Re: [HACKERS] Re: [GENERAL] drop/rename table and transactions

Vadim Mikheev wrote:

> Hiroshi Inoue wrote:
> >
> > > >
> > > > If there's no objection,I would change UnlockRelation() to not release
> > > > the specified lock except AccessShareLock.
> > >
> > > Why don't remove this call from improper places?
> > > I would try to find all calls and understand why
> > > they made...
> > >
> >
> > I was surprized that few people really want DDL commands inside transactions.
> > Are there any reasons to releasing lock before end of transaction except
> > that long term lock for system tuples is not preferable ?
> >
> > I think that UnlockRelation() is unnecessary fundamentally.
> > Mine is the simplest way to achieve this.
> > If there's no problem,I am glad to remove UnlockRelation() calls.
>
> There are! I finally found where I used UnlockRelation() -
> in execUtils.c:ExecCloseIndices(). Please read comments in
> ExecOpenIndices() where LockRelation() is called...

I see it now.

Hmm,index itself doesn't have its time qualification and is out of
transaction control(at least now).

OK,I would examine it one by one.

Regards.

Hiroshi Inoue
Inoue@tpf.co.jp