Horton

Re: Cache lookup errors with functions manipulation object addresses

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Alvaro Herrera <alvherre@2ndquadrant.com>
Cc: Dmitry Dolgov <9erthalion6@gmail.com>, Andrew Dunstan <andrew.dunstan@2ndquadrant.com>, PostgreSQL Developers <pgsql-hackers@lists.postgresql.org>
Date: 2019-10-17T01:37:56Z
Lists: pgsql-hackers

Attachments

On Thu, Sep 26, 2019 at 03:52:03PM +0900, Michael Paquier wrote:
> On Wed, Sep 25, 2019 at 09:21:03AM -0300, Alvaro Herrera wrote:
>> On 2019-Sep-24, Michael Paquier wrote:
>>> + * - FORMAT_TYPE_FORCE_NULL
>>> + *	  if the type OID is invalid or unknown, return NULL instead of ???
>>> + *	  or such
>>
>> I think FORCE_NULL is a strange name for this flag (and its two
>> siblings); I think something like FORMAT_TYPE_INVALID_AS_NULL is
>> clearer.
> 
> Good idea.  I see the point.

Got to think more on this one, and your idea is better.  So changed
this way.

>> I have still to review this comprehensively, but one thing I don't quite
>> like is the shape of the new regression tests, which seem a bit too
>> bulky ... why not use the style elsewhere in that file, with a large
>> VALUES clause to supply input params for a single query? That would be
>> a lot faster, too.
> 
> That makes sense.  Here is how I would write it then:
> WITH objects (classid, objid, objsubid) AS (VALUES
>     ('pg_class'::regclass, 0, 0), -- no relation
>     [ ... ]
>   )
> SELECT ROW(pg_identify_object(objects.classid, objects.objid, objects.objsubid))
>          AS ident,
>        ROW(pg_identify_object_as_address(objects.classid, objects.objid, objects.objsubid))
>          AS addr,
>        pg_describe_object(objects.classid, objects.objid, objects.objsubid)
>          AS descr,
> FROM objects
> ORDER BY objects.classid, objects.objid, objects.objsubid;

That's actually cleaner, so changed as well.  Attached is an updated
patch set with the heap_close() calls removed as per the recent report
from Dmitry.
--
Michael

Commits

  1. Eliminate cache lookup errors in SQL functions for object addresses

  2. Refactor routines for name lookups of procedures and operators

  3. Add new flag to format_type_extended() to get NULL for undefined type

  4. Introduce new extended routines for FDW and foreign server lookups

  5. Refactor routines for subscription and publication lookups

  6. get_relid_attribute_name is dead, long live get_attname

  7. Tweak parser so that there is a defined representation for datatypes