On Mon, Aug 11, 2014 at 8:02 PM, Branko Čibej <br...@wandisco.com> wrote:
> On 11.08.2014 18:37, Stefan Fuhrmann wrote: > > On Sun, Aug 10, 2014 at 7:05 PM, Branko Čibej <br...@wandisco.com> wrote: > >> On 10.08.2014 18:57, Branko Čibej wrote: >> >> >>> FWIW, you should even have the get_memberships function any more. >>> >>> >>> To clarify: I plan to get rid of svn_authz_tng_t::groups at some point. >>> That hash is completely redundant. >>> >> >> Careful! This might get us into scalability troubles. But I reserve >> judgement until you actually came forth with an implementation. >> >> >> The authz_ace_t::members contains exactly the same info, except that you > don't have to look up the group name in svn_authz_tng_t::groups. Since it > both your and my way require one hash lookup to determine if a (group) ACE > pertains to a user, reversing the current groups hash to get memberships is > just a waste of cycles. The groups hash in the authz structure actually > only costs a bit of memory for the hash structure in the result pool, but > otherwise it doesn't give any benefit and I don't believe there's any use > for it. Note that I'm of course not copying hashes around; the group > hashes, user and group names are all singletons. > So, basically, you intend to change the ACE struct from "(user name)->rights" to "(hash of user names)->rights" with the hash actually being either the current per-group instance or an equally singleton per-user instance. That makes perfect sense. I had gotten the impression that you would expand each group right into multiple ACEs, one per user or so. -- Stefan^2.
