Hi Gerrit (and anybody else who wants to comment),
it fits well that you're out of town next week, that means you're not working
on
OpenSG much. :) Can you make sure that you have all your changes committed,
especially in the FC?
As it happens, my wife is out of town next week, too. Which gives me a good
chance to get some work done on OpenSG, and I would like to get the dreaded
FC/CPtr thing done. At the same time, I would really like to simplify the FC
code significantly, as, frankly, the current version is just scary.
So the goals are:
- simplify the code. Remove the mixins and use simple derivation as far as
possible. And don't worry about a little bit of code duplication, if it keeps
things more understandable. Code is not just for compilers, it's for people,
too. ;)
- switch to cptrs only. I don't think the code will change much from the
current
CPTR mode: have an AspectStore refed from the class and use that to keep the
RefCount
- add weak ptrs and make ref counting more efficient. To do that I want to use
the boost::sp_counted_base, which supports weak pointers and also has
specializations for lock-free thread safe counting. It's not quite the way I
would like it to be as you pretty much have to use a pointer to it instead of
deriving the aspect store from it, but maybe I can figure that one out, too.
- remove the PointerBuilders, as they are not necessary any more if we don't
have different modes. This one will go a long way to make the code look more
normal.
- change the fields to use RefPtrs, remove explicit ref handling in the code.
This does not mean using RefPtrs everywhere, just for storage. This should also
go some way towards making the reflexive interface more complete
- add AutoPtrs to avoid ref/unref for just passing out values. This should
avoid
overhead when accessing field data.
- Use fcd2code as much as possible, to avoid having to manually mess with
Fields.
Structure-wise I think I'll derive RC and FC from a base that has common stuff
(not totally sure what it's going to be, but probably not too much). The reason
they're not derived from each other is that RC and FC probably need to handle
RefCounting differently.
Questions that I have:
- Should we get rid of the ...Ptr types? They're not strictly necessary any
more, but without them Allen's memory debugging won't be possible, which i
think
is very helpful. In non-memdebug mode they would be direct CPtrs.
- are the aspect-specific refcounts life? Are they needed, did they fix
something?
- what did I miss why this is a bad idea?
- what did I miss to make this an even better idea?
- any general comments?
Thanks
Dirk
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Opensg-core mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/opensg-core
