在ORDER BY子句中强制转换列别名

Question

Given this:

CREATE TABLE test (
  a int,
  b varchar(255)
);

INSERT INTO test VALUES(1, '100');
INSERT INTO test VALUES(1, '50');
INSERT INTO test VALUES(1, '075');
INSERT INTO test VALUES(1, '+50');
INSERT INTO test VALUES(1, '-50');

SELECT a, b AS c
FROM test AS t
ORDER BY t.a, c;

SELECT a, b AS c
FROM test AS t
ORDER BY t.a, t.b::int;

SELECT a, b AS c
FROM test AS t
ORDER BY t.a, c::int;

The last query returns the error:
ERROR: column "c" does not exist: SELECT a, b AS c FROM test AS t ORDER BY ta, c::int

I want to order a result given by a column alias.
If I additionally want to cast a specific column given by an alias it does not work.

What is the reason for ORDER BY alias is working and ORDER BY alias::int is not?
Is there another way to use column aliases in a cast?

Answer 1

This is a big long for a comment.

Here is an early reference to this issue, from which I quote:

ORDER BY (and also GROUP BY ) permit references to output column names only when they are unadorned . You cannot use them in expressions.

If I had to speculate on the reasons for this, it would have to do with expression evaluation. I think the Postgres engine logically rewrites your query to be something like:

SELECT a, b AS c, c::int as INVISIBLEUNSEENCOLUMN
FROM test AS t
ORDER BY t.a, INVISIBLEUNSEENCOLUMN;

That is, it moves the expression to where expressions are usually evaluated (into what I might call the "select" node in the compiled code). Well, this generates the known column error, because that code doesn't understand aliases defined at the same level. When I want to use column aliases in a query, I often use subqueries, to avoid any confusion.

It is easy enough to work around the problem: use a subquery, use the original column name in the order by , or include the cast value in the select .

Answer 2

This is due to the sequence of events in SQL and the Select clause; where column aliases are provided, is second last

1: FROM
2: the JOINS
3: WHERE clause
4: GROUP BY clause
5: HAVING clause
6: SELECT clause
7: ORDER BY clause

ORDER BY is the last so by that stage the SELECT clause has been executed and column aliases are then available to ORDER BY.

However, all clauses prior to the execution of the select clause don't understand the column aliases.

---

most workaround workaround, "nest" the query, eg

select col_alias from ( select x as col_alias from y ) 

---

EDIT: the only exception to this that I know of is in the product Teradata (there may be others but I've used many)