#pragma section-numbers 2 <> <> <> = Publishing Repositories with hgwebdir.cgi = {{{#!wiki tip The multiple repository CGI server is now described [[PublishingRepositories#multiple|in the Publishing Repositories document]] together with other related information. However, this guide gives some additional useful details for publishing repositories. }}} == Introduction == {{{#!wiki tip Starting with version 1.6 of Mercurial, the ''`hgwebdir.cgi`'' script has been unified with the hgweb.cgi script. Wherever ''`hgwebdir.cgi`'' is referred to in these directions, you can substitute the ''`hgweb.cgi`'' script instead. }}} This page explains how to make a bunch of repositories accessible through CGI using the {{{hgwebdir.cgi}}}-script and a webserver (apache, lighttpd or nginx). Once the script is set up, it is very easy to open new repositories. To quickly serve a single repository, please have a look at [[CGIinstall]]. <> == Pre-requisites == The installed software is: * A webserver that can run CGI scripts such as apache2 (apache2, apache2-common, apache2-mpm-prefork, apache2-utils), lighttpd or nginx (newer than 0.8.48) * But it is not necessary (see below) * [[http://www.aprelium.com/abyssws/|The free version of Abyss Web Server works well under windows]] * some version of mercurial (mine was 0.9 taken from Debian/unstable) * Python: * Ubuntu/Edgy comes with python 2.4 * you will need the python2.4-dev package as well * sudo (in general, I prefer sudo to su) * hgwebdir.cgi (use the one coming with your Mercurial package or download from http://www.mercurial-scm.org/repo/hg-stable/raw-file/tip/hgwebdir.cgi) == Getting proper Mercurial == If you are on Linux, you probably want to install the Mercurial version packaged by your distribution. Like for RPM-based distributions: {{{ $ cd download-directory $ wget http://download.opensuse.org/repositories/devel:/tools:/scm/openSUSE_10.2/x86_64/mercurial-1.0-1.2.x86_64.rpm $ sudo rpm -ihv mercurial-1.0-1.2.x86_64.rpm }}} If you can only get a pre-1.0 version from your distribution, you should probably acquire another version somehow. One way would be to get the latest version from the stable branch: {{{ $ cd working-directory $ hg clone http://mercurial-scm.org/repo/hg-stable/ $ cd hg-stable $ python setup.py build $ sudo python setup.py install }}} /!\ Be sure to check for later versions, these filenames and versions might be out of date. == RHEL4 Specific Instructions == [[RHEL4HgWebDirSetup]] == OpenBSD 4.6 Specific Instructions for chrooted Apache 1.3.29 == OpenBsdWebDirSetup == Windows Specific Instructions == On Windows, your Python version ''must'' match the version used to compile Mercurial. Otherwise, you'll get "Invalid Magic Number" errors when trying to run the CGI. At least one installer for 0.9.5 uses Python 2.4. (I can't verify all of them.) The pre-compiled Windows binaries for Mercurial 1.0.x, 1.1.x, 1.2.x and 1.3.x were compiled with Python 2.5. If you do wish to use Python 2.6, you must build your own Mercurial WindowsInstall. The first line of `hgwebdir.cgi` must point to your Python executable. Ex: {{{#!python #!c:/dev/Python25/python.exe # # An example CGI script to export multiple hgweb repos, edit as necessary ... }}} Some users report for the Mercurial 1.1 installation, Python needs to have the pywin32 package installed. It can be downloaded from: http://sourceforge.net/project/platformdownload.php?group_id=78018 If this package is not installed you will see an error like: "!ImportError: No module named pywintypes". (I didn't need it for Python 1.1.1 though...) === Mercurial Lib Files === Install the Mercurial libraries. The Windows binary puts a .zip file of the libraries in your Mercurial install directory, which I refer to as `MERCURIAL_HOME`. Unzip `MERCURIAL_HOME\library.zip` to a path of your choosing. (e.g. `MERCURIAL_HOME\lib`; for example: `c:\dev\Mercurial\lib`.). Winzip and this file may not play nice together so try using a command line unzip utility like: http://stahlforce.com/dev/index.php?tool=zipunzip Specify the location of the library files in `hgwebdir.cgi` {{{#!python # adjust python path if not a system-wide install: import sys sys.path.insert(0, "c:/dev/Mercurial/lib") }}} If you compiled your own version of Mercurial, you do ''not'' need the `sys.path.insert` line. If you continue to see import errors even after adjusting `sys.path`, you may need to turn on Python's optimize mode so it can import `.pyo` files. You can do this by enabling mod_env and using the following setting in Apache: {{{ SetEnv PYTHONOPTIMIZE x }}} === Adjust Template Files === The CGI assumes templates are located inside of the `lib` directory you just created. (Although you're supposed to be able to specify `templates = c:/dev/Mercurial/lib` in `Mercurial.ini` or `hgweb.config`, neither option worked for me.) Move the `Templates` folder from my `MERCURIAL_HOME\Templates` to `MERCURIAL_HOME\lib\Templates`. === Other Gotchas === The path to your Mercurial repositories cannot also contain the name to the path to the Mercurial CGI. Maybe this is just my weird Apache setup, but when my `hgwebdir.cgi` was installed in `c:/webdata/repos`, this `hgweb.config` failed for me: {{{ [paths] repo1 = c:/webdata/repos.merc/images repo2 = c:/webdata/repos.merc/web/htdocs/ }}} and instead I had to change my directory names to {{{ [paths] repo1 = c:/webdata/merc.repos/images repo2 = c:/webdata/merc.repos/web/htdocs/ }}} I don't know if this is a Mercurial/Python issue or an Apache issue. Either way, hopefully this saves you some time. === Collections of Repositories === The advised way of specifying collection is now the [paths] section. This has been introduced in Mercurial 1.1 For old versions see the next section. ==== Current version ==== {{{ [paths] /trunk = /webdata/hg_repos/trunk/** }}} '''Warning''': Browsing the available repositories '''will be very slow''', as all the files and subdirectories are scanned every time. To avoid scanning more than one subdirectory, use one asterisk (*) instead of two (**). This causes all repositories under `c:/webdata/hg_repos/trunk` to show up (whereas a simple `*` does not look for nested ones). Multiple collections may be specified, however each must have a unique prefix (part before the =). The prefix name may be the root - just the '/'. This will search inside every directory of the repository, this might be extremely slow if executed on a repository containing a working directory. The advice here is to run [[http://mercurial-scm.org/wiki/Update|`hg update null`]] on the served repository to avoid any superfluous recursive search. ==== Mercurial 1.0 ==== There is a bug with `[collections]` not working with Windows drive letters - Mercurial misinterprets the colon as a variable separator. See [[http://www.selenic.com/mercurial/bts/issue852|this bug tracker issue]] for details. This bug was fixed in Mercurial 1.1.x (see next section). Repository collections still work ''if'' they are on the same drive as your CGI though. Use: {{{ [collections] \webdata\hg_repos\trunk = /webdata/hg_repos/trunk }}} The backslashes and forward slashes make a difference. This got all of my repositories to show up that were located in `c:/webdata/hg_repos/trunk`. (Repositories located in lower subdirectories will show up ''but'' won't work. You have to add another line for their subdirectories.) == Directory Structure == Create the necessary directories: {{{ $ sudo mkdir -p /var/hg/repos $ sudo chown -R www-data:www-data /var/hg }}} It's usually a good idea to keep special directories out of the tree served by apache, but for security reasons on openSUSE the