svn到git转换（如何检查存储库质量）

Question

Currently I'm planning to help with a fairly large git conversion for an open-source project. The repository is quite large so trial & error is slow (over 60,000 commits).

There are many questions relating to how a git conversion should be done but almost no details on how to check if the conversion is valid.

Of course there are basics like setting the revisions in both repos and comparing the contents of the repositories, but history, commit messages... moving files, tracking changes between branches etc - become more involved.

So my questions are:

• Which areas should be checked in a newly converted git repository to see that the conversion is correct and succeeded?
• What are the the gotcha's/pitfalls to watch out for?
• Can anyone suggest strategies for evaluating a converted svn project to be sure nothing went wrong during conversion?

---

note : currently we're using reposurgeon however that should have no baring on the answer, though it does mean we have to do a once-off conversion and get-it-right .

Answer 1

If you use the git-svn plugin you can just clone your SVN repository with a local git client which effectively creates a Git repo complete with intact history. Not only is this super easy and super quick but you can trust that you are properly synchronized with the existing svn repo and even pull changes that may have happened after the initial clone.

As for things to look out for, git does not track empty folders. Another gotcha is dealing with binary files that you hopefully aren't storing in your existing repo. You dont want to store large binaries in a git repo typically. There are a few git specific solutions you can google but it might be tricky if you are pulling them by cloning the svn repo.

Answer 2

This is not a definitive answer, just some things we have been doing to check the git conversion is ok.

Find all commits with 3+ parents, its quite unlikely these are valid, though there may be exceptions where it is.

git log --all --min-parents=3

Find all commits with duplicate parents. note that running git filter-branch can clean these up after, but that can sometimes be a really slow process.

git log --all --min-parents=2 --format="format:%H: %P" | egrep ':[^:]* ([0-9a-f]+) [^:]*\1'

Find all commits that are not merges and don't change any files (possibly svn props change):

git log --all --max-parents=1 --format="format:%H" --shortstat | pcregrep -v -M "^[a-z0-9]+\n "

Thanks to Julien Rivaud for the regex commands!