From: Mauricio Fernandez <mfp@...>
Date: 2006-08-13T09:38:17+09:00
Subject: Re: PATCH to make internal Hash class retain order...

On Sun, Aug 13, 2006 at 09:08:14AM +0900, Hal Fulton wrote:
> dblack@wobblini.net wrote:
> >Could it be implemented as an extension that redefines Hash?
> >
> 
> That's a very interesting question. I'd like to think so.
> 
> I'd assume that the *parser* at least retains the oreder
> in which it sees the elements of the literal (long enough
> to pass them to the "guts" of Hash).

I don't think this is easily implementable as an extension without patching
the interpreter.

The order of the associations in the literal is indeed preserved, since
they're put in a linked list inside the NODE_HASH. However, the latter is
evaluated in rb_eval, so the OP would need to patch that, using at least
something like

(untested)

--- eval.c.orig 2006-08-13 02:28:59.000000000 +0200
+++ eval.c      2006-08-13 02:31:39.000000000 +0200
@@ -3750,22 +3750,23 @@
        break;
 
       case NODE_HASH:
        {
            NODE *list;
-           VALUE hash = rb_hash_new();
+           VALUE hash = rb_funcall(rb_const_get_at(rb_cObject, rb_intern("Hash")), 
+                                   rb_intern("new"), 0, 0);
            VALUE key, val;
 
            list = node->nd_head;
            while (list) {
                key = rb_eval(self, list->nd_head);
                list = list->nd_next;
                if (list == 0)
                    rb_bug("odd number list for Hash");
                val = rb_eval(self, list->nd_head);
                list = list->nd_next;
-               rb_hash_aset(hash, key, val);
+               rb_funcall(hash, rb_intern("[]="), 2, key, val);
            }
            result = hash;
        }
        break;
 

in order to turn literals into ordered hashes if Hash is replaced.

Even if you cache the IDs, the speed impact will probably be significant;
it's too late to time this now.

-- 
Mauricio Fernandez  -   http://eigenclass.org   -  singular Ruby