The Right Stuff

Today we changed the URL of our Subversion server at work to a new domain. Subversion and TortoiseSVN offer a separate relocate command for that, which basically updates the local working copy metadata.

Not so for Git. Git keeps SVN metadata in two places: the commit log messages themselves which hold a git-svn-id entry for all commits that have been pushed to SVN, and in the .git/config file.

commit e82751b4872142679ba61e26fc0c57e97c698e8f
Author: agross 
Date:   Thu Jun 25 16:44:55 2009 +0000

    Adding FxCop to the code quality task

    git-svn-id: https://your.svn-server/svn/Crimson/trunk@67 8ed4a44c-bfb4-4748-a28a-fad9255c4788

[core]
	repositoryformatversion = 0
	filemode = false
	bare = false
	logallrefupdates = true
	ignorecase = true
[svn-remote "svn"]
	url = https://your.svn-server/svn/Crimson/trunk
	fetch = :refs/remotes/git-svn

To update the SVN URL it’s required to update the Git configuration file (an easy edit) and also to rewrite the commit log messages, updating the values of git-svn-id to reflect the new SVN server URL. The latter can be achieved with the git-filter-branch command which allows you to dissect the project history in interesting ways.

Having several local Git repositories to update, I went for the scripted solution. I found this article on how to change the SVN repository URL and added some scripting goodness to it (aside from fixing the syntactic errors). It worked pretty good for my ~10 repositories.

• You pass two arguments:
  • The old SVN URL, i.e. http://old.server
  • The new SVN URL, i.e. https://new.server
  • The old URL will be matched against git-svn-id entries with a regular expression, and the matched parts get replaced with the new URL.
• Rewrite the commit log messages, thus updating git-svn-id
• Create a backup copy of .git/config
• Replace the old SVN URL in .git/config
• Delete all metadata Git has aquired about SVN
• Rebase against SVN, recreating the SVN metadata

#!/bin/sh

# Must be called with two command-line args.
# Example: git-svn-relocate.sh http://old.server https://new.server
if [ $# -ne 2 ]  
then
  echo "Please invoke this script with two command-line arguments (old and new SVN URLs)."
  exit $E_NO_ARGS
fi  

# Prepare URLs for regex search and replace.
oldUrl=`echo $1 | awk '{gsub("[\\\.]", "\\\\\\\&");print}'`
newUrl=`echo $2 | awk '{gsub("[\\\&]", "\\\\\\\&");print}'`

filter="sed \"s|^git-svn-id: $oldUrl|git-svn-id: $newUrl|g\""
git filter-branch --msg-filter "$filter" -- --all

sed -i.backup -e "s|$oldUrl|$newUrl|g" .git/config

rm -rf .git/svn
git svn rebase

For about six month I’m using Git as my preferred Source Control Management system. That is, while my colleagues at work still work with Subversion and TortoiseSVN, I am leveraging the powers of Git behind the scenes. That is, I work on local (Git) feature branches that I commit to our SVN trunk every once in a while, typically after I finished working on a feature.

If you want to get started with Git, I recommend watching the GitCasts and try one of the Windows distributions, msysgit or Cygwin. (I switched to Cygwin after the msysgit team dropped SVN integration in their 1.6 release.)

Git heavily relies on the command line, and UI tools like gitk and TortoiseGit are not quite there yet. Using a CLI might be something a lot of Windows developers do not look forward to, but you still can configure support for some UI tools. For example, file diffs and merges can be customized to work with any tool (UI and CLI-based). I use SourceGear DiffMerge (free) as my primary diff and merge tool. It’s far better that what comes with TortoiseSVN or Git itself (again CLI-based).

I could only find sparse documentation on how to make DiffMerge work with Git, so I dove into the world of Shell programming and created a couple of simple scripts. Here is what I came up with.

diffmerge-diff.sh

#!/bin/sh

path="$(cygpath $1)"
old="$(cygpath --mixed --absolute "$2")"
new="$(cygpath --mixed --absolute "$5")"

#echo -e "path\n$path"
#echo -e "old\n$old"
#echo -e "new\n$new"

"/cygdrive/C/Tools/DiffMerge/diffmerge.exe" "$old" "$new" --title1="Old" --title2="New $path"

diffmerge-merge.sh

#!/bin/sh

localPath="$(cygpath --mixed --absolute "$2")"
basePath="$(cygpath --mixed --absolute "$1")"
remotePath="$(cygpath --mixed --absolute "$3")"
resultPath="$(cygpath --mixed --absolute "$4")"

if [ ! -f $basePath ]
then
    basePath="$(cygpath --mixed --absolute ~/diffmerge-empty)"
fi

#echo -ne "local\n$localPath\n"
#echo -ne "base\n$basePath\n"
#echo -ne "remote\n$remotePath\n"
#echo -ne "result\n$resultPath\n"

"/cygdrive/C/Tools/DiffMerge/diffmerge.exe" --merge --result="$resultPath" "$localPath" "$basePath" "$remotePath" --title1="Mine" --title2="Merged: $4" --title3="Theirs"

Setting up Git

1. Put the files above in your home directory, either C:\Cygwin\home\<username> (Cygwin) or C:\Users\<username> (msysgit). Make sure the line endings are LF-only (UNIX style) and the file is saved without the UTF byte-order mark (BOM).
2. Create an empty file named diffmerge-empty in the same directory.
3. Modify the scripts to point to the correct location of DiffMerge (see the last lines). Some more modifications will be needed if you’re from the msysgit crowd, as the cygpath utility is not available there.
4. Open a Git CLI and enter the following commands:

   git config --global merge.tool diffmerge
   git config --global mergetool.diffmerge.cmd "~/diffmerge-merge.sh \"\$BASE\" \"\$LOCAL\" \"\$REMOTE\" \"\$MERGED\""
   git config --global mergetool.diffmerge.trustExitCode false 
   git config --global diff.external "C:/Cygwin/home/<username>/diffmerge-diff.sh"

   This should add the following lines to your .gitconfig:

   [merge]
       tool = diffmerge
   [mergetool "diffmerge"]
       cmd = ~/diffmerge-merge.sh \"$BASE\" \"$LOCAL\" \"$REMOTE\" \"$MERGED\"
       trustExitCode = false
   [diff]
       external = C:/Cygwin/home/<username>/diffmerge-diff.sh

Using DiffMerge

Using DiffMerge is pretty easy, the normal git diff and git mergetool commands work as usual, but will now spawn DiffMerge instead of vi (which is the default). If anything does not work as expected, uncomment the #echo lines and use a tool like Process Monitor and Process Explorer to see where things go wrong.

Happy diffing!