On Wed, Apr 25, 2012 at 2:22 AM, Greg Stein <gstein_at_gmail.com> wrote:
> Hyrum, Daniel, et al,
>
> I just merged libsvn_client/add.c from the ev2-export branch into my
> local working copy, switched from svn_ra_get_commit_editor4() to
> svn_ra__get_commit_ev2(), and ran some of the mkdir URL tests. The
> wiring is fully connected all the way to the FS editor. The tests that
> I tried passed.
>
> The FS editor only supports add_dir and add_file, so not everything
> can switch to the new logic. But import "should" work via ra_local and
> Ev2. With a little more FS editor work, copies and moves could be
> done, too (for 'svn cp/mv URL URL'). (and more and more, of course,
> with further FS editor work)
>
> But short answer: we have true Ev2 wiring from client to ra_local to
> repos to FS.
>
> One note: since I was just testing ra_local, I omitted the three
> callbacks that get_commit_ev2() wants. They're ignored, so I just
> passed NULL. But the other layers will want to insert some shims, and
> they need callbacks. I'm going to craft up some standard callbacks for
> the client library.
>
> I believe the next step(s) is to start merging get_commit_ev2()
> judiciously onto the ev2-export branch. (speed determined by FS editor
> work) Once the callbacks are in place, then the non-ra_local
> providers will function. And that point, stuff can be merged back to
> trunk (depending upon expected performance of the shims on a
> per-commit-flow evaluation).
I've been hoping to replace ra_get_commit_editor4() with
get_commit_ev2() on the branch, but since the client library callers
all use the same set of shims callbacks, I figured I'd replace the
last use of the old editor before cutting over. In that respect, I've
got something that compiles, but fails miserably. It may be a few
days before I have the tuits to fully dig into it.
> Separately, the repos layer's Ev2 commit editor still needs some authz
> checks, but some groundwork has been laid there. Other future work is
> switching mod_dav_svn and svnserve to use the repos Ev2 editor (should
> be supported, as I've been coding the new work against the needs of
> those two subsystems).
>
> It seems we're close to merging much of the ev2-export branch back to
> trunk. (tho not 'export' as that is separate from commit, and it
> relies on unbounded input to the Ev1/Ev2 shim)
Generally, I'd like to see as much of this work on trunk as possible,
given the various performance constraints we may be concerned about.
Thanks for your work!
-Hyrum
--
uberSVN: Apache Subversion Made Easy
http://www.uberSVN.com/
Received on 2012-04-26 00:53:56 CEST