bug#2478: 23.0.91; font antialiasing switch in nextstep broken

[Adrian Robert]

> If someone could try and compile the Emacs.app port
> with -DUSE_LISP_UNION_TYPE, this bug (and probably several others)  
> would
> get caught during compilation.

I just tried this and found errors in core files relating to  
specification of Lisp_Object arguments as "register" variables.  I'm  
not sure the compiler complaint is specific to the version of gcc on  
OS X (4.0.1)?

gcc -c  -Demacs -DHAVE_CONFIG_H  -I. -I/Users/arobert/src/EmacsApp/ 
emacs/src -I/usr/X11R6/include -Dtemacs    -g -O2 -Wno-pointer-sign - 
DUSE_LISP_UNION_TYPE frame.c
frame.c: In function 'delete_frame':
frame.c:1505: error: address of register variable 'force' requested
make: *** [frame.o] Error 1
make: *** Waiting for unfinished jobs....

Then I got an error in nsterm that I can't make sense of:

nsterm.m: In function 'ns_term_init':
nsterm.m:3897: error: incompatible types in assignment

This line is a simple call to a DEFUN taking one arg:

Lisp_Object color_file, color_map, color;
...
color_map = Fx_load_color_file (color_file);