Обсуждение: Are scalar type's in/out functions implicitly STRICT?

So, I create some new scalar type and I don't declare its in/out
functions as STRICT.  But PostgreSQL copes with them such as they
would be STRICT - they never get the NULL value.  And this is not
documented, AFAIK.  Have I missed something or is it a some hard
well-known etc. fact?  Thanks in advance.

Denis Zaitsev <zzz@cd-club.ru> writes:
> So, I create some new scalar type and I don't declare its in/out
> functions as STRICT.  But PostgreSQL copes with them such as they
> would be STRICT - they never get the NULL value.

This is likely true in many places --- for example, COPY has a shortcut
because it wants to substitute \N for nulls.  It would not be a bright
idea to assume that it's true in every place and forevermore.  In
general, if you have a C function and don't want to be bothered with
explicit PG_ISNULL testing, you'd better mark it STRICT.

An example of an easy way to crash an I/O function without such guards
is to invoke it explicitly:

regression=# select textin(null::cstring);textin
--------

(1 row)

regression=# select textout(null::text);textout
---------

(1 row)

        regards, tom lane

On Fri, Feb 28, 2003 at 05:45:44 +0500, Denis Zaitsev <zzz@cd-club.ru> wrote:
> So, I create some new scalar type and I don't declare its in/out
> functions as STRICT.  But PostgreSQL copes with them such as they
> would be STRICT - they never get the NULL value.  And this is not
> documented, AFAIK.  Have I missed something or is it a some hard
> well-known etc. fact?  Thanks in advance.

Are you using the version one calling sequence?
The version zero method doesn't have a way handle nulls other than by
declaring the function strict.

On Thu, Feb 27, 2003 at 09:44:37PM -0500, Tom Lane wrote:
> Denis Zaitsev <zzz@cd-club.ru> writes:
> > So, I create some new scalar type and I don't declare its in/out
> > functions as STRICT.  But PostgreSQL copes with them such as they
> > would be STRICT - they never get the NULL value.
> 
> This is likely true in many places --- for example, COPY has a shortcut
> because it wants to substitute \N for nulls.  It would not be a bright
> idea to assume that it's true in every place and forevermore.  In
> general, if you have a C function and don't want to be bothered with
> explicit PG_ISNULL testing, you'd better mark it STRICT.
> 

Ok, thanks.

On Thu, Feb 27, 2003 at 09:24:44PM -0600, Bruno Wolff III wrote:
> On Fri, Feb 28, 2003 at 05:45:44 +0500,
>   Denis Zaitsev <zzz@cd-club.ru> wrote:
> > So, I create some new scalar type and I don't declare its in/out
> > functions as STRICT.  But PostgreSQL copes with them such as they
> > would be STRICT - they never get the NULL value.  And this is not
> > documented, AFAIK.  Have I missed something or is it a some hard
> > well-known etc. fact?  Thanks in advance.
> 
> Are you using the version one calling sequence?

Yes, I am - v1.

> The version zero method doesn't have a way handle nulls other than by
> declaring the function strict.
> 

Hmm, examples/docs show such a way for v0...