Re: [RFC] Moving from Bazaar (bzr) to git

[Geoffrey Thomas]

On Sat, 22 Jun 2013, Paul Menzel wrote:

> Hopefully everyone will agree, that git is the most popular one in the
> FOSS world (Linux, freedesktop.org, GNOME, KDE, …) and a lot of people
> are now familiar with the basic git commands. So using Bazaar is one
> more thing to deal with for new contributors before making a
> contribution to GRUB. (I know there is git-bzr-ng, but it is still one
> more step.)
>
> Talking to Vladimir and Colin in #grub, they did not oppose a migration
> to git and Vladimir asked me to bring it up on the list.
>
> Does anyone have experience with such a move and how to do it on
> Savannah? Could you please share these? That would be great.

I did a conversion via bzr fast-export | git fast-import a while ago for 
my workplace, so that we could internally use git for development (so we 
all don't have to learn yet another VCS). Our priorities have shifted so 
we haven't hacked on GRUB for a while, but this thread inspired me to dust 
off the conversion, and re-attempt setting up a cronjob to do the import. 
I believe I now have it working and it should push new commits every hour:

  https://github.com/mokafive/grub

I'd be thrilled for GRUB to adopt that history -- let me know if it looks 
reasonable. Until there's a move to git, I intend to keep the URL working, 
so feel free to clone and format-patch from there. I'll try not to break 
history, but if someone points out that something got mis-imported, I'll 
fix that, which will involve rewriting history.

I haven't handled branches, because they don't look very active and 
honestly I don't yet understand bzr branching. I think handling it is 
straightforward, though, so I will figure that out if someone asks.

If you're curious about the details, the scripts I used for the conversion 
are at the end of the email; consider them freely usable with no 
restrictions. Here are the issues I ran into:

* For some reason, git packs the fast-import very, very poorly, causing 
the .git directory to balloon to over a gigabyte (for comparison, the .bzr 
directory is 21M). Running `git gc --aggressive` fixes it and drops .git 
to more like 20M. Note that Github seems to preserve the packing you send 
it, so if you make a new repo on the Github web interface, clones _from_ 
there will be gargantuan if you don't run the aggressive GC before the 
first time you push. I don't seem to need to rerun the GC on incremental 
imports, fortunately.

* A couple of commits (like 
address@hidden / b38045f in my 
import) seem to have an empty author name or email, when fast-exported, 
but a valid committer. So I set the author to the committer.

* .bzrignore needs to become .gitignore, so I changed that in each commit 
so that checking out older commits works reasonably. bzr has some built-in 
defaults that git doesn't, that GRUB makes use of (like *.o), so I added 
those. Also, .bzrignore itself was listed in .bzrignore, which strikes me 
as odd.

* I couldn't easily figure out how to make filter-branch resume from a 
certain point. Fortunately it doesn't take _that_ long to filter the whole 
thing, and is consistent. I think you can do this with .git/info/replace, 
if you really care, though.

There's nothing Github-specific about the import, so it should upload to 
Savannah just fine. I can post the "marks" files, for converting between 
bzr and git commit names, if people find them useful.

As probably implied by the fact that I did this, I'm a fan of moving, both 
because git is significantly more popular and because I find the UI better 
than bzr's (although that may just be familiarity).

--
Geoffrey Thomas
address@hidden

address@hidden grub]$ crontab -l
0 * * * * chronic /home/gthomas/src/grub/import.sh
address@hidden grub]$ cat import.sh
#!/bin/sh

set -e

cd ~/src/grub/bzr
bzr up
cd ~/src/grub/git
bzr fast-export --branch=import --marks=../bzr-marks ../bzr | git fast-import 
--import-marks-if-exists=../git-marks --export-marks=../git-marks --quiet
git reset --hard
../filter.sh
git reset --hard
git push github HEAD:upstream
address@hidden grub]$ cat filter.sh
#!/bin/sh

set -e

export default_ignores=$(python -c "from bzrlib import ignores; print 
\"\\n\".join(ignores.USER_DEFAULTS)")

git update-ref -m "New upstream" refs/heads/upstream import
git checkout upstream
git reset --hard

dir=$(mktemp -d --tmpdir=${XDG_RUNTIME_DIR:-/tmp} filter-branch.XXXXXX)
git filter-branch -f -d "$dir" --env-filter '
    export GIT_AUTHOR_NAME=${GIT_AUTHOR_NAME:-$GIT_COMMITTER_NAME}
    export GIT_AUTHOR_EMAIL=${GIT_AUTHOR_EMAIL:-$GIT_COMMITTER_EMAIL}
   ' --index-filter '
    flag=0
    if git checkout .bzrignore 2>/dev/null; then
        sed "s/^\*\*\///" .bzrignore | grep -v .bzrignore > .gitignore
        echo "$default_ignores" >> .gitignore
        git rm -qf .bzrignore
        git diff .gitignore
        git add .gitignore
        if git show --pretty=format: --name-only "$GIT_COMMIT" | grep -Fxq 
.bzrignore; then
            flag=1
        fi
    fi' --msg-filter '
    cat
    if [ "$flag" = 1 ]; then
        echo
        echo "[git importer: converted .bzrignore to .gitignore]"
    fi'