BUG #3002: PQexecParams only supports some commands; needs improved error reporting, documenting or fixing

"Phil Endecott" <spam_from_postgresql_bugs@chezphil.org> writes:

I understand that PQexecParams may only be used for SELECT, INSERT, DELETE
and UPDATE commands, but not for other commands where parameter substitution
might be useful such as CREATE VIEW and CREATE TABLE AS SELECT. When I used
PQexecParams with CREATE VIEW with one $ parameter I got no error
immediately, but got a "no value found for parameter 1" error when I later
ran a query that used the view.

As near as I can tell, your issue is that you haven't thought carefully
about what "$1" in a CREATE VIEW *means*. Is it supposed to represent a
constant value bound into the view when it's created? Is it supposed to
represent a value supplied whenever the view is used (if so, how do you
disambiguate when the query calling the view uses $1 for its own
purposes?). Or something else?

Given defensible arguments for a particular answer to these definitional
questions, we might think about implementing something, but right now my
instinct is to not go there. I'm not seeing a use-case that would
justify work on this.

regards, tom lane

Hi Tom,

Tom Lane wrote:

"Phil Endecott" <spam_from_postgresql_bugs@chezphil.org> writes:

I understand that PQexecParams may only be used for SELECT, INSERT, DELETE
and UPDATE commands, but not for other commands where parameter substitution
might be useful such as CREATE VIEW and CREATE TABLE AS SELECT. When I used
PQexecParams with CREATE VIEW with one $ parameter I got no error
immediately, but got a "no value found for parameter 1" error when I later
ran a query that used the view.

As near as I can tell, your issue is that you haven't thought carefully
about what "$1" in a CREATE VIEW *means*. Is it supposed to represent a
constant value bound into the view when it's created?

Yes.

I'm just relying on the documentation of PQexecParams, which says
things like: "the primary advantage of PQexecParams over PQexec is that
parameter values may be separated from the command string, thus
avoiding the need for tedious and error-prone quoting and escaping."
This is the benefit that I was hoping to get. I simply expected
PQexecParams to substitute the parameter values at the positions in the
command string where the $ placeholders were, or to behave as if that
is what it had done.

I'm not seeing a use-case that would justify work on this.

In my IMAP mail server, I run a series of "create temporary view"
commands when the IMAP LOGIN command is processed. These views filter
the message database down to just that user's messages, so that
subsequent queries are simplified:

1 LOGIN phil password
----> create temporary view u_messages as select * from messages where owner='phil'

2 SELECT Today
----> select msg_id from u_messages where age(msgdate)<'1 day'::interval

I have a C++ wrapper around libpq. This allows me to write things like:

typedef std::string username_t;
Query<username_t> create_u_messages
("create temporary view u_messages as select * from messages where owner=$1");
...
create_u_messages("phil");

As currently implemented, this uses PQexecParams. I will have to
change it to do parameter substitution itself and then call PQexec,
either unconditionally or after parsing enough of the command to
recognise whether it is supported by PQexecParams.

Regards,

Phil.

I'm resending this as the first attempt didn't make it to the list.

Phil.

Phil Endecott wrote: