I knew I had seen that somewhere recently. At http://newmicros.com/download/software/NMI/, the file NMIY0020.ZIP contains the support files for their NMIY-0020 board, including a 68HC11 Small C cross-compiler. For MS-DOS, I believe.

Edited to add: whoops, I missed that 'legacy' is looking for source code for the compiler. That zip only contains an executable.

_________________
Because there are never enough Forth implementations: http://www.camelforth.com

I can think of two reasons:

1. Exceptionally difficult to port to new architectures, and is too complicated.
2. The compiler proper relies on its own extensions (though an already-existing ANSI C compiler can compile an intermediate version which can compile the remaining portions).

I've had a bit of (dated) gcc-11 experience-- of course it's 32bit nature will look bad if you compare it to 16bit compilers. One of the biggest mistakes by newbs is inappropriately using 32-bit signed ints everywhere. The first experiences in embedded programming is getting your casts right. (The second is not linking in whole of libc..)

When I did the unsigned char magic on my 68hc11 robot code, the resulting object output actually looked rather nice. (And the optimizer in gcc was great for feedback to improve the code..)