cgi scripts only work within the document root. So for openSUSE, which uses user/group wwwrun/www instead of www-data/www-data and does not allow write access to cgi directories for anyone but the apache user it is {{{ $ sudo mkdir -p /srv/www/htdocs/hg/repos $ sudo chown -R wwwrun:www /srv/www/htdocs/hg/repos $ sudo chmod 755 /srv/www/htdocs/hg }}} == Preparing the config == {{{ $ cat > /tmp/hgweb.config [collections] repos/ = repos/ ^D $ sudo -u www-data cp /tmp/hgweb.config /var/hg $ rm /tmp/hgweb.config }}} For openSUSE, replace the sudo line above with {{{ $ sudo -u wwwrun cp /tmp/hgweb.config /srv/www/htdocs/hg }}} To get a look like on http://hg.kublai.com/ with subdirectories? Just create subdirectories with hg repositories in it. To get a look of the top level like on kublai, add the following to hgweb.config: {{{ [web] style = gitweb }}} This also works in repository hgrc's, which are mentioned below. == Two possibilities == You can either use a separate web server such as Apache, lighttpd or nginx, or use the web server built into hg. === Using Apache, lighttpd or nginx with the hgwebdir.cgi script === ==== Putting the right stuff in place ==== Put the script in place (remember, we are still in that `working-directory/hg` :)): {{{ $ sudo -u www-data cp hgwebdir.cgi /var/hg $ sudo -u www-data chmod +x /var/hg/hgwebdir.cgi }}} And again, some changes for openSUSE {{{ $ sudo -u wwwrun cp /usr/share/doc/packages/mercurial/hgwebdir.cgi /srv/www/htdocs/hg $ sudo -u wwwrun chmod +x /srv/www/htdocs/hg/hgwebdir.cgi }}} ==== Configuring apache for use with CGIs ==== Ok, now it's time for apache (see at the end of this paragraph for the openSUSE way of doing this). First of all, do not really change the config of apache directly: {{{ $ sudo mkdir /etc/apache2/hg }}} Create the config with the following contents (e.g by using `sudo vim /etc/apache2/hg/main.conf`): {{{ ScriptAliasMatch ^/hg(.*) /var/hg/hgwebdir.cgi$1 Options ExecCGI FollowSymLinks AllowOverride None }}} This config says that we are going to serve our repositories through '/hg/'. Now make it really available, by changing your favorite site in `/etc/apache2/sites-available`. For this experiment I used `/etc/apache2/sites-available/default`: {{{ ... Include /etc/apache2/hg/main.conf }}} Make sure that everything is OK: {{{ $ sudo apache2ctl configtest Syntax is OK }}} Restart your web server: {{{ $ sudo apache2ctl stop $ sudo apache2ctl start }}} Check if it works by directing your browser to `/hg/`. For openSUSE, just put this in /etc/apache2/conf.d/hg.conf: {{{ ScriptAliasMatch ^/hg(.*) /srv/www/htdocs/hg/hgwebdir.cgi$1 Options ExecCGI FollowSymLinks AllowOverride None }}} and run {{{ sudo rcapache2 reload }}} if this does not complain about config errors you should be done. ==== Configuring Apache with mod_wsgi ==== Using mod_wsgi is recommended over using mod_python. It's one of the faster and more efficient ways of serving hgweb(dir). You can use the hgwebdir.wsgi script (it lives where other Mercurial scripts live, and works with Mercurial 1.0 and later), which references a hgwebdir.conf file in the CONFIG variable: {{{ from mercurial.hgweb.hgweb_mod import hgweb from mercurial.hgweb.hgwebdir_mod import hgwebdir CONFIG = '/var/hg/hgweb.config' application = hgwebdir(CONFIG) }}} This is really an ordinary Python module, but it uses a wsgi extension to make it clear what its use is. You then need to add this to your httpd.conf (or the vhost config): {{{ WSGIScriptAlias / /var/hg/script/hgwebdir.wsgi Order deny,allow Allow from all }}} Here is an alternative configuration for a non rooted directory of the context path. {{{ WSGIScriptAliasMatch ^/hg(.