On Tue, Sep 06, 2011 at 06:04:32PM -0400, Paul Burba wrote: > On Tue, May 24, 2011 at 12:35 PM, <s...@apache.org> wrote: > > Author: stsp > > Date: Tue May 24 16:35:14 2011 > > New Revision: 1127134 > > > > URL: http://svn.apache.org/viewvc?rev=1127134&view=rev > > Log: > > As part of issue #3779, "actual-only nodes need regression tests", > > make 'svn add' detect tree conflict victims that do not exist on disk > > and prevent adding new nodes at that path with a meaningful error message. > >
> Hi Stefan, > > While reviewing some outstanding merge-related issues I noticed that > this change breaks the use-case of incoming replacements on local > deletes. The delete portion of the replacement is handled and a > tree-conflict set by the time the add is done and the above error is > raised. Hmmm, this a problem. I remember fixing all sorts of merge replacement issues in 1.6.x. So this might even be a regression from 1.6.x :( > Not exactly sure how to fix this...I can look at it further tomorrow, > just wanted to get your thoughts if you have time. Does reverting the changes made to libsvn_wc/adm_ops.c fix the problem? I was aiming at changing the behaviour of 'add' and 'mkdir', not 'merge'. Both 'add' and 'mkdir' should be covered by the change to libsvn_client/add.c. I probably just added the check to adm_ops.c as well in case some API callers bypass svn_client_add(). But I did not consider the implications for 'merge'.
