Getting error 42P02, despite query parameter being sent

Στις 16/11/24 12:55, ο/η Max Ulidtko έγραψε:

Greetings, group!

I'm trying to understand a low-level issue. Am evaluating a new client
library for Postgres; it's not particularly popular / mainstream, and
as I've understood so far, sports an independent implementation of PG
binary protocol.

The issue I'm hitting with it is exemplified by server logs like this:

2024-11-16 10:28:19.927 UTC [46] LOG: statement: SET client_encoding =
'UTF8';SET client_min_messages TO WARNING;
2024-11-16 10:28:19.928 UTC [46] LOG: execute <unnamed>: CREATE VIEW
public.foobar (alg, hash) AS VALUES ('md5', $1);

At least for SQL level prepared statements the statement has to be one of :

|SELECT|, |INSERT|, |UPDATE|, |DELETE|, |MERGE|, or |VALUES|

|so CREATE is not valid, and I guess the extended protocol prepared
statements aint no different in this regard.
|

On 11/16/24 03:15, Achilleas Mantzios wrote:

Στις 16/11/24 12:55, ο/η Max Ulidtko έγραψε:

Greetings, group!

I'm trying to understand a low-level issue. Am evaluating a new client
library for Postgres; it's not particularly popular / mainstream, and
as I've understood so far, sports an independent implementation of PG
binary protocol.

The issue I'm hitting with it is exemplified by server logs like this:

2024-11-16 10:28:19.927 UTC [46] LOG: statement: SET client_encoding =
'UTF8';SET client_min_messages TO WARNING;
2024-11-16 10:28:19.928 UTC [46] LOG: execute <unnamed>: CREATE VIEW
public.foobar (alg, hash) AS VALUES ('md5', $1);

At least for SQL level prepared statements the statement has to be one of :

|SELECT|, |INSERT|, |UPDATE|, |DELETE|, |MERGE|, or |VALUES|

|so CREATE is not valid, and I guess the extended protocol prepared
statements aint no different in this regard.

It would seem so. Using psycopg:

import psycopg
from psycopg import sql

con = psycopg.connect("postgresql://postgres:postgres@127.0.0.1:5432/test")
cur = con.cursor()
cur.execute("CREATE VIEW public.foobar (alg, hash) AS VALUES ('md5',
%s)", ['test'])

IndeterminateDatatype: could not determine data type of parameter $1

cur.execute(sql.SQL("CREATE VIEW public.foobar (alg, hash) AS VALUES
('md5', {})").format(sql.Literal('test')))

con.commit()

cur.execute("select * from foobar")
cur.fetchone()

('md5', 'test')

|

2024-11-16 10:28:19.928 UTC [46] DETAIL: parameters: $1 =
'test-param-value'
2024-11-16 10:28:19.928 UTC [46] ERROR: there is no parameter $1 at
character 57

Of course, I /am/ passing a value for parameter $1; and I can trace
that the client lib sends it out on the wire as expected. (Attaching
packet captures.)

Heck, even the PG server itself says, DETAIL: parameters: $1 =
'test-param-value' — so it sees the parameter! But then, immediately
unsees it.

Am I being hit by a PG bug? Is this a known issue?

I'd retested with master version of that client library, and against 6
latest major versions of PostgreSQL server (12 throughout to 17). No
difference across versions spotted; the result is consistently error
42P02.

Is the client library doing something wrong? How can the server claim
there's no parameter $1 immediately after logging its value it has
received?

I did minify a 100-line SSCCE that reproduces the issue and can be shared.

Any advice, or pointers on what to check next besides delving into PG
source, I'd greatly appreciate. Thanks in advance.

Max

--
Adrian Klaver
adrian.klaver@aklaver.com

Achilleas Mantzios <a.mantzios@cloud.gatewaynet.com> writes:

Στις 16/11/24 12:55, ο/η Max Ulidtko έγραψε:

The issue I'm hitting with it is exemplified by server logs like this:

2024-11-16 10:28:19.928 UTC [46] LOG: execute <unnamed>: CREATE VIEW
public.foobar (alg, hash) AS VALUES ('md5', $1);
2024-11-16 10:28:19.928 UTC [46] DETAIL: parameters: $1 =
'test-param-value'
2024-11-16 10:28:19.928 UTC [46] ERROR: there is no parameter $1 at
character 57

At least for SQL level prepared statements the statement has to be one of :
|SELECT|, |INSERT|, |UPDATE|, |DELETE|, |MERGE|, or |VALUES|
|so CREATE is not valid, and I guess the extended protocol prepared
statements aint no different in this regard.

