使用移动和重新定位的提交迁移git svn提交

Question

We are trying to migrate our main source repository from svn to git My first try was to simply do an git svn clone. After cloning was done I find that the first commit is in november 2012. I know for a fact that the code is a couple of years old Examining the svn repo it looks like the first commit to the main directory took place in november 2012, while several files inside the repo have been committed years before that. So how can files inside a directory be committed before the parent directory..

I asked around as I'm new in the company, and no one knows for sure, but this repo has experienced a couple of svn relocate and svn move in an unusal way, for example they developed into a branch then moved that branch as a directory inside trunk, or developed into a branch, then renamed that branch as trunk and so on.

I picked one of those old commits that happened before november 2012 and can be queried using git log and tried

svn up -r 599 http://myrepo/trunk

and received the error:

svn: E195012: Unable to find repository location for 'http://myrepo/trunk' in revision 559

This error is reproducible for all committs before november 2012, ie those that can be queried using svn log, but cannot be checked out

The good news is that I'm able to do something like:

svn diff -r 599 http://myrepo/trunk

My next approach is to run through all commits using svn diff, create patches and applies them to git using the original author, date, etc...

Any better ideas how to extract such corrupted commits?

I do not have physical access to the repo, and I can't use svnadmin

Edit 1:

It turned out I was not querying the root directory, but one level down the tree, that's why I was getting such errors. Anyways I did an svnrdump and succeeded in dumping the whole repository (> 32K commits), also the repo contained corrupted commits which I had to skip.

Further it has been imported locally. This helped me further understand what has happened

Basicaly, the repo used to have a very chaotic structure, something like

svn 
   |_Project1 
         |_subproject1 
               |_branches 
                      |_branch1 
                      |_branch2 
               |_trunk 
               |_tags 
                      |_tagv1 
   |_Non-JavaProject 
         |_subproject 
   |_Project2 
          |_AnotherSubproject 
               |_SubSubproject 
          |_Subproject2 
               |_branches 
               |_tags 
          |_Subproject3 
               |_trunk 
          |_Subproject4 
               |_Subsubproject 
                       |_branches 
                       |_tags 
                       |_trunk

What happened afterwards, is that using a mixture of svn mv and copy, the structure was refined a bit

svn |_mainProject |_trunk |_branches |_tags

So now we have paths like

mainProject/trunk/_Project1
mainProject/trunk/_Project1/subproject1
mainProject/trunk/Project2/AnotherSubproject/SubSubproject 

and the same goes for branches/tags

In other words, I need to to define in .git/config in the fetch and other headers a method so that different directories from mainProject will follow different (multiple) old paths

Is this possible?

Answer 1

In SVN branches, including trunks are essentially just directories in a repository. One way you can figure out where your files were before r599 is by running log on the root of SVN repo in verbose mode. Ie svn log -v http://myrepo/@599 It will show you all commits and the paths they touched before r599 in that repository, including commits to branches, renames, etc. One of those will be the commit that moved directories around. This will help you where the files were before that commit and you could checkout previous revision from its old location.