RE: svn info --recursive isn't reporting tree-conflict-only nodes

On Tue, 2011-05-03 at 18:02 +0200, Bert Huijben wrote:
>
> > -----Original Message-----
> > From: Hyrum K Wright [mailto:hyrum_at_hyrumwright.org]
> > Sent: dinsdag 3 mei 2011 17:49
> > To: Julian Foad
> > Cc: Philip Martin; dev_at_subversion.apache.org
> > Subject: Re: svn info --recursive isn't reporting tree-conflict-only nodes
> >
> > On Tue, May 3, 2011 at 10:19 AM, Julian Foad <julian.foad_at_wandisco.com>
> > wrote:
> > > On Tue, 2011-05-03 at 14:36 +0100, Philip Martin wrote:
> > >> Julian Foad <julian.foad_at_wandisco.com> writes:
> > >>
> > >> > I found a bug in "svn info -R": it doesn't report a node that has a
> tree
> > >> > conflict but otherwise is nonexistent, except if this node is the
> root
> > >> > node of the requested target path.
> > >>
> > >> I think that's an actual-only node. See
> > >>
> > >> http://subversion.tigris.org/issues/show_bug.cgi?id=3779
> > >>
> > >> which says that we need to ensure that commands behave sensibly on
> > >> actual-only nodes.
> > >
> > > Yes.
> > >
> > > The 'info' code uses svn_wc__internal_walk_children(), and makes a
> > > special case of checking for a tree conflict on the target node if that
> > > walk function returns a NOT FOUND error. But 'info' doesn't check for
> > > tree conflicts on other unvisited leaf nodes that may exist as
> > > actual-only nodes.
> > >
> > > One way or another, 'info' is going to have to walk a tree of nodes that
> > > includes actual-only nodes. It could either do this itself or we could
> > > have a walker function that does that. It's logically re-usable
> > > functionality (even if 'info' is currently the only user) so we should
> > > have some kind of walker function that does that. So I'll look at
> > > adding this functionality into svn_wc__internal_walk_children().
> >
> > That's my "gut feeling" as well. ACTUAL-only children are still
> > children (in some sense), and should be included in the set of paths
> > returned by a walk of the children.
>
> In some ways it does, but then every callback has to start with a check 'is
> this an actual only child?'
>
> And most likely we forget that check in a dozen places, because it is so
> unlikely to find these actual only children in test scenarios.
>
> So the result will probably be that Subversion fails for end users...
> because almost none of our functions accept these actual only children. (I
> think only read_info with a conflict argument and the tree conflict
> functions do)

I'm thinking of making the walker do this ONLY when given an additional
flag: "include actual-only nodes". Does that change your perspective?

- Julian
Received on 2011-05-03 18:11:00 CEST