From: Kevin Ballard Date: Mon, 02 Jun 2008 17:45:04 GMT Subject: Re: git-svn pulling down duplicate revisions Message-ID: In-Reply-To: <20080602104225.GA8401@untitled> On Jun 2, 2008, at 3:42 AM, Eric Wong wrote: > Kevin Ballard wrote: >> On Jun 1, 2008, at 10:40 PM, Eric Wong wrote: >> >>> Kevin Ballard wrote: >>>> On Jun 1, 2008, at 10:00 PM, Eric Wong wrote: >>>> >>>>> Kevin Ballard wrote: >>>>>> I started a git-svn clone on a large svn repository, and I >>>>>> noticed >>>>>> that for various branches, it kept pulling down the exact same >>>>>> revisions (starting at r1). In other words, if I had 4 branches >>>>>> that >>>>>> shared common history, their common history all got pulled down 4 >>>>>> times. I double-checked, and the created commit objects were >>>>>> identical. >>>>>> >>>>>> Why was git-svn pulling down the same revisions over and over, >>>>>> when >>>>>> it >>>>>> already knows it has a commit object for those revisions? >>>>> >>>>> Can you give me an example if a repository and command-line you >>>>> used >>>>> that does this? Did you use 'git svn clone -s' or did you >>>>> manually >>>>> specify the branch locations in the repo? >>>>> >>>>> It could even be a lack of read permissions to the repository root >>>>> that would cause things like this. >>>> >>>> The repository is, unfortunately, a private repo so I can't share >>>> it. >>>> I used `git svn clone -s` to clone it. I have the SVN perl bindings >>>> v1.4.4 (according to git svn --version). >>>> >>>> I definitely have read permissions to the repo root. If I specify >>>> to >>>> only fetch -r 12000:HEAD (there's 14000-odd revisions), it doesn't >>>> pull down any duplicates, but when I let it start from the root, it >>>> pulls down hundreds of duplicates for multiple branches. >>> >>> Can you at least send me the 'svn log -v' output for that repo? >>> Feel free to leave out the actual log messages and munge the path >>> names if you can't expose that information. >> >> I'll have to do it tomorrow when I'm at the office. How much log info >> do you need? I can let it run until I see duplicate revisions (it's >> pretty obvious, it starts over again from r1). > > I'll need the revisions where branches were created from > the common ancestor (presumably trunk) and some revisions > before it. > > For debugging problems with restricted repositories, it may be worth > it > to create a repository skeleton cloning tool that just reads the > output > of 'svn log --xml -v' and recreates a new SVN repository with: > > * all log messages stripped > > * all new files are created with just a random string in them (to > throw off rename detection on the git side) > (except symlinks, see below) > > * all path components tokenized and each token replaced with > a dictionary value. Something like: > > @tmp = map { $tok{$_} ||= ++$i; $tok{$_} } split(/\//, $old_path); > $new_path = join('/', @tmp); > > This way all copy history can be preserved > > * all modified files will just get a random byte appended to them > > * all committer names replaced with a dictionary value (similar to > what is done to path components). Isn't there a script somewhere that's supposed to do this? Do you know where it is? Incidentally, I just checked and when I start the git-svn clone, it starts pulling down revisions for the branch 'css_refactor@1559' (odd branch name, but it claimed to find multiple branch points for this 'css_refactor' branch). My guess is when it starts working on the next branch, it doesn't view it as related to css_refactor and starts pulling down the revisions again even though those revisions actually belonged to trunk. -Kevin -- Kevin Ballard http://kevin.sb.org kevin@sb.org http://www.tildesoft.com