Re: Fwd: Re: DB_LOG_AUTOREMOVE: how does it work?

[Howard Chu <hyc@symas.com>]

Hallvard B Furuseth wrote:

> Howard Chu writes:
>  
>
> > Quanah Gibson-Mount wrote:
> >    
> >
> > > <Kurt@OpenLDAP.org> wrote:
> > >
> > >      
> > >
> > > > Maybe we can commit the back-bdb patch behind an ifdef,
> > > > and place the BerkeleyBDB patch in build/berkeleydb42.patch?
> > > >        
> > > Hm... does having the 3rd patch in place generate a different version
> > > string for BDB?  If someone builds OpenLDAP with the OL piece of the
> > > patch, but not the BDB piece of the patch, it simply won't work.  We'd
> > > definitely need a way to test for BDB being a patched version.
> > >      
> > Currently it doesn't, but I guess we can patch the version macros as
> > well. Yes, we definitely need a way to detect the patch, and probably a
> > runtime check to abort the startup if the wrong library somehow gets
> > attached...
> >    
>
> If the bdb patch defines a variable and openldap accesses it, patched
> openldap + unpatched bdb should fail at link time.
>  
Ah yes, the same trick we used to distinguish Cyrus 1.5 from 2.x, good idea.

> I don't remember if the bdb patch is harmless to a caller which does not
> know about the patch - if so, openldap could set the bdb variable and
> bdb could invoke the new behaviour only if the variable is true.
Yes, the BDB patch is harmless, it has to be explicitly invoked by the 
caller. Maybe I can work out a better patch for BDB that works 
transparently, so back-bdb doesn't need to be patched. (That would be 
ideal, since the back-bdb patch is not needed and won't work with BDB 4.3.)

--
 -- Howard Chu
 Chief Architect, Symas Corp.       Director, Highland Sun
 http://www.symas.com               http://highlandsun.com/hyc
 Symas: Premier OpenSource Development and Support