Is “Where IN” with multiple columns defined in Standard SQL?

Question

I\'m working on a query like this:

SELECT * FROM requests where (id,langid) IN (SELECT nid,langid FROM node)

My questions are

does this

Answer 1

I checked that with PostgreSQL and it works (it is officially supported), but it's your responsibility to make id ↔ nid and langid ↔ langid column types compatible (or use explicit casting).

I think it is pretty standard construct. I have SQL:2003 draft and there is in predicate defined (as well as mentioned exists predicate).

8.4 <in predicate>

Function

Specify a quantified comparison.

Format

<in predicate> ::= <row value predicand> <in predicate part 2>
<in predicate part 2> ::= [ NOT ] IN <in predicate value>
<in predicate value> ::=
<table subquery>
| ... (rest is not important here)

EDIT:

As checked works well under MySQL too (version 5.0.90-log). Here is documentation link.

Answer 2

Standard and portable SQL would be EXISTS.. and is semantically the same IN

SELECT *
FROM requests R
WHERE 
    EXISTS (SELECT *
           FROM node n
           WHERE r.id = n.nid AND r.langid = n.langid
           )

The multi-column IN isn't portable to SQL Server or Sybase at least.

Other notes:

• A JOIN may require a DISTINCT and is not the same as IN or EXISTS.
• The final option is INTERSECT which is less commonly supported and works like IN/EXISTS
• IIRC some prehistoric MySQL versions (3.x?) didn't support the correlation for EXISTS

Answer 3

Your SQL code is valid as regards the SQL-92 Standard. You can confirm this for yourself using the Mimer SQL-92 online validator (SQL-99 and SQL:2003 flavours also available). However, because it is a feature of Full SQL-92, it is not as widely implemented as perhaps it should be.

Relationally speaking, the operator in question is a semi-join, for which none of the SQL Standards (and none of the vendors' extensions?) has an explicit syntax.

Answer 4

Your query would work in Postgres. Best I'm aware, not in MySQL.

The portable version for DBs that would support it is:

SELECT * FROM requests where ROW(id,langid) IN (SELECT nid,langid FROM node)

(row is a reserved keyword since SQL:1999.)

A more portable version will be to use exists() as suggested in the other answer.