I have a problem with git.
Basically, here is what I have. I access a svn repository through git. Until now, on python files, everything worked fine.
But lately I also added some pyd, dll and lib files on the repository. THe first update went well. But then, these files have been modified and since then I can't update. These files were added from a windows computer with TortoiseSvn on the svn repository.
If I do a git svn rebase on linux, everything works fine.
If I do a git svn rebase on windows with msysgit (and also tortoisegit), I have the following error : fatal: write error: Invalid argument
If I do a git svn rebase on windows with cygwin, I have the following error : didn't find newline after blob at /usr/lib/perl5/vendor_perl/5.10/Git.pm line 916
I tried several stuff (autocrlf true/false, safecrlf true/false), adding .gitattributes file with the following line *.* -crlf -diff -merge and nothing worked.
I'm a little stuck here so any suggestion would be welcome.
Thanks in advance.
Had identical issue with Msysgit v1.7.2.3, the latest version as at 29 Sep 10, and wanted to share my findings here (Google turns up several cases, but no solutions).
Trying to do "git svn rebase" on a repo (that has this has worked on plenty of times in the past) consistently failed with a "fatal: write error: Invalid argument" after a certain number of commits. The sync would then revert to the beginning again.
I believe this is a bug in Msysgit relating to large(ish) binaries and available memory (on a Win XP SP3 system with 4GB RAM and plenty free HD space). The remote system was the DotNetNuke SVN repo on CodePlex (https://dotnetnuke.svn.codeplex.com/svn).
Initially it was choking on a 330KB "CHM" file (~212th commit, r52261). It consistently did so, even after disabling Avast AV, Google Desktop, etc and verifying that there were no other processes with locks on the repo folder. After a reboot (but opening Outlook, Dreamweaver, etc), it then was consistently and repeatedly failing on a ~15.3MB DLL (~416th commit, same revision).
Finally, after another reboot, disabling Avast, Carbonite and Google Desktop and running no other programs, the sync worked first time.
This seems to point firmly to my conclusion that it was an available memory issue, probably linked to the presence of a largish binary and large number of commits in the revision. Note that I also tried "git fsck", "git svn reset xx" and tweaking the "packSizeLimit" / "usedeltabaseoffset" config vars, without success.
I've found that the best policy for using Git on windows is to tell it to not do anything about line endings.
I don't know if that will help you recover your current git repo, but it's worth a shot.
I set:
[core]
autocrlf = false
Related
On my Windows machine, I have a repository that works nicely with msysgit. But when I try running git status on cygwin, it just hangs, without printing anything until I press Ctrl-C.
How can I investigate this to find out the reason for such behaviour?
The reason for this was the difference in core.autocrlf setting. When install git for windows, I set it to automatically convert line endings in both directions, so the files in the filesystem were actually different from the files saved in git history - but windows git always hid these differences.
Meanwhile, cygwin git was working extremely slow because it detected differences in every line of thousands of text files in my repository.
I'm working in a large corporate enterprise where all the developer machines are Windows 7 Enterprise SP1.
I'm running major migration project from RTC to git (resulting in something like 1200 git repos). There are no Windows servers in the production environment, build or test environments - everything is either Solaris or RedHat. The solution will be rolled out to ~200 developers.
There is generally a push to use more Unix-command line tools and there are a few different alternatives on Windows such as: Cygwin, git-bash, cmder. I have avoided running a full linux VM because that introduces too many other problems (most developers don't have local admin rights and the internet proxy is a constant hassle so I don't want to make networking a bigger problem with NAT).
I've been running Cygwin (mintty 2.6.2) for the past 8 months and it's been ok until today where I hit a very concerning issue with git (v2.8.3) where git status reported a clean working directory even though multiple files and folders had been deleted in the repo. Only after I recreated a folder with the same name as one of the deleted folders did all the deleted files appear correctly with git status.
I'll explain the symptoms and what I was doing but so far I have not reproduced the issue. My suspicion is that the problem lays somewhere between the emulated linux file system and the actual windows file system. Is there any difference in the way the various emulators achieve this?
The specific problem I hit had these symptoms:
Client side:
git status showed a clean working directory
git log showed 3 commits A, B and C (commit C checked out)
The repo contents was one folder containing a file, 2 more files in the root folder, plus the .git folder with contents
git stash list was empty
git branch -a reports only master, HEAD and origin/master
Server side:
The origin contained 13 additional folders, each with one file
The origin also contained commits A, B and C
Only master branch is present
Commit A was the initial empty commit.
Commit B was where all contents was added, 14 folders and 16 files.
Commit C was another empty commit
Commit A and Commit C were both created using tools to assist with the migration. It runs two commands: "git init" and "git commit --allow-empty -m 'initial commmit'"
I could not understand how git status did not report the deleted files (I remember deleting them, but it was some days previously with multiple computer hibernates, and probably a restart or two from resuming my work)
Trying to figure out what had happened I did this:
I created a new file then ran "git add", "git commit" (creating D) and "git push."
Commit D appeared on the server with the new file. The 13 additional folders and files were still present on the server.
I ran "git pull" which returned "already up to date"
I checked the changes for every commit on the client side and on the server side and every diff was the same
I cloned a new copy of the repo and the contents matched the server with the 13 additional files and folders with the additional commit D and extra file
I then recreated a folder with the same name as one of the deleted folders and ran "git status" where finally all of the deleted files and folders were reported correctly.
I cannot explain this any other way except for a serious bug which simply makes it unsafe to use git in Cygwin. I hope the community may have some advice for me in this area and that this is phrased as a clear enough question that the mods don't flag my post.
I will do my best to try and reproduce the problem and update the issue with more info when I have some.
Edit: Update 2016-12-08
My attempts to reproduce the error have been unsuccessful. If I see it again during my work I will update this issue.