Indeed. To some extent this is an implementation limitation: the
parameter is received (and printed if you have logging enabled),
but it's not passed down to utility statements such as CREATE VIEW.
But the reason nobody's been in a hurry to lift that restriction
is that doing so would open a large can of semantic worms. In a
case like CREATE VIEW, exactly what is this statement supposed to
mean? I assume you were hoping that it would result in replacement
of the Param by a Const representing the CREATE-time value of the
parameter, but why is that a sane definition? It's certainly not
what a Param normally does. On the other hand, if CREATE VIEW
stores the Param as a Param (which is what I think would happen
if we just extended the parameter-passing plumbing), that's unlikely
to lead to a good outcome either. There might not be any $1 available
when the view is used, and if there is one it's not necessarily of
the right data type.

So, pending some defensible design for what should happen and a patch
implementing that, we've just left it at the status quo, which is that
Params are only available to the DML statements Achilleas mentioned.

regards, tom lane

Thanks for replies! I understand now.

Just to clarify user-side motivation: I'm taught that concatenating
data into SQL query strings is bad practice and should be avoided. I
know how to do it safely in my particular case; but apparently the
author of this client library was taught the same, and so their
query-builder doesn't provide the "raw" quoted-interpolation
substitution (the analogue to sql.Literal from Adrian example). Instead
this query-builder relies on the parameters mechanism.

Hence, this limitation forces me to rewrite my query into raw SQL, with
hand-quoting of parameter and query string concatenation.

if CREATE VIEW stores the Param as a Param

This makes zero sense to me... I assumed that $1 would get substituted
*at query time*, resulting in effectively VALUES ('md5',
'test-param-value') -- not persisted into the view definition. Which is
yes, the former option, Tom; it is sane because that's what $1 does in
every other query type.

If I stare into the abyss regardless, and consider the latter option,
the one that makes no sense to me... I don't see how could it possibly
ever work.

With substitution at some "later time" (expressly not CREATE VIEW query
time), how could this ever work?

CREATE VIEW foobar_view (alg, hash) AS VALUES ('md5', $1); -- suppose
the Param is persisted into view (?!?)

SELECT * from foobar_view where alg = $1;
— is this a 1- or 2-parameter query?
— what do both $1's refer to exactly?
* there's $1 in select query referring to values in column alg, and
* there's $1 supposedly persisted into VALUES of view definition,
referring to a different column with potentially different type.

This makes no sense to me.

So I'm a bit surprised that the (IMO) straightforward semantics of
substitution-at-query-time is not supported.

Nevertheless, acknowledging the "patches welcome" status quo sentiment.
This is helpful; thanks again.

Max

On сб, лис 16 2024 at 11:51:18 -05:00:00, Tom Lane
<tgl@sss.pgh.pa.us> wrote:

On 11/17/24 02:09, Max Ulidtko wrote:

Thanks for replies! I understand now.

Just to clarify user-side motivation: I'm taught that concatenating data
into SQL query strings is bad practice and should be avoided. I know how
to do it safely in my particular case; but apparently the author of this
client library was taught the same, and so their query-builder doesn't

Why not name the client?

provide the "raw" quoted-interpolation substitution (the analogue to
sql.Literal from Adrian example). Instead this query-builder relies on
the parameters mechanism.

Per the docs:

https://www.psycopg.org/psycopg3/docs/api/sql.html

"
class psycopg.sql.Literal(obj: Any)

A Composable representing an SQL value to include in a query.

Usually you will want to include placeholders in the query and pass
values as execute() arguments. If however you really really need to
include a literal value in the query you can use this object.

The string returned by as_string() follows the normal adaptation rules
for Python objects.

