Subj : Re: Serializing Object Graphs
To   : Eric Brueggemann <eric_brueggemann@yahoo.com>
From : Brendan Eich <brendan@meer.net>
Date : Wed Apr 09 2003 11:29 pm

> I've identified the following slot-setting calls:*
>   jslock.c: js_SetSlotThreadSafe
>   jsobj.c: js_Clear
>   jsobj.c: js_SetRequiredSlot
>   jsobj.h: LOCKED_OBJ_SET_SLOT
> * not using TOO_MUCH_GC or LiveConnect
>
> At the point that obj->slots[slot] is being filled, are the following 
> always true?
> 1. OBJ_SCOPE(obj) == obj 



You mean OBJ_SCOPE(obj)->object == obj, I think, and only if 
OBJ_IS_NATIVE(obj).


> 2. slot <-> (atom | tinyid) can be found in OBJ_SCOPE(obj) 


Yes, currently.  A patch in one of my trees for bug 
http://bugzilla.mozilla.org/show_bug.cgi?id=123668 may relax the 
(corrected) first invariant.


> If so, the modify case should be easy to log and roll back. Still 
> looking into the add and delete cases. Any pointers?


For native objects, js_DefineProperty, js_SetProperty, 
js_DeleteProperty, and js_Clear in jsobj.c, or the lower-level scope 
operations js_AddScopeProperty, js_ChangeScopePropertyAttrs, 
js_RemoveScopeProperty, and js_ClearScope in jsscope.c.

If you care about non-native objects, you may want to think about 
generalizing "above" the object-ops layer.  Stackable ops, perhaps.  But 
I really don't want any overhead in the engine there.

> btw. this is whole scope tree implementation is neat stuff! 


Thanks!

/be

.