Greg Stein wrote on Mon, Apr 23, 2012 at 22:47:43 -0400:
> On Apr 23, 2012 7:20 PM, "Hyrum K Wright" <hyrum.wright_at_wandisco.com> wrote:
> >
> > On Mon, Apr 23, 2012 at 4:55 PM, Greg Stein <gstein_at_gmail.com> wrote:
> > > On Mon, Apr 23, 2012 at 16:21, Hyrum K Wright <hyrum.wright_at_wandisco.com>
> wrote:
> > >> On Mon, Apr 23, 2012 at 12:52 PM, Greg Stein <gstein_at_gmail.com> wrote:
> > >>...
> > >>> Alright. It sounds like svn_ra.h has two purposes: stuff for
> libsvn_client,
> > >>> and stuff for all RA layers to use internally. Ugh.
> > >>
> > >> But the capability strings are part of the public API.
> > >
> > > Says who? Not server Ev2 support. The RA layer can isolate that, so it
> > > should not be "part of the public API".
> > >
> > > Mergeinfo support? Sure. But Ev2? Nah.
> >
> > The entire RA interface is part of the public API. The wire protocols
> > are the public interface to the server, particularly for those who
> > want to write their own clients and server. These strings are part of
> > that definition.
>
> Ah. There we go. That clinches it.
>
> It's too bad we don't have svn_wire.h. Right now, svn_ra serves both
> purposes, so yah... I guess it makes sense to leave it there, unless we'd
> like to start migrating appropriate pieces to a new header. Thoughts?
Haven't followed the thread, but we generally declared only the C API to
be public, while the wire API was considered "internal". (i.e., we'll
have to make it compatible to respect our 1.x-client/1.y-server promise,
but we don't officially support people writing their own wire clients,
etc, not through our libraries)
Received on 2012-04-24 16:09:27 CEST