Example:

s1 = sql.Literal("fo'o")

s2 = sql.Literal(42)

s3 = sql.Literal(date(2000, 1, 1))

print(sql.SQL(', ').join([s1, s2, s3]).as_string(conn))
'fo''o', 42, '2000-01-01'::date

Changed in version 3.1: Add a type cast to the representation if useful
in ambiguous context (e.g. '2000-01-01'::date)
"

It is meant to pass in a value not something else, say an identifier
which is covered by sql.Identifier. The purpose of the sql module is to
build dynamic SQL safely.

Hence, this limitation forces me to rewrite my query into raw SQL, with
hand-quoting of parameter and query string concatenation.

if CREATE VIEW stores the Param as a Param

This makes zero sense to me... I assumed that $1 would get substituted
*at query time*, resulting in effectively VALUES ('md5',
'test-param-value') -- not persisted into the view definition. Which is
yes, the former option, Tom; it is sane because that's what $1 does in
every other query type.

If I stare into the abyss regardless, and consider the latter option,
the one that makes no sense to me... I don't see how could it possibly
ever work.

With substitution at some "later time" (expressly not CREATE VIEW query
time), how could this ever work?

CREATE VIEW foobar_view (alg, hash) AS VALUES ('md5', $1); -- suppose
the Param is persisted into view (?!?)

SELECT * from foobar_view where alg = $1;
— is this a 1- or 2-parameter query?
— what do both $1's refer to exactly?
* there's $1 in select query referring to values in column alg, and
* there's $1 supposedly persisted into VALUES of view definition,
referring to a different column with potentially different type.

This makes no sense to me.

So I'm a bit surprised that the (IMO) straightforward semantics of
substitution-at-query-time is not supported.

Nevertheless, acknowledging the "patches welcome" status quo sentiment.
This is helpful; thanks again.

Max

On сб, лис 16 2024 at 11:51:18 -05:00:00, Tom Lane <tgl@sss.pgh.pa.us>
wrote:

Achilleas Mantzios <a.mantzios@cloud.gatewaynet.com
<mailto:a.mantzios@cloud.gatewaynet.com>> writes:

Στις 16/11/24 12:55, ο/η Max Ulidtko έγραψε:

The issue I'm hitting with it is exemplified by server logs
like this: 2024-11-16 10:28:19.928 UTC [46] LOG: execute
<unnamed>: CREATE VIEW public.foobar (alg, hash) AS VALUES
('md5', $1); 2024-11-16 10:28:19.928 UTC [46] DETAIL:
parameters: $1 = 'test-param-value' 2024-11-16 10:28:19.928
UTC [46] ERROR: there is no parameter $1 at character 57

At least for SQL level prepared statements the statement has to be
one of : |SELECT|, |INSERT|, |UPDATE|, |DELETE|, |MERGE|, or
|VALUES| |so CREATE is not valid, and I guess the extended
protocol prepared statements aint no different in this regard.

Indeed. To some extent this is an implementation limitation: the
parameter is received (and printed if you have logging enabled), but
it's not passed down to utility statements such as CREATE VIEW. But
the reason nobody's been in a hurry to lift that restriction is that
doing so would open a large can of semantic worms. In a case like
CREATE VIEW, exactly what is this statement supposed to mean? I assume
you were hoping that it would result in replacement of the Param by a
Const representing the CREATE-time value of the parameter, but why is
that a sane definition? It's certainly not what a Param normally does.
On the other hand, if CREATE VIEW stores the Param as a Param (which
is what I think would happen if we just extended the parameter-passing
plumbing), that's unlikely to lead to a good outcome either. There
might not be any $1 available when the view is used, and if there is
one it's not necessarily of the right data type. So, pending some
defensible design for what should happen and a patch implementing
that, we've just left it at the status quo, which is that Params are
only available to the DML statements Achilleas mentioned. regards, tom
lane

--
Adrian Klaver
adrian.klaver@aklaver.com

On Sun, Nov 17, 2024 at 3:10 AM Max Ulidtko <ulidtko@gmail.com> wrote:

I assumed that $1 would get substituted *at query time*

The "select" portion of a "create view" query does not get executed during
processing of "create view" and it is the "select" execution part of the
executor code that performs parameter substitution. IOW, "at query time"
never actually happens, the query is only considered as a literal to be
parsed.

David J.

