Re: Updates of SE-PostgreSQL 8.4devel patches (r1268)

Bruce Momjian wrote:
> KaiGai Kohei wrote:
>>>>> Also, having the per-row value always be present in the row and
>>>>> controlled by the bitmask seems ideal;  it avoids having to add a CREATE
>>>>> TABLE option.
>>>> Sorry, I don't understand why it related to a CREATE TABLE option.
>>>> System columns are always allocated for any tables?
>>> Does a table use storage for the security column if no SQL-level
>>> security value is supplied for a given row?
>> When Row-level ACL is enabled on the table and user suppies a tuple
>> without any specific ACLs, it requires security field, because the
>> length of HeapTuple is decided at heap_form_tuple() which is invoked
>> prior to fetching the user supplied ACLs.
>>
>> When Row-level ACL is disabled (by pg_class.reloptions) on the table,
>> the storage for security field is not necessary.
> 
> It is possible to re-call heap_form_tuple() once we know we need a
> security field;  I talked Tom about that.  We can worry about it later.

Is it really possible?

It seems to me the simple_heap_insert() allows to put a value on
the preallocated fixed length field (like "oid"), but does not assume
a situation that whole of the HeapTuple is replaced. Thus, when PGACE
hook is invoked from inside the simple_heap_insert() and it replaces
the given tuple, the caller does not know it and believes the older
tuple was inserted. Does it makes a problematic behavior, isn't it?

Thanks,
-- 
KaiGai Kohei <kaigai@kaigai.gr.jp>