On 10/31/2010 04:40 PM, Alex Hunsaker wrote:
> On Sun, Oct 31, 2010 at 12:00, Andrew Dunstan<andrew(at)dunslane(dot)net> wrote:
>> On 10/31/2010 11:44 AM, Tom Lane wrote:
>>> Good catch, patch reverted (and regression test added).
>> Well, I guess that answers the question of why we needed it, which nobody
>> could answer before. I'm not sure I exactly understand what's going on here,
>> though - I guess I need to look at it closer. At least I think we need a
>> code comment on why the trigger flag is needed as part of the hash key.
> The stack trace is:
> #0 0x0000000000000000 in ?? ()
> #1 0x00000000006c18e9 in InputFunctionCall (flinfo=0x2a039a0,
> str=0x0, typioparam=0, typmod=-1)
> #2 0x00007ff6d2bdf950 in plperl_func_handler (fcinfo=0x7fff4743bec0)
> at plperl.c:1729
> which happens because prodesc->result_in_func.fn_addr (flinfo) is
> NULL. That happens because when we are a trigger we don't setup
> input/output conversion. And with the change we get the same
> proc_desc for triggers and non triggers, so if the trigger function
> gets called first, any call to the direct function will use the same
> proc_desc with the wrong input/out conversion.
How does that happen given that the function Oid is part of the hash key?
> There is a check so that trigger functions can not be called as plain
> functions, but it only gets called when we do not have an entry in
> plperl_proc_hash. I think just moving that up, something the like the
> attached should be enough.
This looks like the right fix, though.
In response to
pgsql-hackers by date
|Next:||From: Alex Hunsaker||Date: 2010-10-31 21:24:20|
|Subject: Re: why does plperl cache functions using just a bool for is_trigger|
|Previous:||From: Dimitri Fontaine||Date: 2010-10-31 21:12:12|
|Subject: Re: Simplifying replication|