Re: [HACKERS] ERROR: infinite recursion in proc_exit - Mailing list pgsql-hackers

From Tom Lane
Subject Re: [HACKERS] ERROR: infinite recursion in proc_exit
Date
Msg-id 2743.941920829@sss.pgh.pa.us
Whole thread Raw
In response to Re: [HACKERS] ERROR: infinite recursion in proc_exit  (Bruce Momjian <maillist@candle.pha.pa.us>)
Responses Re: [HACKERS] ERROR: infinite recursion in proc_exit  (Bruce Momjian <maillist@candle.pha.pa.us>)
List pgsql-hackers
Bruce Momjian <maillist@candle.pha.pa.us> writes:
>     /* ----------------
>      *  if proc_exit_inprocess > 1, then it means that we
>      *  are being invoked from within an on_exit() handler
>      *  and so we return immediately to avoid recursion.  
>      * ----------------
>      */
>     if (proc_exit_inprogress > 1)  
>         return;

No, no, no, noooo!!!

proc_exit MUST NOT RETURN.  EVER, UNDER ANY CIRCUMSTANCES.

If it does, that means that elog(STOP) can return under some
circumstances.  The callers of elog() are not expecting that,
and they are likely to screw things up even worse if elog returns
control unexpectedly.

AFAICS, this set of problems cannot be fixed by localized patching in
proc_exit.  We have to globally change the way in which errors are
processed after proc_exit has begun execution.
        regards, tom lane


pgsql-hackers by date:

Previous
From: Keith Parks
Date:
Subject: New psql compile problem.
Next
From: Bruce Momjian
Date:
Subject: Re: [HACKERS] New psql compile problem.