Στις 16/11/24 18:09, ο/η Adrian Klaver έγραψε:

On 11/16/24 03:15, Achilleas Mantzios wrote:

Στις 16/11/24 12:55, ο/η Max Ulidtko έγραψε:

Greetings, group!

I'm trying to understand a low-level issue. Am evaluating a new
client library for Postgres; it's not particularly popular /
mainstream, and as I've understood so far, sports an independent
implementation of PG binary protocol.

The issue I'm hitting with it is exemplified by server logs like this:

2024-11-16 10:28:19.927 UTC [46] LOG: statement: SET client_encoding
= 'UTF8';SET client_min_messages TO WARNING;
2024-11-16 10:28:19.928 UTC [46] LOG: execute <unnamed>: CREATE VIEW
public.foobar (alg, hash) AS VALUES ('md5', $1);

At least for SQL level prepared statements the statement has to be
one of :

|SELECT|, |INSERT|, |UPDATE|, |DELETE|, |MERGE|, or |VALUES|

|so CREATE is not valid, and I guess the extended protocol prepared
statements aint no different in this regard.

It would seem so. Using psycopg:

import psycopg
from psycopg import sql

con =
psycopg.connect("postgresql://postgres:postgres@127.0.0.1:5432/test")
cur = con.cursor()
cur.execute("CREATE VIEW public.foobar (alg, hash) AS VALUES ('md5',
%s)", ['test'])

IndeterminateDatatype: could not determine data type of parameter $1

cur.execute(sql.SQL("CREATE VIEW public.foobar (alg, hash) AS VALUES
('md5', {})").format(sql.Literal('test')))

con.commit()

cur.execute("select * from foobar")
cur.fetchone()

('md5', 'test')

I dont know python but this does not look like a solid prepared statement.

https://www.psycopg.org/psycopg3/docs/advanced/prepare.html

Does not seem to have used the prepared statement circuitry.

On 11/17/24 11:44, Achilleas Mantzios wrote:

Στις 16/11/24 18:09, ο/η Adrian Klaver έγραψε:

On 11/16/24 03:15, Achilleas Mantzios wrote:

Στις 16/11/24 12:55, ο/η Max Ulidtko έγραψε:

Greetings, group!

I'm trying to understand a low-level issue. Am evaluating a new
client library for Postgres; it's not particularly popular /
mainstream, and as I've understood so far, sports an independent
implementation of PG binary protocol.

The issue I'm hitting with it is exemplified by server logs like this:

2024-11-16 10:28:19.927 UTC [46] LOG: statement: SET client_encoding
= 'UTF8';SET client_min_messages TO WARNING;
2024-11-16 10:28:19.928 UTC [46] LOG: execute <unnamed>: CREATE VIEW
public.foobar (alg, hash) AS VALUES ('md5', $1);

At least for SQL level prepared statements the statement has to be
one of :

|SELECT|, |INSERT|, |UPDATE|, |DELETE|, |MERGE|, or |VALUES|

|so CREATE is not valid, and I guess the extended protocol prepared
statements aint no different in this regard.

It would seem so. Using psycopg:

import psycopg
from psycopg import sql

con =
psycopg.connect("postgresql://postgres:postgres@127.0.0.1:5432/test")
cur = con.cursor()
cur.execute("CREATE VIEW public.foobar (alg, hash) AS VALUES ('md5',
%s)", ['test'])

IndeterminateDatatype: could not determine data type of parameter $1

cur.execute(sql.SQL("CREATE VIEW public.foobar (alg, hash) AS VALUES
('md5', {})").format(sql.Literal('test')))

con.commit()

cur.execute("select * from foobar")
cur.fetchone()

('md5', 'test')

I dont know python but this does not look like a solid prepared statement.

https://www.psycopg.org/psycopg3/docs/advanced/prepare.html

Does not seem to have used the prepared statement circuitry.

The second example is not and was not meant to be. It was meant to show
how you could dynamically create an SQL statement when it will not
accept parameters.

FYI, this was run using psycopg(3) which does things differently, by
default, then psycopg2. For full explanation see:

https://www.psycopg.org/psycopg3/docs/basic/from_pg2.html

Server-side binding

--
Adrian Klaver
adrian.klaver@aklaver.com