Re: CVS update: subversion/subversion/libsvn_fs tree.c

Greg Stein <gstein@lyra.org> writes:
> On Mon, Mar 26, 2001 at 07:59:15PM -0000, kfogel@tigris.org wrote:
> >...
> > --- tree.c 2001/03/23 01:21:52 1.65
> > +++ tree.c 2001/03/26 19:59:15 1.66
> >...
> > + /* ### kff todo: is there any compelling reason to get the nodes in
> > + one db transaction? Right now we don't; txn_body_get_root() gets
> > + one node at a time. */
>
> I don't think so. They're all readonly, and the algorithm allows for them to
> move/shift as they will. There isn't an invariant that they are all
> synchronized in some fashion, so there isn't a need to ensure they are read
> in one, synchronous shot.

Actually, the only restriction on the roots passed to svn_fs_merge is
that TARGET_ROOT must be a transaction root. All three of them could
be transaction roots.

svn_fs_merge needs to be a single transaction, to protect it against
people deleting parents of nodes it's working on, etc.
Received on Sat Oct 21 14:36:26 2006