*) /var/hg/script/hgwebdir.wsgi$1 Order deny,allow Allow from all }}} Since you're allowing some permissions to the directory the .wsgi script is in, you probably don't want to put the script in a directory that also contains your repositories. NOTE: the default hgwebdir.wsgi uses 'hgweb.config' as configuration path. This may not be what you expect. If you find yourself with a working but empty installation, try to set this to the full path where your hgweb.config is situated NOTE: I had an issue in Mercurial 1.6.4 with the path where hgweb.wsgi can find the file hgweb.conf. They were in the same directory but this declaration didn't work config='hgweb.conf' but this with the full path worked config='/var/www/root_apache/hgweb.conf'. ==== Configuring apache with mod_python ==== You can also use mod_python, though mod_wsgi is probably better. Here are three howto's: http://www.selenic.com/pipermail/mercurial/2007-May/013222.html http://blog.slucas.fr/en/oss/hgweb-mod_python http://www.aventinesolutions.nl/mediawiki/index.php/Quick_Tip:_Getting_Started_with_Mercurial (on Fedora 6) ==== Configuring lighttpd ==== You can either update the existing /etc/lighttpd/lighttpd.conf file, or create /etc/lighttpd.conf/hg.conf and include that file from lighttpd.conf. First, you need to check if mod_rewrite and mod_cgi are enabled in the config file, and add them to server.modules if they haven't already been added: {{{ server.modules += ( "mod_cgi" ) server.modules += ( "mod_rewrite" ) }}} Next, configure rewrite rules that map URLs to the hgwebdir.cgi script. With the following added to your config file, URLs starting with either hg or mercurial will map to hgwebdir.cgi: {{{ url.rewrite-once = ( "^/hg([/?].*)?$" => "/hgwebdir.cgi$1", "^/mercurial([/?].*)?$" => "/hgwebdir.cgi$1" ) }}} Then, configure a URL match that invokes hgwebdir.cgi: {{{ $HTTP["url"] =~ "^/hgwebdir.cgi([/?].*)?$" { server.document-root = "/var/hg/" cgi.assign = ( ".cgi" => "/usr/bin/python" ) } }}} Make sure that everything is OK: {{{ $ sudo lighttpd -t -f /etc/lighttpd/lighttpd.conf Syntax OK }}} Restart the web server: {{{ $ sudo /etc/init.d/lighttpd restart }}} Check if it works by directing your browser to `/hg/` or `/mercurial/`. === Configuring Lighttpd push support === To enable support for pushing to remote repositories, you are required to add an extra $HTTP check to your vhost. The command when pushing that requires authorization is 'unbundle', so what we do is check to see if it is within the URL: {{{ $HTTP["querystring"] =~ "cmd=unbundle" { auth.require = ( "" => ( "method" => "basic", "realm" => "Mercuial Repo", "require" => "valid-user" ) ) } }}} This example uses an Apache2 .htpasswd file. You can add the following variables to your lighttpd.conf: {{{ auth.backend = "htpasswd" # auth method auth.backend.htpasswd.userfile = "/path/to/file" # passwd file. Syntax: USERNAME:ENCRYPTED_PASS }}} An easy way to create this file: {{{ htpasswd -c /path/to/file USERNAME }}} You will be prompted for password input. Be sure to add mod_auth to the lighttpd modules: {{{ server.modules += ( "mod_auth" ) }}} === Configuring nginx === Here is a configuration for nginx (newer than 0.8.48, in older versions there was a bug where fastcgi_pass was not inherited inside the limit_except block) which uses fastcgi to talk to the hgwebdir. nginx does not spawn hgwebdir by itself, so you have to take care of that yourself (with for example spawn-fcgi). Once you have done that, you can use something similar to the following configuration. It assumes that you have Mercurial installed in /srv/mercurial/ for static files, probably this is not true for your installation. {{{ server { listen 80 default; server_name example.com; root /srv/www; location / { return 403; } location /hg/ { fastcgi_pass 127.0.0.1:9000; include fastcgi_hg; fastcgi_intercept_errors off; limit_except GET HEAD { auth_basic 'Example'; auth_basic_user_file /srv/hg.htpasswd; } } location /hg/static/ { rewrite /hg/static/(.