Daniel Shahaf wrote on Tue, Jun 07, 2011 at 21:40:00 +0300:
> Bert Huijben wrote on Tue, Jun 07, 2011 at 15:09:12 +0200:
> > > -----Original Message-----
> > > From: danielsh_at_apache.org [mailto:danielsh_at_apache.org]
> > > Sent: dinsdag 7 juni 2011 14:27
> > > To: commits_at_subversion.apache.org
> > > Subject: svn commit: r1132972 -
> > > /subversion/trunk/subversion/libsvn_fs_fs/revprops-db.sql
> > >
> > > +++ subversion/trunk/subversion/libsvn_fs_fs/revprops-db.sql Tue Jun 7
> > > 12:27:03 2011
> > > @@ -28,7 +28,13 @@ pragma auto_vacuum = 1;
> > > create table revprop (revision integer UNIQUE not null,
> > > properties BLOB not null);
> > >
> > > -create index i_revision on revprop (revision);
> > > +/* Unreleased 1.7-dev libraries also contained an index:
> > > + CREATE INDEX i_revision ON revprop (revision);
> > > +
> > > + This was removed since the UNIQUE statement already constructs
> > > + its own index.
> > > + */
> > > +
> >
> > Shouldn't revision be the primary key instead of just unique?
> >
>
> IIRC there is no functional difference in the case, but I agree that
> spelling out PRIMARY KEY will be better, will look into that.
This time the EXPLAIN's show a significant difference, which to my
untrained eye seems to be in favour of PRIMARY KEY. (They're attached.)
I've made the change in r1133237 (and asked a question in the commit),
but if someone has concerns please voice them.
Received on 2011-06-08 05:45:32 CEST