Bert Huijben wrote on Sun, May 26, 2013 at 19:57:42 +0000: > You open an ra session to one canonical url and then get a session to another > one. > > > At that point you can't trust the relative path apis, nor reparent the > session to something using based on the other url. >
In other words, your problem is that you can't reliably figure out what the "repos root URL" portion of the URL you passed to svn_ra_open() is, right? If so I'd be +1: ensuring the session URL is a prefix of the input URL is indeed the easiest way to fix this, and it doesn't constrain us going forward. > > This breaks checkout, diff, or even simple ‘svn ls’ to these url > forms. > > > The biggest problem is that one of the sources is libsvn_wc, which we > can’t just change.. But that is one of the sources. Most other sources > already canonicalize using svn_uri_canonicalize(). > > > Also note that the applying the ‘localhost’ rule I mentioned as one of > the examples already got a -1 in a recent mail for applying in the > generic uri canonicalize code. > > > Mark Phippard and at least one other user reported this regression on > dev@s.a.o recently. > > > > The proposed change is reverting a 1.8 behavior change with unwanted > side effects. Not relaxing a rule. > > > Making it strict will take quite some work; which as I mentioned > includes a wc.db format bump. But this also includes option parsing > changes and applying more canonicalizations than standardized. > I don't see why that is the case, but never mind. Understanding the problem in terms of svn_ra_open4()'s contract is good enough for me at the moment. Thanks, Daniel > > Bert > > > > Sent from Windows Mail > > > > From: Daniel Shahaf > Sent: Sunday, May 26, 2013 9:25 PM > To: Bert Huijben > Cc: dev@subversion.apache.org > > Bert Huijben wrote on Sun, May 26, 2013 at 20:51:10 +0200: > > > > > > > -----Original Message----- > > > From: danie...@apache.org [mailto:danie...@apache.org] > > > Sent: zondag 26 mei 2013 08:49 > > > To: comm...@subversion.apache.org > > > Subject: svn commit: r1486397 - /subversion/branches/1.8.x/STATUS > > > > > > Author: danielsh > > > Date: Sun May 26 06:48:49 2013 > > > New Revision: 1486397 > > > > > > URL: http://svn.apache.org/r1486397 > > > Log: > > > * STATUS: Ask a question about r1485127. > > > > > > Modified: > > > subversion/branches/1.8.x/STATUS > > > > > > Modified: subversion/branches/1.8.x/STATUS > > > URL: > > > http://svn.apache.org/viewvc/subversion/branches/1.8.x/STATUS?rev=1486 > > > 397&r1=1486396&r2=1486397&view=diff > > > ========================================================== > > > ==================== > > > --- subversion/branches/1.8.x/STATUS (original) > > > +++ subversion/branches/1.8.x/STATUS Sun May 26 06:48:49 2013 > > > @@ -104,6 +104,10 @@ Candidate changes: > > > Restores functionality that was lost in 1.8 compared to 1.7. > > > Votes: > > > +1: rhuijben (for 1.8.1 or soak restart) > > > + +0: danielsh (looks right, but all this does is cause > > > sess->repos_url > > > + to be set to a non-canonical value; I wonder if the > > > right > > > + fix is to add a canonicalize() call in front of a > > > strcmp() > > > + in libsvn_client) > > > > I wish it were that simple. > > > > The usage of file://localhost/some/location to a repository on > > /some/location has been hardcoded in ra_local since before 1.0, but is > > not part of our canonicalization rules in svn_uri_canonicalize(). > > > > The 'simple fix' to change the canonicalization rules by including > > this rule would require a working copy format bump, as this prefix is > > currently perfectly valid as a repository location. > > What is the libsvn_wc problem that setting > svn_ra_session_t.repos_root_url to a different value fixes? (You > haven't said that yet either in the log message or in this thread; you > only said what URLs are involved in it.) > > It is still not clear to me why we are fixing a libsvn_wc issue by > changing an implementation detail of libsvn_ra_local. It sounds like > libsvn_wc could invent its own notion of "equal URL" which ignores drives > letters and considers file:/// and file://localhost/ equivalent, with no > changes to libsvn_ra*. > > With the new code, will svn_ra_session_t->repos_root_url always be set > to a canonical value? The code before your change would. > > Thanks for the additional background. > > Daniel > > > And then for this > > example the api works for us, but is not really part of any RFC. > > > > That 'localhost' refers to the local machine is a convention on many > > platforms, but not really a standard. > > > > On Windows we also have file:///some/path urls that refers to > > E:\some\path when E: happens to be the current harddrive. And this has > > similar conversion problems. > > > > > > I think that for 1.8 we have to keep the same one to many relation > > between local dirents and repository urls, as in 1.0-1.7 :( > > > > For 1.9+ we might be able to apply new canonicalization rules via > > a proper upgrade, but doing this now would just delay 1.8, at no real > > gain for our users. > > > > Bert > > > > > > > > > > > > > > * r1485447, r1485449 > > > Make 'SQLITE_VERSION=3.7.15.1 ./get-deps.sh' work. > > > > > > >
