RE: [HACKERS] libpq and SPI

Hello all,

> -----Original Message-----
> From: owner-pgsql-hackers@postgreSQL.org
> [mailto:owner-pgsql-hackers@postgreSQL.org]On Behalf Of Tom Lane
> Sent: Monday, March 15, 1999 3:21 AM
> To: Clark Evans
> Cc: pgsql-hackers@postgreSQL.org
> Subject: Re: [HACKERS] libpq and SPI 
> 
> 
> > What is the problem? I'll research a SPI patch.
> 
> Check the hackers thread (last month I think) titled "libpq and SPI";
> the problem occurs when one submits a utility statement rather than
> a plannable query via SPI.  Apparently what is happening is that the
> backend emits a 'T' message before it invokes the called statement
> and then emits a 'D' message afterwards --- so if the called statement
> causes a 'C' message to come out, libpq gets unhappy.  This seems
> to be clearly a violation of the FE/BE protocol to me, so I don't think
> it's libpq's fault.
> 
> Reasonable fixes might be to postpone the sending of 'T' till after
> the invoked statement is executed, or to modify the traffic cop so
> that a utility statement invoked from SPI doesn't send 'C'.
> 
> I know a little bit about the parts of the backend that communicate with
> the frontend, but nothing about SPI, so I'm not well prepared to solve
> the problem by myself.
>

Probably it's not the problem of SPI.
Specific utility commands(CREATE USER/ALTER USER/DROP USER
/CREATE DATABASE/DROP DATABASE) break FE/BE protocol 
when they are executed inside the PostgreSQL function call.

They call pg_exec_query() (tcop/postgres.c) in their implementation. 
In many cases the destination of pg_exec_query() is "Remote" 
which means that results are sent to frontend process.  But the 
results shouldn't be sent to the frontend directly inside the execution 
of PostgreSQL function.

ProcessUtility() (tcop/utility.c) function processes utility commands 
and takes the destination of command results as its second parameter. 
So should we call pg_exec_query_dest(query_string,  the destination 
parameter of ProcessUtility(), FALSE) instead of pg_exec_query(),
shouldn't we ? 

Thanks.

Hiroshi Inoue
Inoue@tpf.co.jp