Skip to content

Latest commit

 

History

History
56 lines (39 loc) · 3.81 KB

s2g.md

File metadata and controls

56 lines (39 loc) · 3.81 KB
Raw

s2g - svn-to-git with mixed histories

The Tools

  • s2g-map.php /path/to/folder REV creates a list of revisions and what path the folder had in that revision. REV defaults to HEAD but can be any valid revision number.
  • s2g-import.php REVFILE AUTHORFILE takes a file with a list of revision to path mappings and imports those to a git repository.
  • s2g-am.php PATCHFILE sets environment variables and runs git am to simulate author == committer. Used for transferring post-svn history from existing git repo.

Real-World Example: apertium-spa-arg

First map current latest version:
./s2g-map.php /trunk/apertium-spa-arg | tee map

Looking through the log, the first few revisions are not useful, as they merely create the folder and create an incorrect file and then some skeleton files.

Edit map to remove 61667, 61668, 61669.

Whether 61669 is valuable comes down to how dirty the history would be with that revision in the log. 61669 is lacking a lot of the files that exist in earlier copies of the copied-from folders, so because of how the import works, including 61669 would break linear histories for all those files. I chose to exclude 61669 for that reason.

Then add to the map the revisions from the copied-from folders:
./s2g-map.php /trunk/apertium-es-an | tee -a map

This says the oldest revision is 31754, but the log for 31754 shows that's a newly created empty folder, and the subsequent revision 31755 copies in a bunch of files from /incubator/apertium-es-an.

Edit map to remove 31754, or the import will happily make a revision where all files vanished.

Next up, figure out what's the oldest revision /incubator/apertium-es-an exists in. This requires some manual digging through the log from 31754 backwards, and we can determine 31740 is what deletes the folder, so we want 31739:
./s2g-map.php /trunk/apertium-es-an 31739 | tee -a map

Splice all those revisions to one git repo:
./s2g-import.php map svn-authors.txt

...except that throws an error because 10206 moved the parent /incubator folder from /branches/incubator, which s2g-map.php does not handle.

Wipe /tmp/s2g-* and edit map to change 10198 /incubator/apertium-es-an to 10198 /branches/incubator/apertium-es-an and rerun:
./s2g-import.php map svn-authors.txt

We now have a usable git repo in /tmp/s2g-*/git and we need to merge back the changes made in git since the original svn conversion:
cd /tmp/s2g-*
git clone [email protected]:apertium/apertium-spa-arg.git
cd apertium-spa-arg

Determine when the first conversion was done, and export those changesets:
git format-patch -o ../ -k c160a824^..HEAD
cd ../git

Import the changesets one-by-one and fix errors if they happen:
~apertium/apertium-on-github/s2g-am.php ../0001-*.patch
...
~apertium/apertium-on-github/s2g-am.php ../0009-*.patch

When amending, it can be useful to recreate the changes but keep the old timestamp, plus ensure author == committer. This is done via e.g.:
export 'GIT_COMMITTER_DATE=Sat, 17 Mar 2018 02:02:24 -0500' 'GIT_COMMITTER_NAME=Sushain Cherivirala' '[email protected]'
git commit --author 'Sushain Cherivirala <[email protected]>' --date 'Sat, 17 Mar 2018 02:02:24 -0500' -m 'Update/add .gitignore and .gitattributes'

Ensure all looks good, add remote and force-push changes:
git remote add origin [email protected]:apertium/apertium-spa-arg.git
git push origin master -f