Bugzilla – Full Text Bug Listing
| Summary: | [git mirror] git mirror does not correspond to state in CVS HEAD | ||
|---|---|---|---|
| Product: | Community | Reporter: | Istvan Ballok <Istvan.Ballok> |
| Component: | Git | Assignee: | Eclipse Webmaster <webmaster> |
| Status: | RESOLVED FIXED | QA Contact: | |
| Severity: | blocker | ||
| Priority: | P3 | CC: | austin.riddle, b.muskalla, fwaibel, Istvan.Ballok, rsternberg, ruediger.herrmann |
| Version: | unspecified | ||
| Target Milestone: | --- | ||
| Hardware: | PC | ||
| OS: | Windows 7 | ||
| Whiteboard: | |||
|
Description
Istvan Ballok
I think this is the right slot. FYI in the commit: c04153eac45f2dae8a024896c5c1b03b83b86149, the Scrollable.js was added to the rap git mirror repository. This git commit corresponds to the CVS commit that introduced the 1.2 reversion of this file: http://dev.eclipse.org/viewcvs/index.cgi/org.eclipse.rap/runtime.rwt/org.eclipse.rap.rwt.q07/js/org/eclipse/swt/widgets/Scrollable.js?logsort=date&view=markup&revision=1.2&root=RT_Project i.e., the Scrollable.js is no longer missing in these git commits: c04153eac..HEAD However, of course, it does not solve the problem, that in the 74 commits that cover a 4 week long interval: a00ad92fd59b86f..c04153eac45f2d, the Scrollable.js in its reversion 1.1 is missing. Hello, I hope no one minds me escalating the importance of this bug. I am doing work on a GEF/Draw2D port in RAP and the Git mirrors are incredibly important for allowing me to stay in sync between the projects. As it is now, I am at a standstill because I can't get anymore updates from the RAP Git mirror. I can see that certain remedies have been applied in the following bugs: https://bugs.eclipse.org/bugs/show_bug.cgi?id=322148 https://bugs.eclipse.org/bugs/show_bug.cgi?id=322885 https://bugs.eclipse.org/bugs/show_bug.cgi?id=332365 https://bugs.eclipse.org/bugs/show_bug.cgi?id=298528 Is there any relief in sight for these kind of problems? Is that something short term that can be done in this case? Thanks. I can rebuild the RAP mirror, if that will help. Your best bet for more 'permanent' fix is to either: use CVS like RAP does or to convince RAP to migrate to Git. -M. (In reply to comment #4) > I can rebuild the RAP mirror, if that will help. > > Your best bet for more 'permanent' fix is to either: use CVS like RAP does or > to convince RAP to migrate to Git. > > -M. A rebuild of the mirror might at least resolve the current blockage. We are planning on migrating RAP to Git with our 2.0 release which is still a while off. (In reply to comment #4) How long will the rebuild take? It's done now. -M. (In reply to comment #7) > It's done now. > > -M. Thanks....I tried to fetch and now I receive errors about not being able to load the git URL: git://dev.eclipse.org/org.eclipse.rap/org.eclipse.rap.git There was a delay between the mirror rebuild and it's sync to the front end. -M. For the last 2 days now, I cannot fetch from the Git mirror at all. The error I receive from eGit is: Cannot get repository refs. Connection reset. Is the mirror down? (In reply to comment #10) I also forgot to mention that previous to the aforementioned error, new changes could not be obtained by fetching. So it seems that the out of sync problem still exists. (In reply to comment #11) > (In reply to comment #10) > > I also forgot to mention that previous to the aforementioned error, new changes > could not be obtained by fetching. So it seems that the out of sync problem > still exists. Hello, I can second that. Milestone 7 is *not* present in the git mirror. florian It is unfortunate that it has come to this. Since the Git mirror is seemingly not functional and at best is unreliable, I have moved to the more drastic solution of doing a CVS/Git merge on my machine. Following an approach found online, this entailed checking out from CVS, copying my .git folder into that checkout adding a .cvsignore for the .git directory and a .gitignore for the CVS subfolders. This has enabled me to synchronize CVS with my Git working directory. It is distasteful, but at least I can continue working reliably. (In reply to comment #13) > It is unfortunate that it has come to this. If you have any suggestions on what we can do to make the Git mirrors of CVS reach a level approaching 'reliable' without rebuilding them from scratch every night, then by all means, suggest away. git-cvsimport is functional at best. I just tried this: $ git clone git://dev.eclipse.org/org.eclipse.rap/org.eclipse.rap.git Cloning into org.eclipse.rap... remote: Counting objects: 