Re: SQL query syntax question

Ah i see, i does make sense when you talk about FROM having access to
relations that have been "created" in some way.

My current goal is to extract all (or as many as possible) external
dependencies from an arbitrary query.
I've created a parser grammar using ANTLR4 and now i need to analyse the
generated syntax trees and that involves this kind of scope and alias
tracking.

Thanks.
Alexander Levsha

On Mon, Apr 4, 2016 at 10:54 AM, David G. Johnston <
david(dot)g(dot)johnston(at)gmail(dot)com> wrote:

> On Monday, April 4, 2016, Alexander Levsha <levsha(dot)alexander(at)gmail(dot)com>
> wrote:
>
>> Hello all.
>> I'm currently developing an SQL parser/analyzer for internal PostgreSQL
>> tool we use at my workplace.
>> Naturally, i use psql in conjunction with PostgreSQL documentation to
>> develop and verify my work.
>>
>> Recently i've noticed the following oddity in query syntax/execution:
>>
>> select (select c1 from tt) from t1 tt;
>> --ERROR: relation "tt" does not exist
>> --LINE 1: select (select c1 from tt) from t1 tt;
>>
>>
>> select (select tt.c1 ) from t1 tt;
>> --c1
>> ------
>> --(0 rows)
>>
>>
>> Why doesn't the first one work when the second one works fine? In other
>> words, why do subquery's traget list and range table list use different
>> namespaces to resolve table references/aliases? Is this an unintended
>> behaviour, implementation detail or is there an actual reason for this?
>>
>> Cannot speak to restrictions imposed by the SQL standard but the
> exhibited behavior seems logical given the nature of subqueries.
>
> The relation named in the FROM clause must exist in the database schema or
> have been previously "created" using a CTE. Within a subquery all from
> clause entries behave the same but you can reference a column by name (with
> usually optional table and schema prefix) if it exists in the containing
> scope. Thus there is no need to complicate things by requiring (or
> allowing) the outer relation names to be targeted by a FROM clause in a
> subquery.
>
> Correlated subqueries require target list resolution to behave in that
> manner. It doesn't seem useful to complicate range table resolution
> lacking a similar need.
>
> What is the motivation for the inquiry?
>
> David J.
>