Horton

Thread

Commits

  1. Mark multirange_constructor0() and multirange_constructor2() strict

  1. multirange constructor strictness

    Peter Eisentraut <peter.eisentraut@enterprisedb.com> — 2021-04-21T20:56:54Z

    The multirange constructors created in makeMultirangeConstructors() are:

multirange_constructor0 -> not strict
multirange_constructor1 -> strict
multirange_constructor2 -> not strict

And both multirange_constructor1 and multirange_constructor2 contain 
code like

/*
  * These checks should be guaranteed by our signature, but let's do them
  * just in case.
  */
if (PG_ARGISNULL(0))
     ereport(ERROR,
             (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
              errmsg("multirange values cannot contain NULL members")));

In case of multirange_constructor2 the "should be guaranteed" comment is 
not actually true right now.  In case of multirange_constructor1, maybe 
this should be downgraded to an elog or assert or just removed.

Is there a reason why we can't make them all three strict or all not 
strict?  (Obviously, it doesn't matter for multirange_constructor0.)  Is 
the fact that multirange_constructor2 is variadic the issue?  Maybe at 
least some more comments would be helpful.

  2. Re: multirange constructor strictness

    Alexander Korotkov <aekorotkov@gmail.com> — 2021-04-22T11:00:38Z

    On Wed, Apr 21, 2021 at 11:57 PM Peter Eisentraut
<peter.eisentraut@enterprisedb.com> wrote:
> The multirange constructors created in makeMultirangeConstructors() are:
>
> multirange_constructor0 -> not strict
> multirange_constructor1 -> strict
> multirange_constructor2 -> not strict
>
> And both multirange_constructor1 and multirange_constructor2 contain
> code like
>
> /*
>   * These checks should be guaranteed by our signature, but let's do them
>   * just in case.
>   */
> if (PG_ARGISNULL(0))
>      ereport(ERROR,
>              (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
>               errmsg("multirange values cannot contain NULL members")));
>
> In case of multirange_constructor2 the "should be guaranteed" comment is
> not actually true right now.  In case of multirange_constructor1, maybe
> this should be downgraded to an elog or assert or just removed.
>
> Is there a reason why we can't make them all three strict or all not
> strict?  (Obviously, it doesn't matter for multirange_constructor0.)  Is
> the fact that multirange_constructor2 is variadic the issue?  Maybe at
> least some more comments would be helpful.

Thank you for noticing.  I'll take care of it today.

------
Regards,
Alexander Korotkov