*) /$1 break; root /srv/mercurial/mercurial/templates/static; expires 30d; } location ~ /\. { deny all; } } }}} This configuration is using TCP on port 9000 to talk to the fastcgi daemon. It requires authentication (for push). And it servers static files directly. fastcgi_hg file is as follows: {{{ include fastcgi_params; fastcgi_split_path_info ^(/hg)(.*)$; fastcgi_param SCRIPT_NAME $fastcgi_script_name; fastcgi_param PATH_INFO $fastcgi_path_info; fastcgi_param AUTH_USER $remote_user; fastcgi_param REMOTE_USER $remote_user; }}} You probably need Mercurial with [[http://selenic.com/hg/rev/e5d3c7083d91|this changeset]] to use it with nginx properly. === Standalone === Simply run {{{ sudo -u www-data hg serve --webdir-conf /var/hg/hgweb.config }}} and enjoy this speedy method of serving multiple repos. It should be faster than using Apache. == You are done == Hooray! == Final Bits == No openSUSE specifics here. You should know the differences by now (apache user and doc path). === Create a new repository === {{{ $ sudo -u www-data hg init /var/hg/repos/ }}} === Provide more information about it === Add the following to the `/var/hg/repos//.hg/hgrc` file: {{{ [web] contact = Bilbo Baggins # Whom to contact, plain text, # no fancy stuff description = My precious! # Nice description what this is about, # you can include HTML (like ) }}} === Allow pushing to the repository === By default, nobody is allowed pushing. To allow pushing to everybody, add the following line to the `/var/hg/repos//.hg/hgrc` file: {{{ [web] allow_push = * }}} To allow only selected users to push changes, add the following line to the `/var/hg/repos//.hg/hgrc` file: {{{ [web] allow_push = frodo, sam }}} These are virtual users (for instance, as defined using a .htpasswd file), and not real system users. === Deny pushing to the repository === Most likely you will want to use it together with `allow_push = *`. If you want allow pushing to everybody, but a selected list of people, add the following line to the `/var/hg/repos//.hg/hgrc` file: {{{ [web] deny_push = saruman }}} === Allow pushing only over a non-secure channel === (I still need to check how it works :) ) By default, pushes are allowed only over https. If you are certain and do not want to enforce https for pushes, add the following line to the `/var/hg/repos//.hg/hgrc` file: {{{ [web] push_ssl = false }}} === Customize the look === Add the following to the `/var/hg/repos//.hg/hgrc` file: {{{ [web] # enable snapshot downloads allow_archive = gz zip bz2 }}} === Change the URL using baseurl and URL rewriting === For example, we might not want the hgwebdir.fcgi in our URLs if we have a dedicated or virtual host for our repositories. In the hgweb.config, add the following: {{{ [web] baseurl = / }}} hgwebdir will now write all links as `/x/y` instead of `/hgwebdir.fcgi/x/y`. Now we only need a rewrite rule. For Lighttpd, add (if hgwebdir.fcgi resides in the server's document root): {{{ url.rewrite-once = ( "^/?(.*)$" => "/hgwebdir.fcgi/$1" ) }}} For apache, add this to your .htaccess file. This example is based on hgwebdir.fcgi residing on your server as such "hg.example.com/cgi-bin/hgwebdir.fcgi" {{{ # Taken from http://www.pmwiki.org/wiki/Cookbook/CleanUrls#samedir # Used at http://ggap.sf.net/hg/ Options +ExecCGI RewriteEngine On #write base depending on where the base url lives RewriteBase /cgi-bin RewriteRule ^$ hgwebdir.fcgi [L] # Send requests for files that exist to those files. RewriteCond %{REQUEST_FILENAME} !-f # Send requests for directories that exist to those directories. RewriteCond %{REQUEST_FILENAME} !-d # Send requests to hgwebdir.cgi, appending the rest of url. RewriteRule (.*) /cgi-bin/hgwebdir.fcgi/$1 [QSA,L] }}} == See also == * PublishingRepositories * [[SharedSSH]] ---- CategoryHowTo CategoryWeb