Re: Getting error 42P02, despite query parameter being sent
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 : 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.kla...@aklaver.com
Re: Getting error 42P02, despite query parameter being sent
Achilleas Mantzios 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 : 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
Getting error 42P02, despite query parameter being sent
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 : 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
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
query42P02-with-prepstatement-on.pcap
Description: application/vnd.tcpdump.pcap
query42P02-with-prepstatement-off.pcap
Description: application/vnd.tcpdump.pcap
Re: 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 : 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.
|
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
