svn commit: samba r4785 - in branches/SAMBA_4_0/source: . script

[Simo Sorce]

On Tue, 2005-01-18 at 12:19 +1100, Andrew Tridgell wrote:
> Simo,
> 
>  > I think I can do a lot of talloc_steal() or talloc_reference() now that
>  > we moved ldb to talloc, that code has not changed when I made the
>  > interim conversion to talloc.
> 
> a steal probably makes most sense, or just a pointer assignment if
> you're keeping both contexts around for the same time.

I think I'll o for a steal.

>  > The strdup where necessary as I search all the attributes for all the
>  > schema referenced by the current operation and need to keep them around
>  > while searching the tree 
> 
> I don't really understand why this requires a strdup. You can keep the
> search results for as long as you need them, or steal bits of them if
> you want to just keep some parts.

yeah but I'm not sure I want to keep 10-12 search results around, it's
easier to steal/reference the pointer.

> I'm not really so concerned about the speed of the strdup calls
> (although that is a factor), its the fact that it forces you to add a
> lot more error checking and bailout logic.

yeah not pretty, but necessary at this stage

>  > (1-2 search for each objectclass involved, no optimizations yet)
> 
> yep - we will need some optimisations once you've got it to the stage
> of passing all our existing tests.

yeah, once we are confident the implementation is ok then we will be
able to look at which code points are critical and try to speed them up.

>  How is that going btw?

Well, slowly as I have not much time, but it's progressing, I still miss
attribute syntax checking and need to do a lot more tests before I'm
satisfied.

> I imagine
> you've probably hit quite a few places in our rpc server code where we
> don't fill in all the required attributes :-)

Yeah, but that's only part of the problem. In fact in AD it is legal to
not fill up all attributes, in some cases you can just pass the
objectclass and few other attributes and all others get auto-generated.
So, after the first formal implementation I want to make some deep tests
on this issue to find out how to program a new module to be called
before schema checking that will fill up required attributes in a
similar way as AD does (that's why I was looking for a way to order
modules btw, order matters in this case).

Simo.

-- 
Simo Sorce    -  idra at samba.org
Samba Team    -  http://www.samba.org
Italian Site  -  http://samba.xsec.it