On Apr 30, 2012 9:26 PM, "Hyrum K Wright" <hyrum.wri...@wandisco.com> wrote: > > On Mon, Apr 30, 2012 at 5:20 PM, Greg Stein <gst...@gmail.com> wrote: > > On Mon, Apr 30, 2012 at 17:58, <hwri...@apache.org> wrote: > >>... > >> @@ -902,9 +956,13 @@ repos_to_repos_copy(const apr_array_head > >> message, ctx, pool)); > >> > >> /* Fetch RA commit editor. */ > >> - SVN_ERR(svn_ra__register_editor_shim_callbacks(ra_session, > >> - svn_client__get_shim_callbacks(ctx->wc_ctx, > >> - NULL, pool))); > >> + shim_callbacks = svn_delta_shim_callbacks_default(pool); > >> + SVN_ERR(svn_ra__register_editor_shim_callbacks(ra_session, shim_callbacks)); > >> + shim_callbacks->fetch_props_func = fetch_props_func; > >> + shim_callbacks->fetch_base_func = fetch_base_func; > >> + shim_callbacks->fetch_kind_func = fetch_kind_func; > >> + shim_callbacks->fetch_baton = path_infos; > > > > Shouldn't the registration occur *after* the values are assigned? I > > recognize that this "works", but it certainly looks strange, and it > > would definitely break if the structure is copied into RA-private > > storage. > > We can debate the semantics of call-by-value and call-by-reference in > this instance, but I think your suggestion makes sense: r1332489
Oh, I have no problem with the approach. We have lots of APIs that state "must live at least as long..." with the implicit by-ref. But we do usually try to finish initializing the thing before passing a reference :-) Thanks for the tweak. Cheers, -g