128729, done. remote: Compressing objects: 100% (29948/29948), done. remote: Total 128729 (delta 78068), reused 128729 (delta 78068) Receiving objects: 100% (128729/128729), 62.05 MiB | 1.77 MiB/s, done. Resolving deltas: 100% (78068/78068), done. warning: remote HEAD refers to nonexistent ref, unable to checkout. $ cd org.eclipse.rap/ $ git checkout master Checking out files: 100% (12447/12447), done. Branch master set up to track remote branch master from origin. Already on 'master' $ ls -l total 36 drwxrwxr-x 8 4096 May 11 2011 incubator drwxrwxr-x 14 4096 May 11 2011 obsolete drwxrwxr-x 14 4096 May 11 2011 releng drwxrwxr-x 5 4096 May 11 2011 runtime.rwt drwxrwxr-x 9 4096 May 11 2011 runtime.rwt.test drwxrwxr-x 18 4096 May 11 2011 runtime.ui drwxrwxr-x 7 4096 May 11 2011 runtime.ui.test drwxrwxr-x 14 4096 May 11 2011 sandbox drwxrwxr-x 14 4096 May 11 2011 tooling I can obviously clone, and checkout.... What do I need to do? (In reply to comment #15) In my case the problem is with fetching. I consistently get no new changes. In your case after cloning, does master point to a changeset that is recent, like earlier this week? (In reply to comment #16) > (In reply to comment #15) > > In my case the problem is with fetching. I consistently get no new changes. In > your case after cloning, does master point to a changeset that is recent, like > earlier this week? $ git log -1 commit 94f8e8de3693c90fa5d84815eafd6ec928c81782 Author: rsternber <rsternber> Date: Wed Mar 30 19:37:40 2011 +0000 formatted copyright Mirroring the other way around: Does any eclipse project walk the migration path with http://www.kernel.org/pub/software/scm/git/docs/git-cvsserver.html as Austin proposed here https://bugs.eclipse.org/bugs/show_bug.cgi?id=331780 ? (In reply to comment #14) > (In reply to comment #13) > > It is unfortunate that it has come to this. > Sorry, didn't mean to seem so dramatic. > If you have any suggestions on what we can do to make the Git mirrors of CVS > reach a level approaching 'reliable' without rebuilding them from scratch every > night, then by all means, suggest away. git-cvsimport is functional at best. I definitely think that rebuilding the mirrors every night is not a solution. I believe rebuilding the mirror causes downstream consumers to have to rebase or even reclone their local repos to get in sync. The behavior I have been experiencing seems to be sporadic. For instance, I went for a long time without receiving any changes, and then when RAP 1.4M6 was tagged, it came down from the mirror. But 1.4M7 has yet to do so. I am not familiar with how the mirrors are synced. Denis, could you fill me in? No worries, Austin. I understand your frustration.
I had discontinued importing the RT repository because of problems I could not resolve with the import. Alas, I must have been distracted since I didn't delete the repos.
I'm running it now, and I'll report back the problem. In the meanwhile, this is the actual script I use:
#!/bin/bash
umask 0002
# exit; # droy -- this thing explodes after a while for whatever reason
BASEDIR=/(snip)
TARGETDIR=/(snip)
for i in $(ls -1 $BASEDIR | egrep "^org.eclipse." | awk -F. {'print $1 "." $2 "." $3'} | sed -e 's/-.*//' | sort | uniq); do
echo "Processing $i"
if [ ! -d $TARGETDIR/$i ]; then
mkdir $TARGETDIR/$i
fi
# Iterate all directories in /cvsroot/rt that start with $i
for j in $(ls -1 $BASEDIR | egrep "^$i"); do
echo "Importing $j into $i";
export GIT_DIR=$j.git
git-cvsimport -v -i -p -x -o master -d :local:/cvsroot/rt $j -C $TARGETDIR/$i
done
done
The import for RT seems to have run successfully. A git clone / git show reveals this: $ git show commit f8c22210852ab8576807074d9e10356efed2eb88 Author: rsternber <rsternber> Date: Wed May 11 20:48:34 2011 +0000 It looks like it's in good shape -- for now :/ (In reply to comment #20) > The import for RT seems to have run successfully. A git clone / git show > reveals this: > $ git show > commit f8c22210852ab8576807074d9e10356efed2eb88 > Author: rsternber <rsternber> > Date: Wed May 11 20:48:34 2011 +0000 > > > It looks like it's in good shape -- for now :/ Thanks a lot. The git mirror now contains the 1.4 M7 tag(s). Ok, let's dare to close this as fixed. Please reopen if you see it get skewed again. The Git sync runs daily at 13:00 Eastern time and takes several hours to complete all the CVS projects. OK, thanks Denis. > Ok, let's dare to close this as fixed. Please reopen if you see it get skewed
> again.
Actually setting this to "fixed". Regardless, CVS is deprecated at Eclipse, so these cloned repos will be going away soon.
|