Gerd Möllmann <gerd.moellmann@gmail.com> writes:

> This is with master fdac10b216f7b47e2eea129d2a96807a0c2055f3 on
> macOS, built with ASAN.
>
>   $ /Users/gerd/emacs/savannah/master/configure --cache-file /var/folders/1d/k_6t25f94sl83szqbf8gpkrh0000gn/T//config.cache.master --without-tree-sitter --with-native-compilation=no CC=clang 'LDFLAGS=-fsanitize=address -fno-omit-frame-pointer' 'CFLAGS=-Wgnu-imaginary-constant -Wunused-result -g -fno-omit-frame-pointer -g -O0 -fsanitize=address -fno-omit-frame-pointer'
>
> Recipe:
>
> - emacs -nw -q
> - M-x xterm-mouse-mode RET
> - M-x make TAB
> - Move the move over the completion candidates
>
> => ASAN abort in note_mouse_highlight, xdisp.c:36108
>
> The line number may vary. Looking at that in the debugger, I see
>
> 	default:
> 	  /* This should not happen.  */
> 	  if (cursor != FRAME_OUTPUT_DATA (f)->nontext_cursor)
> 	    cursor = FRAME_OUTPUT_DATA (f)->nontext_cursor;
>
> nsterm.h defines FRAME_OUTPUT_DATA(f) as
>
>   #define FRAME_OUTPUT_DATA(f) ((f)->output_data.ns)
>
> and since we are not in a GUI frame, this is no good. Analogous defines
> are in xterm.h etc., so the problem is not limited to macOS.
>
> (I built with ASAN because I observe weird things with mouse stuff in
> conjunction with tty child frames, and wanted to check what's up with
> that in master.)

I'd like to propose a change like in the attached patch (from my tty
child frames branch). It fixes the case I mentioned above.