On Wed, May 16, 2012 at 11:27 PM, Greg Stein <[email protected]> wrote: > > On May 15, 2012 6:04 PM, "Hyrum K Wright" <[email protected]> wrote: >>... > > >> This research comes as a result of the final test failures on the >> ev2-export branch for commit: tree conflicts tests 4 and 8. Both of >> these set up the tree conflict scenario, and in doing so occasionally >> expect an out-of-date error, which we aren't currently generating. >> For instance, test 4 creates a file in r3, updates the working copy to >> r2 (thereby deleting the file), does a local copy to the same location >> as the previously created file, and then attempts to commit. The >> commit should fail, but the FS editor isn't returning any out-of-date >> errors. > > Right. The add/copy should fail because the node already exists. I relied on > that failure, rather than issuing an OOD. We could add that empty-target > verification.
I think we should. The add/copy isn't failing at all in it's current state, which means some check isn't happening as we expect it to. I think the OOD check is the reasonable place for this failure (and the tests seem to indicate it's what we've historically done). >> I think the problem is that we aren't calling can_modify() enough, or >> with the right revision argument. For instance, in the case of a >> copy, we only call can_modify() on the destination path if we are >> replacing it. It feels like we should ensure we can modify the target >> in all instances, not just if we need to replace it. This may also >> apply to other editor callbacks in this file. > > Sure. When REPLACES_REV is INVALID, we can call check_path and verify > nothing exists. > > Seem about right? Sounds reasonable. -Hyrum -- uberSVN: Apache Subversion Made Easy http://www.uberSVN.com/
