Re: git-svn bug?

I've got a repository I've converted over to git from svn. (using git-svn. since there's only been one branch, I figured I could skip git-svnimport).

For quite a while, all I did was fetch/rebase from the svn repository to my git repository; all of my own work was committed to the git repository; none of the changes were commited to the svn repository.

Then came the time to commit changes from my git repository to the svn repository.

Being somewhat cautious, I created an empty 'dummy' svn repository and familiarize myself with using git-svn to commit from git -> svn.

I ran: git-svn fetch git-svn rebase remotes/git-svn (already updated) git-svn dcommit (to push my changes to the svn repository)

Everything seemed to work fine with the dummy repository.

Encouraged, I did the same with the 'real' repository, and received the following error: (Using git 1.4.3.5, svn 1.4.0 on a Gentoo box) fatal: Not a valid object name 92e2e0c50bbbacb0a3426b2c0f8b3e043eb4830a~1 32768 at /usr/lib64/perl5/5.8.8/Memoize.pm line 269

(using git 1.4.4, svn 1.3.0 on a SLES 9 SP3 box) fatal: Not a valid object name 92e2e0c50bbbacb0a3426b2c0f8b3e043eb4830a~1 32768 at /usr/lib/perl5/5.8.3/Memoize.pm line 269

(using git 1.4.4, svn 1.3.1 on a SLES 10 box) fatal: Not a valid object name 92e2e0c50bbbacb0a3426b2c0f8b3e043eb4830a~1 32768 at /usr/lib/perl5/5.8.8/Memoize.pm line 269

I had NFS mounted the git repository, so the SLES 9 was local the other two were NFS. In any event, the error seems to be essentially identical on every platform I've tried.

Additionally, I had created three branches for the purpose of pushing my changes to svn: master (used for my own development) svn (created using git checkout -b svn remotes/git-svn -- basically only what is in svn) merge (git checkout -b merge svn; git rebase master)

`git diff-tree merge master | wc -l` returns 0 (which I assume means no changes) `git diff-tree svn merge | wc -l` returns 44 (again, I assume this means 44 changes)

I couldn't find an object named "92e2e0c50bbbacb0a3426b2c0f8b3e043eb4830a~1" in .git/

Aside from my cluelessness, is there anything else wrong?

"Troy Telford" <ttelford.groups@gmail.com> writes:

Troy, do you have object 92e2e0c5? Is it a root commit (i.e. a commit that does not have a parent)?

The only place that mentions ~1 in git-svn seems to be inside dcommit but it seems that it unconditionally appends ~1 to the rev. I do not know how the code guarantees it does not go down to the root commit.

Eric, any clues?

On Wed, 15 Nov 2006 14:43:30 -0700, Junio C Hamano <junkio@cox.net> wrote:

I'll have to admit I'm stabbing in the dark on how to get the correct answer this, but here goes:

* `git cat-file -t 92e2e0...` returns 'commit'
* 'git cat-file -p 92e2e0...` returns: (minus the header/footer asterisks)
*********************************************
tree 4b825dc642cb6eb9a060e54bf8d69288fbee4904
author unknown <unknown> 961088898 +0000
committer unknown <unknown> 961088898 +0000

New repository initialized by cvs2svn. *********************************************

Sorry for the late replies, I've been caught up with other things.

Troy Telford <ttelford.groups@gmail.com> wrote:

dcommit expects to be run on a git-svn fetch-ed HEAD that is linear superset of remotes/git-svn. That is: remotes/git-svn..HEAD should (ideally) contain no merges, and no root commits. git-svn currently does no checking for root commits, but it should.

This commit is missing the git-svn-id: line at the bottom. If you simply left it out (private svn repository info), can you check that the URL in this line is actually for the SVN repository you want to commit to?

It seems like your usage of dcommit would actually cause the issue you're experiencing to be triggered on the dummy repository, and not the real one. My other guess would be that you somehow merged commits from your dummy svn repo into your master branch.

On Fri, 17 Nov 2006 01:55:10 -0700, Eric Wong <normalperson@yhbt.net> wrote:

I didn't remove anything, but I did double-check and the URL is correct. But the following may shed light on why there is no git-svn-id anywhere: IIRC, this is how my current repository came to be (from the very beginning): 1.) Way back when, before I even started on the project, it started life as a CVS repository 2.) Was converted from CVS -> SVN in early '05 (pre-git) 3.) I converted from SVN->git in Nov/Dec '05 (using git-svnimport. I'm not sure git-svn was available at the time.) 4.) The svn repository is still around, and I need to interoperate with the svn repository on occasion. I read about the new (at the time) 'git-svn', and decided to give it a try. 5.) I start with my pre-existing git repository, running:

   git svn init <url>
   git svn fetch
   git checkout -b master svn
   git rebase remotes/git-svn

I need to work on being more clear; sorry about that.  Here's what I did  
with my 'dummy' repository
1. create a new (empty) svn repository
2. imported it into a new git repository using git-svn
3. added a few files that were just sitting in $HOME, then modified them,  
removed some, added others, etc.  (using both git-svn and subversion)
4. verified everything was working as I expected it to.  (and if not,  
figure out why I was wrong).

My 'dummy' repo was imported using git-svn. My 'real' repo was imported using git-svnimport.

Having not read any of the code, I'm just taking a wild guess; but is it reasonable to say that since the repository was originally imported to git using git-svnimport (rather than git-svn), git-svn doesn't have some of the data it needs to push to the remote svn repo?

Would it be reasonable to use git-svn to import the SVN repository into a new git repo, and then rebase from the old git-svnimport'ed repo into the new git-svn imported one? (did that even make sense?!?)

Troy Telford <ttelford.groups@gmail.com> wrote:

Exactly, git-svn needs the git-svn-id: lines at the end of each commit for 'commit' and 'dcommit' to work. So using a repo created by git-svnimport will not work.

'commit-diff' will work, however it's intended as a low-level command ('commit' should be that, too).

Yes, something along those lines would work.

"Troy Telford" <ttelford.groups@gmail.com> writes:

I always send commits as:

git-svn dcommit remotes/git-svn..master

Possibly replacing master with whatever git branch I'm working on. One nice thing about this approach is that you can sort of preview the commit as:

  git diff remotes/git-svn..master

[but that just shows overall diffs and git-svn dcommit will apply each commit in master separately to svn]

Seth Falcon <sethfalcon@gmail.com> wrote:

Or just:

   git-svn dcommit remotes/git-svn..

to send the current branch.

Or see the log and patch of each commit on the current branch:

   git log -p remotes/git-svn..

Shawn Pearce <spearce@spearce.org> writes:

Yes, but be warned. As I just discovered, dcommit always commits up to the HEAD of the current branch. So I just did:

git-svn dcommit remotes/git-svn..HEAD~3

And the ~3 was ignored. :-( Not a bug, as the doc says that dcommit operates on HEAD, but it would be convenient to have this work -- especially since reviewing with diff (or better log -p) seems natural.

I get essentially the same error: fatal: Not a valid object name 92e2e0c50bbbacb0a3426b2c0f8b3e043eb4830a~1

Since I actually checked out a branch, I assume that

     git diff remotes/git-svn..svn

should be empty, right?

And if that's the case `git diff svn..master` should be identical to `git diff remotes/git-svn..svn`?

If I write the diff to a file, it is about 294k in size. Also worth mentioning is that I'm also tracking binary files in the git repository as well. It's for an RPM tree, so I have the (binary) package source tarballs (in a few cases >100 MB tarball) and their corresponding (text) patches and .spec files.