), but at least it doesn't crash. Thanks again!