Re: Introducing "toggle references"

[Tim Janik <timj gtk org>]

On Mon, 25 Apr 2005, Owen Taylor wrote:

> Here's a patch implementing the proposed API (with the GDataset
> flags hack). A test case is included.
>
> As always, if the patch is approved, I'll move the docs to the tmpl
> files to match the rest of libgobject.

without looking into the issue of toggle references per se, skimming
over the patch i do have two immediate comments on the datalist part:

1) you do:
     GData *tmp = G_DATALIST_GET_POINTER (data);
     modify (&tmp); // invokes user callbacks
    G_DATALIST_SET_POINTER (data, tmp);
  this breaks reentrancy for datalists, getters/setters/clearance
  all have to work correctly in recursive call scenarios.
  so you actually need to move the GET/SET-POINTER logic down into
  the modify() functions of your patch.
2)
	+++ glib/gdataset.h     25 Apr 2005 22:57:13 -0000
	@@ -39,6 +39,15 @@ typedef void            (*GDataForeachFu

	 /* Keyed Data List
	  */
	+#define G_DATALIST_FLAGS_MASK 0x3
	+
	+#define G_DATALIST_GET_FLAGS(datalist)         \
	+  ((gulong)*(datalist) & G_DATALIST_FLAGS_MASK)
	+#define G_DATALIST_SET_FLAGS(datalist, flags) G_STMT_START {                           \
	+  *datalist = (GData *)((flags) |                                                      \
	+                       ((gulong)*(datalist) & ~(gulong)G_DATALIST_FLAGS_MASK));        \
	+} G_STMT_END
	+
   not being able to predict future datalist modifications, flag uses,
   notification requirements or alignment schemes, the GET_FLAGS and
   SET_FLAGS macros should actually be turned into real functions on
   the datalist.
   i see no problem with leaving G_DATALIST_FLAGS_MASK as a macro, since
   ensuring future modifications to it will only grow the number should
   be easy to maintain and keeps backwards compatibility.

> Regards,
> 						Owen

---
ciaoTJ