On Fri, Jun 22, 2012 at 7:47 PM, Les Mikesell <lesmikes...@gmail.com> wrote: > On Fri, Jun 22, 2012 at 12:00 PM, Ruhe Julian <jr...@axway.com> wrote: >> >>> I'm sort of surprised a checkout works that way either. >> >> I hope this is an argument for expecting svn:externals to do the same. > > No, it is an argument for you to more closely align your > expectations/usage to what subversion actually does.
Interesting thread. Like Les, I was convinced at first that you were asking something that SVN simply cannot do (given its current design). However, then I noticed that you said: "I want an error on my request "-rHEAD path@peg" if the object was deleted, moved or recreated." So I wondered what the actual behavior of checkout or update would be in those cases (since you compared with behavior of those two subcommands). A quick test: [[[ C:\Temp\svntest>svnadmin create testrepos C:\Temp\svntest>svn mkdir -mm file:///C:/Temp/svntest/testrepos/dir Committed revision 1. C:\Temp\svntest>svn mv -mm file:///C:/Temp/svntest/testrepos/dir file:///C:/Temp/svntest/testrepos/dir2 Committed revision 2. C:\Temp\svntest>svn checkout -rHEAD file:///C:/Temp/svntest/testrepos/dir@1 svn: E160013: File not found: revision 2, path '/dir' ]]] Aha, so checkout cannot follow moves either. That's expected, because svn doesn't know how to do forward history traversal (as already discussed in this thread). But if I understand correctly, that's perfectly valid behavior for your use case: same behavior as checkout. If the object has moved etc, an error is fine. Just to be sure, I checked what happens if you don't move the target of the checkout, but perform some other change below it. Sure enough, checkout does the right thing: it checks out the correct item, at the operative revision: [[[ C:\Temp\svntest>svn mkdir -mm file:///C:/Temp/svntest/testrepos/dir2/subdir Committed revision 3. C:\Temp\svntest>svn co -rHEAD file:///C:/Temp/svntest/testrepos/dir2@2 A dir2\subdir Checked out revision 3. C:\Temp\svntest>dir /b dir2 subdir ]]] Since externals (dir externals at least) are much like an embedded checkout with some sugar on top, I think it should be doable to implement this behavior for externals too. Conclusion: I think you have a valid feature request. Go ahead and file an issue. -- Johan
