Arc Forumnew | comments | leaders | submitlogin
2 points by almkglor 4126 days ago | link | parent

Okay, now that I've actually gotten to see it - I think you forgot to push codegen.arc, since the current version on git don't actually generate code that calls DBL2OBJ.

Edit: Also, it might be useful to organize the primitives as "functions", like what I did with cons_fun().

So:

  #define CONS() {obj d = POP(); TOS() = cons_fun(TOS(), d);}
This should allow us to express some primitive calls directly:

  (%cons foo (%car bar))
  ==>
  PUSH(cons_fun(foo, car_fun(bar)));
Also, it might be useful to merge 'aquote and 'alit structures into a single structure.


1 point by sacado 4125 days ago | link

Oops, thank you :/ I just pushed codegen.

I'll work on what you said about primitive functions in a few days (Monday or Tuesday I guess).

-----

2 points by almkglor 4120 days ago | link

Err, no. I just realized while on vacation that it's not safe to use memory-allocating primitives as functions. The problem is with nested conses:

  QUOTE_CONSTANT[0] = cons_fun(FIX2OBJ(1),cons_fun(FIX2OBJ(2),NILOBJ))
Let's call the cons_fun(FIX2OBJ(1) ...) call cons_fun1, and the other as cons_fun2. cons_fun2 is called first (because it's more inner). The value is then pushed into the C stack. Then cons_fun1 is called. However, what if GC is triggered at exactly cons_fun1? Then the cons cell returned by cons_fun2 is on the C stack, not the Arc stack; the C stack is not scanned by the GC! This means we actually lose memory areas.

Probably means I have to modify the way constants are stored.

-----

1 point by sacado 4120 days ago | link

Hmmm, good point. The stack, GLOBALS and QUOTE_CONSTANTS are all explored when GC is triggered. Maybe the solution would be to add a temporary storage (that should be taken into consideration by GC too) before putting things in the constants, or something like that ?

-----

1 point by almkglor 4120 days ago | link

> Maybe the solution would be to add a temporary storage (that should be taken into consideration by GC too) before putting things in the constants, or something like that ?

Called the "Arc stack", maybe? ^^

Currently this is what is done:

  QUOTE_CONSTANTS[0] = cons_fun(FIX2OBJ(1),cons_fun(FIX2OBJ(2),NILOBJ))
It should probably be:

  PUSH(FIX2OBJ(1));
  PUSH(FIX2OBJ(2));
  PUSH(NILOBJ);
  CONS();
  CONS();
  QUOTE_CONSTANTS[0] = POP();
Of course, the thing about having _fun functions is, they are safe for non-allocating functions. So we could safely directly translate:

  (car (cdr foo))
to:

  PUSH(car_fun(cdr_fun(LOCAL(3)/*foo@42*/)));
Edit: of course, even though it's named "_fun" doesn't mean that it is a function:

  #define car_fun(o) (((pair *)(o))->car)

-----

2 points by almkglor 4120 days ago | link

Okay, converted init_constants() to use the Arc stack instead of the C stack. Fear the pointer arithmetic foo-ness when using a processor with special PUSH and POP instructions (which are arguably dying out, because RISC processors handle stacks using explicit pointer ariths).

-----