On Oct 13, 2007, at 7:50 PM, Mark Hatle wrote:

> Jeff Johnson wrote:
>> 2) Supporting asm locking on a myriad of embedded tool chains. Of  
>> course
>> fcntl locking with BDB would have solved exactly the same problem.
>
> I tried this a month or so ago and it was incredibly unreliable.  I
> didn't have the time to track down the issues.
>

Really. I'm surprised. I'll take a look at fcntl locking in bdb down  
the road a bit.

FWIW, the headerGetExtension changes uncouple tag store from
header store, so a reasonable SQL schema design (other than
the dirt simple inverted lists with manual join keys) is now feasible.
Loading an entire header to access RPMTAG_NAME is, well, a lot
of effort for not much gain.

My bet is still on bdb however ;-)

> The symptom was that you could open and work w/ the database once..  
> then
> the next time you tried it went into a lock recovery loop until you  
> kill
> -9'ed RPM.  (none of this was over NFS BTW.)  So for fcntl locking in
> db4 to be usable, someone will really need to try and document the
> gotchas.  (Unfortunately I didn't have time.)
>

Thanks for the info.

73 de Jeff