Stefan Sperling wrote: > On Tue, Jan 08, 2013 at 11:03:41AM +0100, Neels Hofmeyr wrote: >> I've just used 'svn patch'. There were missing target files, but I only >> noticed a lot later; because all the text conflicts had .svnpatch.rej >> files, but the missing targets were only listed in 'svn patch' output. >> >> So in my particular case it would have helped to have .svnpatch.rej >> files for missing targets. I would have noticed the lack of a target >> and I could have immediately looked at the diff chunks without >> having to browse through the original patch file. Does that >> generally make sense? It would sometimes need to create parent folders >> to place a rej file... >> >> ~Neels > > Yes, it should probably do that. > However, simple user errors such as using the wrong --strip argument > can create a *lot* of skips. So I'd prefer a single > 'svnpatch.skipped.rej' > within the target dir, rather than littering the working copy with > a lot of 'svnpatch.rej' files. > > Note that UNIX patch creates a single Oops.rej files for files which > don't exist. > > Assume for a second that 'svn patch' already had the ability to flag > conflicts in the working copy (this is a planned feature). Should we > then still create an unversioned skip file or do something else?
My preference: We should create conflicts in the WC, containing sufficient info for the user, including a reference to the original patch file when possible, and *not* write reject file(s) into the WC for skipped targets. Also I would seriously consider doing the same for all conflicting or otherwise rejected hunks, not writing any .rej files at all (after giving due thought to backward compatibility and so on). - Julian