I've never seen such behaviour with Git on Cygwin. Actually currently I use Git for Windows from within Cygwin and it also works fine. I used to use the Cygwin Git, but I had the feeling that it is slower than Git for Windows, such I switched to Git for Windows used from within Cygwin and it works great.
If you at some point update the Windows boxes to Windows 10, you can also consider another option, the Windows Subsytem for Linux which is an Ubuntu based virutal Linux environment developed by MS together with Canonical. It is still in the process of getting mature and not fully usable yet in my opinion, but there you then have a natively supported virutal Linux environment where you can use apt-get and so on.
Issue you mentioned are really unexpected, and I doubt Cygwin can cause it. But you have following options
git comes with git bash, which support all major unix commands and it looks completely like unix shell. I am using git version 2.9 on windows 10 and heavily use major unix command like grep, sed & find , and they all work excellent. Even it support vi but I don't use it
git comes with git CMD, and same git commands will work on windows command prompt as they do on unix. You shouldn't need a separate emulator with this.
Though you mention you are using Windows 7, but now Windows 10 comes with native support for Unix Bash
You can use gnuwin32 but I doubt it will be better than Cygwin
I have a Mercurial repository that is located on a NAS (Buffalo TeraStation). It is mounted on a Ubuntu machine and is mapped onto a Windows 7 machine.
I have previously been able to clone, update, etc. from both the Linux machine and the Windows machine, but now am having problems. Too many changes have been made (updating versions of hg, thg, etc.) to identify a specific point where things started to fail.
I can read and write files from both machines from/to the NAS. I can use "cp" to copy entire repositories, but if I attempt to clone using hg (or thg) it fails.
When cloning an existing (in the NAS) repository to the Ubuntu machine, all of the files copy over, but the operation is aborted with the message:
abort: Operation not permitted: (repository_path)/.hg/store/.phaseroots-94sdvj
[command returned code 255 Thu Jul 30 17:39:45 2015
When cloning an exiting (in the NAS) repository to the Windows machine, the cloning hangs (with no files transferred).
I have tried various [trusted] settings to no avail.
I have tried performing the hg clone command manually and it works properly when done with "sudo". However, the cloned repository then is owned by "root".
I'm pretty sure this is a permissions problem, but have run out of ideas. Any guidance would be appreciated!
One thing can think of is support for hardlinks on the NAS you are using.
For example some of the versions of Mercurial had problems with windows shares. For more information here.
I have no explanation for the problem on the Linux machine, but the problems on the Windows machine might be caused by this exact setup (cloning from a Linux drive to a Windows machine).
Years ago, I had similar problems (although with different error messages) when pushing from my Windows machine to repositories on a shared (Linux) NAS drive - see the link for more details.
But that was over five years ago and I didn't try it anymore since then, though.
As part of our code repository, we have a symlink which is internal to the working tree.
Zend -> ZendFramework1.10/library/Zend
This works fine for all the developers running Linux or OS X, but we're now getting some people trying to use the repository on Windows.
The functionality of the symlink can be replicated by deleting the link git creates, and using mklink to create the equivalent directory junction.
However, git now sees this as the deletion of the symbolic link, and the addition of a proper directory.
I'm looking for a way to have the two co-exist, is there a way to tell the Windows machines to ignore the Zend directory, even though it's technically versioned. If this breaks when the files in that directory change then so be it, but it'd be nice to be able to work with the rest of the repo without having to worry about the link.
You can use git-update-index to tell git to ignore changes to the file:
git update-index --assume-unchanged Zend
You could probably use cygwin on the machines running windows.
As Magnus Skog has suggested, git under cygwin copes correctly with the symlinks. I switched away from Git for Windows for this reason alone. However you need to weigh up this advantage against the overhead of setting up the cygwin environment for your Windows users (particularly for those unfamiliar with *nix and the command line; for example there are a number of outstanding issues when trying to use Cygwin and TortoiseGit.)
I've been playing around with git at home and I really like the idea of local commits. Being able to commit smaller changes without spreading my potentially broken code to everyone is nice, as well as being able to revert smaller changes because of the more frequent commits.
I've been reading about the git-svn command, but I'm not sure I entirely understand how it works. We work on Visual Studio 2008 projects, and run VisualSVN which handles file renames, moves, and all that for us from within the IDE.
What I want to know is: Is it possible for me to commit to a local git repository but also commit to the remote SVN repository as well? I'd like to keep VisualSVN change tracking and committing from within the IDE, but also be able to use git to temporarily store changes. Are they likely to get in each other's way?
It works beautifully. go for it. Just don't check your .git folder into svn.
edit: erm, when I do it though, I don't bother with git-svn. I just treat the local working directory for svn as any other directory, and I tend not to care much about the previous SVN history.
I've used git-svn to keep "updating" from the remote repository, but haven't used it to commit to an svn repository, so I can't help you about that part.
What you do is simple, with all settings on default:
>git svn init <url......>
>git svn fetch
When you do that, it fetches it to a "remote" branch called "git-svn".
To merge it with your current branch:
>git merge git-svn
You may run into some issues if you're using git-svn after the fact. What I mean by that is: you already have checkout the project using svn, then you also created a git repository in your local working svn directory. Then you use git-svn on top of that.
Two issues I had to deal with:
Line endings. svn might convert line endings to windows, while git-svn will preserve them to unix style. So you might get tons of conflicts due to the line ending differences.
So to be sure, use a tool to convert line endings on all files to unix (or windows, depending on what line ending is used in the svn repo).
svn keyword expansion. e.g. $Id$
git-svn will not expand these keywords, while svn will. So you'll have conflicts there as well. Again, use a tool (or write a script) that converts all instances of $Id .......crap.....$ to just $Id$`