Tom Lane wrote:
> KaiGai Kohei <kaigai@ak.jp.nec.com> writes:
>> Heikki Linnakangas wrote:
>>> What's the point of doing SELECT FOR UPDATE if you're not actually going
>>> to UPDATE the row? Having separate permissions for SELECT FOR UPDATE and
>>> UPDATE seems useless.
>
>> I wonder why SELECT FOR UPDATE need ACL_UPDATE, although the statement
>> itself does not modify any of the given relation.
>
> Because it blocks competing transactions in exactly the same way as an
> UPDATE does. I agree with Heikki --- there is no apparent value in
> having a separate permission bit for this. Given that AclMode is 3/4ths
> full already, I'm not for inventing new privilege types without a very
> strong use-case.
>
> A separate bit for SELECT FOR SHARE might possibly make sense given the
> strength-of-locking argument. But doing both would eat half of the
> available bits, and bring nearer the day that we need a different
> representation for AclMode.
I would not like to have a discussion how many permission bits will be
necessary in the future, because nobody can say something ensured.
We have an another approach that defines ACL_SELECT_FOR_SHARE as
an alias of ACL_SELECT, and applies it on SELECT FOR SHARE statement.
(Needless to say, the targets are already listed, so it might not necessary
to put a ACL_SELECT_FOR_SHARE bit explicitly.)
In the LOCK statement, it checks ACL_SELECT privilege for shared locks and
discriminate between shared and exclusive locks. It seems to me quite natural.
Thanks,
--
KaiGai Kohei <kaigai@kaigai.gr.jp>
