Обсуждение: TODO Alter Table Rename Constraint
Hi Everybody!
I looked up this todo, and figured out a plan, how the implementation could be written.
The main challenge is to keep constraints and indexes (for unique and PK constraints) consistent.
Fortunately this is already realized in the case of an index. So at ALTER INDEX RENAME the consistency is given by an extra check in tablecmds.c (Line 2246), where a function is finally called, RenameConstraintById (pg_constraint.c Line 604).
My idea is, to do that analog for ALTER CONSTRAINT RENAME too. So renaming the constraint is going to be done in tablecmds.c as for indexes, tables, sequences, views. And after checking whether the renametype is constraint, an extra rename has to be done for the index. Getting the index can be done with the function get_constraint_index (pg_depend.c Line 475). Now it should be possible to do the same as in RenameConstraintById.
Is that so legal? Is anything else to be considered?
I appreciate any suggestion, and maybe some help too, as I'm not so familiar with the source.
Thanks in advance,
Viktor
(Student of technical unverstity of Vienna)
On Tue, Nov 9, 2010 at 10:50 AM, Viktor Valy <vili0121@gmail.com> wrote: > Hi Everybody! > I looked up this todo, and figured out a plan, how the implementation could > be written. > The main challenge is to keep constraints and indexes (for unique and PK > constraints) consistent. > Fortunately this is already realized in the case of an index. So at ALTER > INDEX RENAME the consistency is given by an extra check in tablecmds.c (Line > 2246), where a function is finally called, RenameConstraintById > (pg_constraint.c Line 604). > My idea is, to do that analog for ALTER CONSTRAINT RENAME too. So renaming > the constraint is going to be done in tablecmds.c as for indexes, tables, > sequences, views. And after checking whether the renametype is constraint, > an extra rename has to be done for the index. Getting the index can be done > with the function get_constraint_index (pg_depend.c Line 475). Now it > should be possible to do the same as in RenameConstraintById. > Is that so legal? Is anything else to be considered? I think the biggest problem is handling inherited tables properly, especially in complex inheritance hierarchies where there are multiple, separate paths from the top of the hierarchy to the bottom. See here for a couple of relevant test cases: http://archives.postgresql.org/pgsql-hackers/2010-07/msg01570.php I believe that the rename needs to fail if any table in the inheritance hierarchy rooted at the target table also inherits the constraint from someplace outside that hierarchy; or if any table in that hierarchy has a local copy of the constraint that got merged with the inherited one. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
Thanks for your answer!
I'm not really familiar with inheritance, but I wonder how this issue is handled in other cases, for instance renaming an index, which invokes internal a constraint rename too. Is that relevant or is the renaming of constraints so special?
Is there a solution for the test-cases you have posted? Or is this yet a problem?
Viktor
2010/11/9 Robert Haas <robertmhaas@gmail.com>
On Tue, Nov 9, 2010 at 10:50 AM, Viktor Valy <vili0121@gmail.com> wrote:I think the biggest problem is handling inherited tables properly,
> Hi Everybody!
> I looked up this todo, and figured out a plan, how the implementation could
> be written.
> The main challenge is to keep constraints and indexes (for unique and PK
> constraints) consistent.
> Fortunately this is already realized in the case of an index. So at ALTER
> INDEX RENAME the consistency is given by an extra check in tablecmds.c (Line
> 2246), where a function is finally called, RenameConstraintById
> (pg_constraint.c Line 604).
> My idea is, to do that analog for ALTER CONSTRAINT RENAME too. So renaming
> the constraint is going to be done in tablecmds.c as for indexes, tables,
> sequences, views. And after checking whether the renametype is constraint,
> an extra rename has to be done for the index. Getting the index can be done
> with the function get_constraint_index (pg_depend.c Line 475). Now it
> should be possible to do the same as in RenameConstraintById.
> Is that so legal? Is anything else to be considered?
especially in complex inheritance hierarchies where there are
multiple, separate paths from the top of the hierarchy to the bottom.
See here for a couple of relevant test cases:
http://archives.postgresql.org/pgsql-hackers/2010-07/msg01570.php
I believe that the rename needs to fail if any table in the
inheritance hierarchy rooted at the target table also inherits the
constraint from someplace outside that hierarchy; or if any table in
that hierarchy has a local copy of the constraint that got merged with
the inherited one.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
On Wed, Nov 10, 2010 at 6:32 AM, Viktor Valy <vili0121@gmail.com> wrote: > Thanks for your answer! > I'm not really familiar with inheritance, but I wonder how this issue > is handled in other cases, for instance renaming an index, which invokes > internal a constraint rename too. Is that relevant or is the renaming of > constraints so special? Indexes can't be inherited, so the problem doesn't arise in that case. > Is there a solution for the test-cases you have posted? Or is this yet a > problem? We had a bug related to the handling of ALTER TABLE .. ADD/DROP CONSTRAINT for those test cases, which I fixed. I think we still have a similar problem with ALTER TABLE .. ADD/DROP ATTRIBUTE, which I haven't fixed because it's hard and I haven't had time, and no one seems to care that much. My point was just that whatever patch you come up with for ALTER TABLE .. RENAME CONSTRAINT should probably be tested against those cases to see if it behaves correctly. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
OK, I see. Thanks for mentioning it.
Are there other problems with the suggestion? Or should the work like that?
Viktor
2010/11/10 Robert Haas <robertmhaas@gmail.com>
On Wed, Nov 10, 2010 at 6:32 AM, Viktor Valy <vili0121@gmail.com> wrote:Indexes can't be inherited, so the problem doesn't arise in that case.
> Thanks for your answer!
> I'm not really familiar with inheritance, but I wonder how this issue
> is handled in other cases, for instance renaming an index, which invokes
> internal a constraint rename too. Is that relevant or is the renaming of
> constraints so special?We had a bug related to the handling of ALTER TABLE .. ADD/DROP
> Is there a solution for the test-cases you have posted? Or is this yet a
> problem?
CONSTRAINT for those test cases, which I fixed. I think we still have
a similar problem with ALTER TABLE .. ADD/DROP ATTRIBUTE, which I
haven't fixed because it's hard and I haven't had time, and no one
seems to care that much. My point was just that whatever patch you
come up with for ALTER TABLE .. RENAME CONSTRAINT should probably be
tested against those cases to see if it behaves correctly.
--
On Fri, Nov 12, 2010 at 4:28 AM, Viktor Valy <vili0121@gmail.com> wrote: > OK, I see. Thanks for mentioning it. > Are there other problems with the suggestion? Or should the work like that? I think you'll just need to give it a try and see how it goes. I think we've covered most of the possible sticking points that I know about, but of course there could be some I don't know about. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company