I'm trying to get a git post-receive hook working on Windows.
I'm using Git 1.7.9 (Msysgit) and have a repo locally and a bare repo on a remote server. I can fetch, commit, push etc. I've set up a post-receive hook that should checkout the files into a working folder (part of the deployment process) but it doesn't seem to work.
Here's what I do:
Change a file, stage and commit it
Push to remote server - successfully
Expect to see the echo - don't see the echo
Check working folder on server - latest files are not there
Log onto the server and run the hook script manually - latest files are checkout out into the working folder.
I changed the hook so it does nothing except echo a message and I've read that I should see this in my console after pushing. But this is not being displayed so I can only assume the hook is not being fired off.
I'm pushing over HTTP with git dot aspx on the server handling the request and pusing via the gui locally. After that failed I tried Bonobo and the hook doesn't work when pushing via the gui or a bash console.
I'm assuming someone has this working somewhere but after two days of searching all I have found are solutions that don't help or people with the same problem that has gone unanswered.
(I'm a git newbie btw).
Cheers.
Update
I'm starting to think it may be to do with permissions - but Unix permissions, rather than NTFS. When #eis mentioned permissions I had assumed NTFS. But after more digging it seems that Git on Windows still checks for Unix file perms.
So I suspect the issue is that the post-receive file is not executable as when I do a ls -o it's -rw-r--r-- (644 I believe). If I try and change this through bash and chmod 777 post-receive then do ls -o the permissions are the same.
The strange this is that as soon as I edited post-receive (with notepad++) the execute bit gets removed. (my test script that ends in .bat does retain its execute bits though...)
BTW, the user I'm logged on as is the owner of the files (according to ls -o) and yet I can't set the permissions.
Starting to get really confused now. Am I missing something really obvious?
Update 2
Neither chmod 777 post-receive nor chmod a+x post-receive work. I took a new, clean post-receive file, uploaded it the to the server and checked the permissions and it had execute. If I rename the file (to remove sample) in Windows then execute is removed. If I do it in bash with mv execute is retained.
But, whenever I edit the file (in Windows or in bash with vi) then execute gets removed.
So, the problem now is why does it remove the execute bits when I edit the file?
Hopefully this is the final hurdle and the cause of it not executing...
You are going to have to patch git to make this work. The checks in builtin/receive-pack.c are for access(path, X_OK). In msysgit this diverts to mingw_access which throws away the X_OK bit as it is simple not supported on Windows.
On windows, we have no flag to specify a file is executable. Systems often do some emulation of this. For instance, tcl will look for any extension in the PATHEXT environment variable to decide that a file is executable. We can't do that here as the hook names are hardcoded without any extensions.
Instead, I suggest changing the access test to just check the file exists and then call execv on the path. The mingw version of this (in compat/mingw.c) looks for script files and will read the shbang line and launch an appropriate interpreter (sh, perl etc). So modifying builtin/receive-pack.c:run_update_hook should let this work for you. Currently the hook running uses start_command and I think that should call down to execv for you.
In short, change the access test and it will probably work.
When using msysgit on the server and pushing via a file share hooks work without problem now. Maybe this was fixed in mysysgit since the answer was written. I didn't look into it.
I also noticed that the original question stated git dot aspx and Bonobo were being used which use GitSharp.dll. This would mean the application is not shelling out to the git.exe and hooks would not be handled the same way.
For example, the GitSharp.dll used in git dot aspx has it's own hook post-receive hook implementation which could be performed in C#:
public void Receive(Stream inputStream, Stream outputStream)
{
using (var repository = GetRepository())
{
var pack = new ReceivePack(repository);
pack.setBiDirectionalPipe(false);
//setup post receive hook here
pack.setPostReceiveHook(new PostRecieveHook());
pack.receive(inputStream, outputStream, outputStream);
}
}
public class PostRecieveHook : IPostReceiveHook
{
public void OnPostReceive(ReceivePack rp, ICollection<ReceiveCommand> commands)
{
//Do PostRecieve Hook Work Here
}
}
I hope to help others with confusion between libraries that are implementations of Git and applications that call out to the actual git.exe.
