Re: [Bodington-developers] special groups

[Jon M. <jo...@te...>]

The general idea of 'special groups' is that at install time you can 
decide to create up to (64?) groups which you know will very frequently 
occur in ACLs. As a user navigates the web site their membership of 
groups will be checked frequently. Since the user has logged in and is 
actively accessing the site you can be certain that the User record is 
loaded - because there is a reference to it in the httpsession object 
etc. So, since special group membership is coded in the User object, 
checking membership of the group requires no database access at all.

The trade off is that although taking a user and finding out if they 
belong is quick, as Matthew points out, taking a special group and 
listing the users that belong is more difficult. A potential improvement 
would be to put a record in 'members' for all groups, including special 
ones, making sure that the two sources of data always agree.

Jon


Matthew Buckett wrote:

>Paul Trafford wrote:
>  
>
>>At 15:09 08/03/2006, Antony Corfield wrote:
>>
>>    
>>
>>>Can someone, Jon perhaps, remind me what special_groups_a etc. are
>>>used for in users table?
>>>      
>>>
>>You can deduce memberships of the system generated groups.  The total
>>value is made up of components totted up from:
>>
>>2    allusers
>>4    allstaff
>>8    allstudents
>>16   allothers
>>32    campus.users
>>64    campus.staff
>>128    campus.students
>>256   campus.other
>>
>>e.g.
>>special_group a=102 => campus.staff, allstaff, allusers, campus.users
>>special_group a=170 => campus.students, allstudents, allusers, campus.users
>>    
>>
>
>I think User.addToGroup() and User.removeFromGroup do the dirty work.
>They are used so that you can have very large groups without having a
>large number of entries in the members table. One problem with them on
>PostgreSQL is that you can't have an index on them that improves
>performance when doing bitwise operations which means that the DB always
>has todo a full table scan doing the bitwise operation for each row.
>
>  
>