Hi,

Big fan of micropython, so for I decided to port it to an 10 year old gadget, just for the fun of it, here are my first-hand impression of how that went.
The platform is a ColdFire CPU, the last iteration of the Motorola 68K family, but it is still a "new" archtecture as far as micropython :-)
Setting up GCC as cross-compiler is old hat, no comments there.
I started from the ports/minimal files.
The first thing I noticed is that there is very little inline-commentary to explain what goes on, and which bits needs fleshing out.
I was lucky, in the sense that the platform has both a debugger and a watchdog, otherwise I would probably not have guessed that I ended up in nlr_jump_fail() which by default is just a while(1){;}
The reason I ended up in nlr_jump_fail() was that I needed -malign=int in CFLAGS to get 32 bit alignment. That is of course something peculiar to the 68K which by default is only 16bit aligned, but it might be a good idea to add a compile-time check to break early if things do not align on 32 bit (using compile-time asserts) because the "diagnostics" from this mistake are very confusing.
Once I got ports/minimal running, the obvious thing was to add to it.
A "What to do once ports/minimal runs" section would have been nice :-)
The process was frustrated because ports/minimal/Makefile has OBJ += $(PY_CORE_O) instead of OBJ += $(PY_O) so nothing from extmod etc. actually gets compiled, until you spot that detail. (Unless there is a good reason, I suggest changing it.)
If there is documentation about MICROPY_CONFIG_ROM_LEVEL and the many other "inclusion" #defines, I could not find it.
What depends on what was sorted out by Prototyping (The development methodology formerly known as Trial&Error.)
I'm still not done and it would be nice to have a table with the various options, their dependencies and a rough estimate of how much space they take up.
Having to configure long-int two places (Makefile & mpconfigport.h) is pointless. Set it as a variable i the Makefile and -D it into mpconfigport.h ?
Trying to use the longlong instead of mpz caused link errors relating to read_r, write_r etc. Never figured that one out and stuck with mpz.

Next stumbling block was the GC, which I found particularly under-documented.
It took some hours before I arrived at this gchelper_68k.s:

    .section .text
    .align 4    
    .global gc_helper_get_regs_and_sp
    .type gc_helper_get_regs_and_sp, %function  
gc_helper_get_regs_and_sp:
    movl        %a6, %sp@-
    movl        %sp@(8),%a6
    movl        %a0,%a6@+
    movl        %a1,%a6@+
    movl        %a2,%a6@+
    movl        %a3,%a6@+
    movl        %a4,%a6@+
    movl        %a5,%a6@+
    movl        %sp, %d0
    movl        %sp@+,%a6
    rts
    .size gc_helper_get_regs_and_sp, .-gc_helper_get_regs_and_sp

(Feel free to include it if you want to support retro-computing.)

With that in place, I had a stable port, and have started having fun with it.

Hat-tip to everybody involved!