On Thu, Mar 15, 2018 at 06:48:36PM +0900, Kyotaro HORIGUCHI wrote: > So, we should reject to define function in the case. We don't > accept the GUC element if it is just a placeholder. > > The attached is a rush work of my idea. Diff for pg_proc.h is too > large so it is separated and gziped. > > It adds a column named "proconfigislist" of array(bool) in > pg_proc. When defined function has set clauses, it generates a > proconfig-is-list-or-not array and set. It ends with error if > required module is not loaded yet. Perhaps > GetConfigOptionFlags(,false) shouldn't return 0 if no option > element is found but I don't amend it for now.
I think your approach has a vulnerability too. I believe that a non GUC_LIST_INPUT extension GUC which was used to create a function may become GUC_LIST_INPUT variable. If I'm not mistaken nothing stops from that. In this case values in proconfigislist won't be valide anymore. -- Arthur Zakirov Postgres Professional: http://www.postgrespro.com Russian Postgres Company
