constrains of array

Hi.

I use version from cvs.

I want to have array of referencies to another table, so I do:
cms=# create table a (a int primary key); create table b (b int[]
references a);
NOTICE: CREATE TABLE/PRIMARY KEY will create implicit index 'a_pkey' for
table 'a'
CREATE
NOTICE: CREATE TABLE will create implicit trigger(s) for FOREIGN KEY
check(s)
CREATE

But now I can't add anything to table b:
cms=# insert into a values (1);
INSERT 52069 1
cms=# insert into b values ('{1}');
ERROR: Unable to identify an operator '=' for types 'int4' and '_int4'
You will have to retype this query using an explicit cast

I expect either of following:
1) REFERENCIES from array to scalar should create correct trigger (check
every entity of array)

Actually, no. Technically, it should be seeing if the exact same array is
on the other side. AFAIK, There is no definition of the RI constraint
in the spec that works that way. We've talked about implementing such
a beast as an extension, but there are some issues about indexing that
we need to look at before we can do that in general.

2) It should be error in *creation* of table if there is no comparasion
operator for constrain check

Possibly, although it currently doesn't to allow you to add the operator
after you do the references. The benefits of that might be outweighed by
the problems if you don't add the operator.

3) Error in insert should say something about trigger, otherwise user have
to guess what is wrong with his query.

Definately true, but unfortunately i'm not sure if it's possible in the
current implementation (it does another query inside the trigger, and
that's what's erroring -- I don't think it knows it's in the trigger at
that time).

Stephan Szabo <sszabo@megazone23.bigpanda.com> writes:

2) It should be error in *creation* of table if there is no comparasion
operator for constrain check

Possibly, although it currently doesn't to allow you to add the operator
after you do the references. The benefits of that might be outweighed by
the problems if you don't add the operator.

I can't see any good reason not to require the operator to pre-exist.
In fact, there's a good argument that we should require the two columns
to have the exact same datatype.  Otherwise, equality may be a pretty
fuzzy concept.  Think about varchar vs bpchar comparison, for example
--- shall we consider trailing blanks significant?  Which column will
drive the choice?

In any case, it's certainly a bad idea that the system accepted an
FK constraint relating int[] to int.

regards, tom lane

On Tue, 12 Dec 2000, Tom Lane wrote:

Stephan Szabo <sszabo@megazone23.bigpanda.com> writes:

2) It should be error in *creation* of table if there is no comparasion
operator for constrain check

Possibly, although it currently doesn't to allow you to add the operator
after you do the references. The benefits of that might be outweighed by
the problems if you don't add the operator.

I can't see any good reason not to require the operator to pre-exist.

The only case I could see would be if there was some case where you had
equality operators that needed to be defined after the table that had
the references constraint (not sure if that could ever happen). You
could use alter table in these cases though.

In fact, there's a good argument that we should require the two columns
to have the exact same datatype.  Otherwise, equality may be a pretty
fuzzy concept.  Think about varchar vs bpchar comparison, for example
--- shall we consider trailing blanks significant?  Which column will
drive the choice?

I think the spec only requires them to be comparable I believe (I'd
assume that the match predicate rules would apply), so would an equality
operator be sufficient to tell that?

Stephan Szabo <sszabo@megazone23.bigpanda.com> writes:

In fact, there's a good argument that we should require the two columns
to have the exact same datatype.

I think the spec only requires them to be comparable

Oh, in that case never mind ... but I'd still favor checking for
existence of the comparison operator right away.

regards, tom lane

On Tue, 12 Dec 2000, Tom Lane wrote:

Stephan Szabo <sszabo@megazone23.bigpanda.com> writes:

In fact, there's a good argument that we should require the two columns
to have the exact same datatype.

I think the spec only requires them to be comparable

Oh, in that case never mind ... but I'd still favor checking for
existence of the comparison operator right away.

Okay, will do. :)