Phing

PHP::Impact blog has very nice article about remote files synchronization. Given FileSyncTask library uses rsync, console utility, which is de-facto standard for syncing files. Everything sounds great, but things get worse, when you try to sync files in shared hosting environment, where only FTP is available.

Code in the hole some months ago posted solution which uses Net_FTP (FTP functionality wrapper) package to upload files to server. I decided to try given solution, so I modified it to download source code from SVN first and then upload them. My tests code:

<?xml version="1.0" ?>
<project name="project" basedir="." default="build">
 
<property file="build.properties" />
 
<fileset dir="${sync.projectdir}/" id="files.main">
	<include name="**/*" />
        <!-- netbeans project folder -->
	<exclude name="nbproject/**" />
</fileset>
 
<target name="build" depends="svnexport,deploy-application-files" />
 
<target name="deploy-application-files">
	<echo msg="Deploying application files" />
	<phingcall target="deploy">
	    <property name="deploy.fileset.refid" value="files.main"/>
	</phingcall>
</target>
 
<target name="deploy">
        <ftpdeploy
	    host="${ftp.host}" port="${ftp.port}"
	    username="${ftp.username}" password="${ftp.password}"
	    dir="${ftp.dir}">
	    <fileset refid="${deploy.fileset.refid}" />
	</ftpdeploy>
</target>
 
<target name="svnexport">
	<delete dir="${sync.projectdir}" />
 
	<svnexport
	    username="${svn.username}" password="${svn.password}"
	    nocache="true" force="true"
	    repositoryurl="${svn.repo}" todir="${sync.projectdir}"
        />
</target>
 
</project>

Project was successfully uploaded, but it took more than half an hour (I have university Internet which is very fast). Problem is, it doesn’t check for files changes – it just uploads every single file in fileset. I don’t blame anyone, FTP is not made for syncing files and uploading huge amount of tiny files is very slow. We need other solution.

I have come up with another idea, which works by:

1. Get last-revision-id from production server
2. Download modified files (difference from last-revision-id to HEAD) from SVN repository
3. Upload modified files
4. Change last-revision-id to current ID

It works in theory and can be really simply implemented with Phing, but it’s still not very good solution. There is too much trust for one file (last-revision-id). Unless you can make sure that nobody will touch any files inside server without using Phing build script, things can easily brake. That’s why I haven’t tried it.

Have you used anything like this? Currently I’m thinking about moving to different server, because these limits (no ssh) for bigger projects creates massive headaches. I have came up with some hacks, as written above, but they are too sensitive and just mimics actual synchronization.

For example, if you change your CSS code, you want users to get it immediately and not after their browser cached version expires. One of possible ways, to force browser to load new version, is to add version number as a query string.

Normal HTML stylesheet code looks like this:
<link rel=”stylesheet” href=”/style.css” type=”text/css” media=”screen” />,
and with added version number:
<link rel=”stylesheet” href=”/style.css?91” type=”text/css” media=”screen” />.

Browsers will always grab new stylesheet after each url change – you can increase version number by +1, add date, etc. Let it be our goal – use SVN functionality to automatically tag stylesheets.

First thing to do, is to insert $Rev$ in source code of selected file (mine: style.css). Then go to console and set SVN property:

svn propset svn:keywords "Rev" style.css

Now, after each commit to repository, this keyword will be replaced by $Rev: NUMBER $, where NUMBER is current revision number. Next thing is to use PHP to read revision number from style.css source and append it to style URL.

Reading revision number is the easiest part, because if you store revision number in first line, you can quite efficiently grab only that part of source and preg_match/simple match number. Sample script looks like this:

//reading stream
$handle = fopen("style.css", "r");
//read first line, TODO: check if it's not empty, etc.
$first_line = fgets ($handle);
//extract revision number, chosen format: "/* $Rev: 1424314 $ */"
$revid = substr($first_line, 9, -6);
//print generated link
print "style.css? " . $revid;

You can easily modify and transform it to function, but key concept will be the same. However, this method is useful only to show revision number of one file, because revision number changes only after commit. For whole project revision number there are better solutions.

I would love to hear your ideas, because I brainstormed this one only a few days ago and I still think it can be made more efficiently. One of possible enchantments would be using build script, but I’m still searching for more simpler solutions.