Isn't init_irels() dangerous ? - Mailing list pgsql-hackers

From Hiroshi Inoue
Subject Isn't init_irels() dangerous ?
Date
Msg-id EKEJJICOHDIEMGPNIFIJMEEMDCAA.Inoue@tpf.co.jp
Whole thread Raw
Responses Re: Isn't init_irels() dangerous ?  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-hackers
Hi all,

In InitPostgres()(postinit.c) I see the following code.
       RelationCacheInitialize();    /* pre-allocated reldescs created here
*/       InitializeTransactionSystem(); /* pg_log,etc init/crash recovery
here */

init_irels() is at the end of RelationCacheInitialize() and
accesses system tables to build some system index
relations. However InitializeTransactionSystem() isn't
called at this point and so TransactionIdDidCommit()
always returns true. Time qualification doesn't work
properly under such a situation.
It seems that init_irels() should be called after
InitializeTransactionSystem() was called.

Comments ?

Regards.
Hiroshi Inoue



pgsql-hackers by date:

Previous
From: Sébastien Bonnet
Date:
Subject: libpq enhancement for multi-process application
Next
From: "Hiroshi Inoue"
Date:
Subject: Is PQreset() proper ?