lua-users home
lua-l archive

[Date Prev][Date Next][Thread Prev][Thread Next] [Date Index] [Thread Index] 

On Sep 14, 2009, at 10:44 PM, Mark Hamburg wrote:


On Sep 14, 2009, at 5:26 AM, steve donovan wrote:


On Mon, Sep 14, 2009 at 2:16 PM, Luiz Henrique de Figueiredo
<lhf@tecgraf.puc-rio.br> wrote:
One important reason for not having "set:values" is that it'd be sugar 
for "function (...) return set:values(...) end" and this would imply
the creation of a hidden closure.


Yes, it did feel like an incompatible proposal.
So, is there some support for Mark Hamburg's closure sugar in Rio ;) ?
To be fair, though I'm not sure to whom, I think someone promoted it before I did. ;-) 

But the full details of my proposal included:

1. Binds early.
2. Binds strongly. (Weak closures are a separate implementation problem.) 3. Only generates a closure when not used at a call site though this should be purely an optimization.
The piece of syntax I will take credit for promoting since I don't think I'd seen it elsewhere before is: 

	obj:[method](...)

and in this case:

	obj:[method]
Thereby allowing the method name to be an expression while at the same time not forcing the repetition of obj in an expansion of the colon operator.
I believe this is actually moderately easy to implement as a patch on the current Lua implementation. (I'd have to go find my notes on how to do so.) 

In primaryexp (lparser.c), the colon operator case becomes:

      case ':': {  /* `:' NAME funcargs */
        expdesc key;
        luaX_next(ls);
		/* PATCH: Add support for obj:[ exp ] syntax for calling methods. */
		if (ls->t.token == '[') {
		  yindex(ls, &key);
		}
		else {
		  checkname(ls, &key);
		}
        luaK_self(fs, v, &key);
/* Compiles each expression evaluating them into a sequence of registers. */ switch (ls->t.token) { /* PATCH: Check for a call before going to the funcargs case, otherwise curry. */ 
		  case '(': case TK_STRING: case '{': { /* funcargs */
		    funcargs(ls, v);
			break;
		  }
		  default: {
		    currymethod(ls, v);
			return;
		  }
        }
        break;
      }
I'm afraid I don't have an implementation for currymethod. Without that support, the inner switch just becomes a call to funcargs(ls,v).
But to get the rest of obj:[ method ] working one also needs to revise luaK_self in lcode.c: 

void luaK_self (FuncState *fs, expdesc *e, expdesc *key) {
  int func;
  luaK_exp2anyreg(fs, e);
  freeexp(fs, e);
  func = fs->freereg;
  luaK_reserveregs(fs, 2);
  luaK_codeABC(fs, OP_SELF, func, e->u.s.info, luaK_exp2RK(fs, key));
  freeexp(fs, key);
fs->freereg = func + 2; /* PATCH: Make sure that freeing the key won't have just cost us a register. Why??? */ 
  e->u.s.info = func;
  e->k = VNONRELOC;
}
As the comments, suggest it would be good to have someone who knows the register behavior in the Lua code generator better than I review this code. 

Mark