Re: [PATCH v2 2/2] alter: zap space_vtab::commit_alter

[Vladimir Davydov <vdavydov.dev@gmail.com>]

On Fri, Apr 06, 2018 at 05:28:47PM +0300, Konstantin Osipov wrote:
> * Vladimir Davydov <vdavydov.dev@gmail.com> [18/04/06 17:25]:
> > On Fri, Apr 06, 2018 at 05:10:01PM +0300, Konstantin Osipov wrote:
> > > * Vladimir Davydov <vdavydov.dev@gmail.com> [18/04/06 16:20]:
> > > 
> > > > space_vtab::commit_alter is implemented only by memtx, which uses it
> > > > to set bsize for a new space. However, it isn't necessary to use
> > > > commit_alter for this - instead we can set bsize in prepare_alter and
> > > > reset it in drop_primary_key, along with memtx_space::replace. Let's
> > > > do it and zap space_vtab::commit_alter altogether, because the callback
> > > > is confusing: judging by its name it should be called after WAL write,
> > > > but it is called before.
> > > >  void
> > > >  DropIndex::alter(struct alter_space *alter)
> > > >  {
> > > > -	space_drop_primary_key(alter->new_space);
> > > > +	if (old_index_def->iid == 0)
> > > > +		space_drop_primary_key(alter->new_space);
> > > >  }
> > > 
> > > I keep wondering why you ditched a nice comment.
> > > If you think it was bad, please write a better one.
> > 
> > This comment is only pertinent to memtx so I just moved it to
> > memtx_space_drop_primary_key.
> 
> > > > -	/*
> > > > -	 * If it's not the primary key, nothing to do --
> > > > -	 * the dropped index didn't exist in the new space
> > > > -	 * definition, so does not exist in the created space.
> > > > -	 */
> 
> What about this hunk?
> 
> Anyway, it's not self-explanatory. Why aren't we doing anything
> here for the secondary index and have to take an extra step for
> primary? 

Because there's an engine that needs it, no? If someone needs more
details, they can look into engine implementation, memtx in this case.

Regarding the particular comment you mentioned. I removed it, because
now it contradicts the code: now we call drop_primary_key if iid == 0,
not if new_space->index[0] == NULL (I explained why in the comment to
the patch). I could "fix" the comment, but then we'd have

        /*
         * If it's not the primary key, nothing to do.
         */
         if (old_index_def->iid == 0)
                return;
         space_drop_primary_key(alter->new_space);

Such a comment would be useless IMHO.