Community technical support mailing list was retired 2010 and replaced with a professional technical support team. For assistance please contact: Pre-sales Technical support via email to sales@march-hare.com.
Hi All,
I have a problem with being unable to setup POSTCOMMIT to work as desired.
I already searched documentation and entire web site & mailing list archive, and
have read the following article, and a number of mailing list posts:
http://www.cvsnt.org/wiki/PostCommit
Here is my setup:
- CVS is used for web development.
- CVS server is CVSNT on a Linux server (same machine is web server).
- Clients are using TortoiseCVS on Windows.
- Web server document root needs to be updated immediately when anything is
committed to CVS.
- Here is how everything is setup on Linux server:
- There are multiple repositories setup - for multiple projects.
- All repositories are setup under one system user ("cvsadmin").
- Each project is setup as a different system user (i.e. "project1",
"project2", etc).
- Programmers have FTP and shell access to all projects - this is required for
various purposes.
Here is how it works now when changes are committed:
1) Changes are committed to CVS.
2) Programmer logs in via SSH as a project's system user, and runs a shell
script to update repository:
------------------------------
#!/bin/bash
rm -rf /project1/tmpdir/*
cd /project1
/usr/local/bin/cvs -d :pserver:export_user:PASSWORD at myserver.com/project1 \
export -f -r HEAD -d tmpdir .
cd tmpdir
cp -R * ../wwwroot/
------------------------------
It works great, but we need to automate it.
What I need, is to set it up so when anything is committed in a repository, web
server document root would automatically get updated too. I guess, my problems
are because repository and project are different system users.
I understand that if I setup repository as same system user as project, I can
setup a shell script (similar to above) to be executed on POSTCOMMIT. However,
there are reasons to not do it - in particular, because CVS users have shell
access, and will be able to access CVSROOT directly - I don't want that.
As a solution, I tried setting up the following script to POSTCOMMIT for a
repository:
-------------------------------
#!/usr/bin/expect
# run "cvs_update" shell script - similar to shell script above
spawn su -s /bin/bash project1 -c "cd /project1; ./cvs_update"
sleep 2
expect phrase:
send "PASSWORD\r"
sleep 2
-------------------------------
However, this seems to be an ugly way, and it does not work in TortoiseCVS for
some reason... When I do a "commit" via TortoiseCVS, and "su"'s password prompt
appears in TortoiseCVS log window, it prints "ok" and just hangs forever!
Like this:
-------------------------------
......
Password: ok
-------------------------------
I guess, it is a bug of TortoiseCVS, since it works great if I commit to
repository from a standard Linux CVS client (located on same Linux server).
I don't like this method much anyway - as system user's password is disclosed in
POSTCOMMIT script. Seems to be insecure.
Can anyone suggest anything better?
THANK YOU!
Alexander.