On Sat, Mar 15, 2014 at 3:03 PM, Bert Huijben <b...@qqmail.nl> wrote:
> > > > -----Original Message----- > > From: Bert Huijben [mailto:b...@qqmail.nl] > > > > > -----Original Message----- > > > From: Philip Martin [mailto:philip.mar...@wandisco.com] > ... > > > > > > SVN_INVALID_REVNUM is less than the baseline revision but the check > > > makes no sense. > > > > I haven't gotten to the full details yet, but... > > > > The check would make sense for a 'checkout' that is required before any > > change in a subtree, which is exactly what happens in http v1. > > And it also works for leaves of the tree, BUT NOT for parent directories > > when one of its descendants is already modified... which is what happens > in > > v2. > > > > I added a python test for this issue in r1577739. > > (Based on Brane's script, but using just one working copy) > > One additional thought. > > Editor V2 aims to remove the depth first restriction for the other RA > layers > (opening the intermediate dirs; which allows the out of date checks), which > is exactly what currently catches this class of out of date error on the > other ra layers. > > Is there some way the FS layer could still deliver that original revision > after making parts of the transaction mutable? > The information is trivially available inside FSFS, FSX and probably BDB (e.g. via noderev predecessor ID) - as long as the respective path has not been deleted within that transaction. So far, there is no API for it. > That would make this easy to fix and I think we will also need it for a > true > editor v2 commit. > > Without it the EditorV2 commit won't be able to trigger this out of date > error in even more of the scenarios, it was designed for... > > > Interesting thought that this matches the problems with the tree conflict > handling on the client side... I didn't think of that before this issue. > -- Stefan^2.