Related
I have just made a new repository. My server is Ubuntu 12.04 32 bit.
I want my commits to be live as soon as I have committed them. This is achieved by making a working copy in my public html directory, and having the post-commit hook update that working copy. I have set that up using the instructions here:
http://www.frenssen.be/content/using-subversion-automatically-update-live-website
The issue is that my post-commit will not run when my project commits. My project commits just fine. I know that the script is a valid script too, because I can run it with
env ./post-commit
inside of the hooks directory and it properly updates.
I thought it might be a permissions thing so I have made extensive use of chown to ensure that www-data is the owner of:
the repo /home/svn/repo2
the working copy /var/www/html
the hook /home/svn/repo2/hooks/post-commit
the update binary /home/svn/autoupdate/autoupdate
But still nothing. My commits are successful, but the hook never runs. My hook script looks like this:
#!/bin/sh
/home/svn/autoupdate/autoupdate
All of the googling seems to be pointing to a permissions issue but I can't figure this one out.
Can you repeat this guy's example, and see if it works at all?
mikewest.org/2006/06/subversion-post-commit-hooks-101
Do you get an error when you commit?
If your post-commit hook actually failed, I think you would see evidence of that in whatever client you were using to perform the checkin. I'm almost suspecting that post-commit is not firing at all. Just to double check, the file needs to actually be called post-commit, all lowercase, and not post-commit.tmpl.
I have created jekyll site. Regarding the deployment I don't want to host on github pages. To host private domain I came know from documentation to copy the all files from _site folder. That's all wicked.
Question:
Each time I add new blog post, I am running cmd>jekyll build then I am copying newly created html to hosted domain. Is there any easy way to update without compiling each time ?
The reason, Why I am asking is because it will updated by non technical person
Thanks for the help!!
If you don't want to use GitHub Pages, AFAIK there's no other way than to compile your site each time you make a change.
But of course you can script/automate as much as possible.
That's what I do with my own blog as well. I'm hosting it on my own webspace instead of GitHub Pages, so I need to do these steps for each update:
Compile on local machine
Upload via FTP
I can do this with a single click (okay, a single double-click).
Note: I'm on Windows, so the following solution is for Windows.
But if you're using Linux/MacOS/whatever, of course you can use the tools given there to build something similar.
I'm using a batch file (the Windows equivalent to a shell script) to compile my site and then call WinSCP, a free command-line FTP client.
WinSCP allows me to store session configurations, so I saved the connection to my server there once.
Because of this, I didn't want to commit WinSCP to my (public) repository, so my script expects WinSCP in the parent folder.
The batch file looks like this:
call jekyll build
echo If the build succeeded, press RETURN to upload!
pause
set uploadpath=%~dp0\_site
%~dp0\..\winscp.com /script=build-upload.txt /xmllog=build-upload.log
pause
The first parameter in the WinSCP call (/script=build-upload.txt) specifies the script file which contains the actual WinSCP commands
This is in the script file:
option batch abort
option confirm off
open blog
synchronize remote -delete "%uploadpath%"
close
exit
Some explanations:
%~dp0 (in the batch file) is the folder where the current batch file is
The set uploadpath=... line (in the batch file) saves the complete path to the generated site into an environment variable
The open blog line (in the script file) opens a connection to the pre-saved session configuration (which I named blog)
The synchronize remote ... line (in the script file) uses the synchronize command to sync from the local folder (saved in %uploadpath%, the environment variable from step 2) to the server.
IMO this solution is suitable for non-technical persons as well.
If the technical person in your case doesn't know how to use source control, you could even script committing & pushing, too.
There are a number of options available which are mentioned in the documentation: http://jekyllrb.com/docs/deployment-methods/
If you are using Git, I would recommend the Git Post-Receive Hook approach. It simply builds the site after the new code is received:
GIT_REPO=$HOME/myrepo.git
TMP_GIT_CLONE=$HOME/tmp/myrepo
PUBLIC_WWW=/var/www/myrepo
git clone $GIT_REPO $TMP_GIT_CLONE
jekyll build -s $TMP_GIT_CLONE -d $PUBLIC_WWW
rm -Rf $TMP_GIT_CLONE
exit
Since you mentioned that it will be updated by a non-technical person, you might try something like rack-jekyll to automatically rebuild when new files are FTP'd.
I'm using CollabNet Subversion Edge on Windows 2008 … and trying to auto-deploy (so update from repo to folder) when any commits are made by developers using Tortoise SVN.
I've got a post-commit hook file in the correct repo /hooks folder. The file is named post-commit.bat
The file has one line -
"C:\Program Files\TortoiseSVN\bin\tortoiseproc.exe" /command:update /path:"c:\wamp\www\thewebsite*" /closeonend:1 /outfile:"c:\csvn\update-logs\thewebsite-out.txt"
When I commit anything, it's timing out if I have the file present. If the file is not present, the commits work without any problem. So that tells me the post-commit file is being called … and it's got a problem!
Anyone got a sample post-commit Windows batch file that can help me? Or know how to solve my particular problem?
You should try testing your script by simply calling it from the command line and passing in the repo and version parameters. This might give you some more insight as to why it is timing out. For example:
script.bat PATH_TO_REPO REPO_VERSION
Also instead of using Tortoise, use the native SVN client library. In your script you can navigate to the folder you want to update, and call the "svn update" command. This will be more straightforward and not have to go through Tortoise just to make the update command back to the native library.
Try something like this:
cd "c:\wamp\www\thewebsite"
svn update
If your SVN server requires permissions you may need to pass these in your script as well.
I am currently setting up a git schema as described here.
The server in question is running Windows Server 2008 R2, using copSSH and msysGit.
The jist of it is a pair of git repos: one bare repo that is pushed to and then another non-bare repo that contains a live application.
The repo that is pushed to contains a post-update hook that looks as follows:
echo
echo "**** Pushing data to Live Server."
echo
cd /cygdrive/c/Repositories/Live || exit
unset GIT_DIR
git pull hub master
exec git-update-server-info
Of note: Live is the non-bare repo which is, as the name implies, live. Live.git is the bare repo which contains the above hook and is pushed to. I was unable to cd to the repo without /cygdrive/, which was uncovered by echoing $pwd in a diagnostic push.
Pull is successfully called in the Live repo, the problem comes in locating the Live.git repo in that update. I get "odd ball" errors that read like:
remote: Pushing data to Live Server.ommand not foundnd not foundle or directory
remote: : not foundpdate-server-infoand not foundand not founddommand not found
hub is defined as a remote repository via the local git shell as both /c/Repositories/Live.git and /cygdrive/c/Repositories/Live.git.
I've replaced "hub" with direct folder paths, even trying things like "../Live.git"
Everything I have tried has ended in a git update failure. All of the attempted paths for Live.git in the update command were attempted via a git bash with the working directory set to the Live repo and worked.
Thank you for any leads or help!
An update:
My schema requires that I pull from the hub repository, it would take an unnecessary amount of room to explain why, but such is the case.
I have done more testing and found more out about my issue:
When I push via SSH to my repository, the prime hub is successfully mounted and the pull command is executed. When said pull command tries to resolve the hub repository it fails (as I detailed above), but it turns out with a series of ls commands in the hook that no folders can be seen. I even ran an ls on /cygdrive/c/ and can only see a folder called usr. I have ensured that the copSSH user has full administrative privileges on the entire file system (only for testing and debugging purposes).
I do not know why my paths have to use the /cygdrive/ path, I do not have cygwin installed besides the cygwin packaged in copSSH. Is this indicative of a problem in my configuration? or is this normal for copSSH + msysGit setups? How is it possible that I can mount the prime repository from the hook, even if ls will not list the repositories folder?
You shouldn't need /cygdrive in paths with msysgit, unless you also have cygwin installed (in which case the combination of msysgit mingw and cygwin might be the source of your problem)
You shouldn't try to run git pull from a repo hook. While some people will say that it works fine, it does not and the behavior can be unpredictable due to various factors.
Instead you should use a method such as this one. http://toroid.org/ams/git-website-howto
That method uses a post-receive hook in a bare repo that calls git checkout to update a directory.
Im trying to use git extensions and I really like it so far, but I don't manage to push to github. The following command works fine in git bash:
git push "origin" master:master
and then when I push with git extensions I get this:
C:\Program Files\Git\bin\git.exe push "origin" master:master
Permission denied (publickey).
fatal: The remote end hung up unexpectedly
Done
In the git settings it just has openSSH selected which I would like to keep because putty seems more hassle.
The remote addres I have added in gitextensions, actually the only thing I have done from the bash just to test is the command above and gerenating the SSH keys. All else is set up in git extensions. So I suppose everything must be set up correctly. Eg, the email adress and name are correct...
ssh git#github.com
also connects fine, from bash
edit:
So I can reproduce the error by putting the command from git-extensions in cmd. Appearantly there is a difference between running from git bash and from cmd and git-extensions runs this command like a windows command... any clues?
update:
If I choose git-bash from the menu in git-extensions I get a window that is exactly the same as when I right click in explorer on my repo folder and choose git bash here. Now, in the one opened from explorer I can push and in the one opened from extensions I get the public key problem.
Things to check for:
Presence of HOME environment variable.
Presence of %HOME%\.ssh\ and RSA keys there.
When you run git from command promt it is preferred to run git.cmd, because it fixes HOME automatically:
#if not exist "%HOME%" #set HOME=%HOMEDRIVE%%HOMEPATH%
#if not exist "%HOME%" #set HOME=%USERPROFILE%
To fix gitextensions, define HOME variable and set it to same value as USERPROFILE
Ok, I solved it.
I opened the git bash from git extensions and a git bash using the explorer shell extensions. I then ran ssh -vvv git#github.com from both to see the difference.
It turned out that because I changed the location of the git config file ( I tend to reformat every now and then, so c:\documents and settings... is notoriously bad for storing anything I might want to keep persistent), when run from git extensions git was also looking for the ssh keys in a different location. So I added the .ssh folder where my git cofig file is and now it works fine. I think git-extensions sets the HOME variable for git to whatever you set as path for the config file.
Now, that was not obvious...
The correct answer is the one from 'max' (set your HOME env var manually), but it may help some to understand why the problem is happening (as Git gets more users around the world it's going to be very common).
Cygwin sets $HOME to /home/yourname, but that variable is not known in the Windows environment. So if you open a bash window and do env | grep HOME you'll see all three 'HOME' variables mentioned here, and you might wonder why Gitextensions doesn't use your proper cygwin HOME - which is because its .bat file invocation doesn't see it - it only sees what you see from doing 'set' in a windows console.
It's mystifying why it doesn't do this evaluation later and get the proper cygwin path since it knows how to invoke bash, but (at least in versions up to 2.41) you have to do this manual change in the settings or in .gitconfig.
When you fire up git bash directly you'll land in your home folder for MSYS. You need to make sure you have your ssh key (id_rsa ?) in the .ssh subfolder (relative to the home folder).
Glad you solved the problem. Since this sounds like a serious problem I'm interested in the difference between git-bash when started from GitExtensions. In GitExtensions there is a setting that might fix this. The %HOME% directory can be changed in GitExtensios. By default it will be set to %HOMEDRIVE%%HOMEPATH%, but you can override this. Changing this probably solves your problem, since you suggest this is the problem. To change this open the settings dialog and go to the tab "git". In the section "Environment" you can set the %HOME% path.
I will appreciate it if you let me know if this also solves the problem. I'm also interested in what caused this in the first place. Maybe I can improve the check for a valid HOME directory.
Yes, not setting the HOME varible was the issue for me too.
Set the HOME variable as %USERPROFILE% and regenerate the Private and Public keys, then try cloning—it should work now.
I see that the HOME variable is set to USERPROFILE in git extensions. I can add and commit but not push/pull from the remote repository. I am unable to connect to any remote repository. We use http (not ssh) for the connection. However, using git bash I am able to connect to the remote and do push/pull. I had put my issue GIT extensions does not connect to remote but git bash does and was directed to this issue. However, I am unable to resolve my issue. Any ideas?