On Sat, Mar 21, 2015 at 7:27 AM, Bruce Momjian wrote: > On Thu, Nov 20, 2014 at 05:22:02PM +0900, Michael Paquier wrote: >> While playing with the GIN code for an upcoming patch, I noticed that >> when inserting a new entry in a new index, this code path is not able >> to make the difference if the index is in a build state or not. >> Basically, when entering in ginEntryInsert@gininsert.c GinBtree built >> via ginPrepareEntryScan does not have its flag isBuild set up >> properly. I think that it should be set as follows to let this code >> path be aware that index is in build state: >> btree.isBuild = (buildStats != NULL); >> >> Note that the entry insertion code does nothing with isBuild yet, so >> it does not really impact back-branches. However, if in the future we >> fix a bug in this area and need to make distinction between a fresh >> index and an old one well there will be problems. For those reasons, >> this correctness fix should be perhaps master-only for now (perhaps >> even 9.4 stuff as well). > > Where did we leave this?
I recall Heikki mentioning me that the code paths where ginPrepareEntryScan is called do not make use of isBuild, so it does not matter much now to not fix it... But *if* there is a new feature implemented in gin that makes use of the flag isBuild there will be problems, so I am of the opinion to push a fix for correctness. -- Michael -- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers