>
> Looks right to me.
>

Many thanks for your review!

>> @@ -940,10 +940,9 @@ #define DEFINE(name, value) [name] = &&insn_ ## name,
>>  	  }
>>
>>  	CASE (Bsave_restriction):
>> -	  record_unwind_protect (save_restriction_restore,
>> -				 save_restriction_save ());
>> -	  record_unwind_protect (narrowing_locks_restore,
>> -				 narrowing_locks_save ());
>> +	  record_unwind_protect (save_restriction_and_narrowing_locks_restore,
>> +				 Fcons (save_restriction_save (),
>> +					narrowing_locks_save ()));
>>  	  NEXT;
>
> Shouldn't the value returned by `save_restriction_save` include the 
> narrowing locks already?
>
> IOW rather than changing this `bytecode.c` code, the locks handling 
> should be "hidden" inside `save_restriction_restore` and 
> `save_restriction_save`, don't you think?
>

You mean, the attached patch?  That's probably even better, indeed. 
(Again it passes make and make check, with and without native 
compilation.)