Re: BUG #8242: No way to debug "subquery must return only one column" error

On Monday, June 24, 2013 8:59 PM Борис Ромашов wrote:
> Amit, where should I post to force developing this feature ?

You can post this to pgsql-hackers, but I think it would be more better if you can check if any other database support that feature.
I feel you need a more strong case for any developer to work on it and community to agree on it. This is just my personal opinion, so please feel free to work the way you think is best.

2013/6/24 Amit Kapila <amit(dot)kapila(at)huawei(dot)com>
On Monday, June 24, 2013 1:23 PM Борис Ромашов wrote:
>> Why do you want to know the exact row due to which this happens, and what you want to do with it?
> Suppose I have a query that should select something and this query has some subquery that is (possibly by error) supposed to return only one row, i.e. > I mean that it fetches smth unique. But for some row it appears not to be unique.
> I will get that error. But I don't know about my error, I still suppose this to be unique.
> In this case - how can I debug this ? I don't know which row was corrupting uniqueness.
> Moreover, let's assume I have more than one subquery. In this case this is even more complicated to debug error, because I need to check each subquery
> for each row.
It is not straightforward, but you can know by trying some logic like below:
Declare the cursor with corresponding subquery
For i In 1..10 Loop -- this loop is corresponding to outer query values
While(Fetch new row)
{
If fetch returns row more than once then print it.
}

For the part which subquery is giving problem, you might need to break the query into smaller parts and check.

At the moment I am not able to think of any other better way.

>> I don't think there is any way, you can know exactly for which this error occurred.
> Why? Query executor knows what it executes and which row is now. Why cannot it log this info ?
What I mean was that AFAIK currently there is no way to know that, if we enhance the way you are suggesting, then it can possible.
PostgreSQL does something similar for duplicate key, it prints the value for which duplication happens.
postgres=# insert into tbl values(4,2);
ERROR: duplicate key value violates unique constraint "tbl_c1_idx"
DETAIL: Key (c1)=(4) already exists.

With Regards,
Amit Kapila.

2013/6/24 Amit Kapila <amit(dot)kapila(at)huawei(dot)com>
On Friday, June 21, 2013 1:24 PM Борис Ромашов wrote:
> I just realized that I wanted to ask about another error.
> more than one row returned by a subquery used as an expression
> not about
> subquery must return only one column

> 2013/6/21 Борис Ромашов <boraldomaster(at)gmail(dot)com>
> Tom, suppose you haven't understood what the problem I'm facing with.
> Let me explain deeper.
> Try to execute the following 2 queries.
> select (select generate_series(1,2));
> select (select generate_series(1,1));
> They differ only in data, both of them are well-written, so there is not the problem in parsing.
> But first query gives (even in psql)
> ERROR: more than one row returned by a subquery used as an expression

> Certainly - instead of generate_series I could write any usual query that fetches some data from database.
> And if this query returns one row - everything is correct.
> And certainly, instead of selecting from dual (that is how it is called in Oracle) - I could construct more complex external query such that subquery
> could return "more than one row" for just in some exact row (not in each row) of external record set.
> Example
> select id, (select friend.id from user friend where friend.id = user.id) user from user
> This query fetches all users with their friends assuming that every user has only one friend.
> But if some of them will have 2 friends - this query will fail with
> ERROR: more than one row returned by a subquery used as an expression
> And I will have no chance to guess - which user exactly this happened for.
I don't think there is any way, you can know exactly for which this error occurred.
The main reason is that this error occurs when an expression subquery returns more than one row when it is not expected.
In some cases it is okay even if subquery expression returns more than one row, for example:
postgres=# select 1 In (select generate_series(1,2));
?column?
----------
t
(1 row)

postgres=# select 4 In (select generate_series(1,2));
?column?
----------
f
(1 row)

postgres=# select 1 = (select generate_series(1,2));
ERROR: more than one row returned by a subquery used as an expression
postgres=#

Why do you want to know the exact row due to which this happens, and what you want to do with it?

With Regards,
Amit Kapila.