For these reasons and for general flexibility, my current project wanted to be able to easily swap the underlying triple store from ARC to Jena as needed so I needed to investigate how to expose a Jena triple store as a SPARQL endpoint. After working this out, I now really really appreciate how easy ARC makes this.

Jena doesn’t appear to ship with the ability to expose the ARQ SPARQL processor as a SPARQL endpoint and hence you need to make use of a separate piece of software called Joseki. The following is the list of things I needed to do to get this working in my environment. Note that your setup may have different requirements and also I may have completely misunderstood the best way of doing this!

1. Setup a database to use as your triple store and get a JDBC driver so Joseki can interact with it from Java
2. Download and extract Joseki
3. Add the JDBC driver to the Joseki classpath (e.g. for Windows by adding the following line to bin\joseki_path.bat: set CP=%CP%;C:\my_jdbc_driver\my_jdbc_driver.jar)
4. Add the following to joseki-config.ttl:

    <#myProjectUpdate>
      rdf:type            joseki:Service ;
      rdfs:label          "My Project SPARQL/Update" ;
      joseki:serviceRef   "sparql/myproject/update" ;
      joseki:dataset      <#myProject> ;
      joseki:processor    joseki:ProcessorSPARQLUpdate .
   
    <#myProjectRead>
      rdf:type            joseki:Service ;
      rdfs:label          "SPARQL" ;
      joseki:serviceRef   "sparql/myproject/read" ;
      joseki:dataset      <#myProject> ;
      joseki:processor    joseki:ProcessorSPARQL_FixedDS .
   
    <#myProject>
      rdf:type            ja:RDFDataset ;
      rdfs:label          "My Project" ;
      ja:defaultGraph     <#myProjectDB> .
   
    <#myProjectDB>
      rdf:type            ja:RDBModel ;
      ja:connection       [
                            ja:dbType "MySQL" ;
                            ja:dbURL           ;
                            ja:dbUser         "myproject-database-username" ;
                            ja:dbPassword     "myproject-database-password" ;
                            ja:dbClass        "com.mysql.jdbc.Driver"
                           ] ;
      ja:reificationMode    ja:minimal ;
      ja:modelName        "DEFAULT" .
       

5. Set the JOSEKIROOT environment variable to the location you extracted Joskei
6. Run Joseki (from it’s directory) by executing bin/rdfserver.bat

Note that I wanted to be able to make use of SPARUL to update data using the SPARQL endpoint. In ARC I can use SPARQL+ (which is effectively the same for my purposes) on the same endpoint as normal SPARQL queries. For Joseki however, I needed to expose two different endpoints, one for standard SPARQL queries and one for updating.

The one thing I haven’t yet worked out how to do it to be able to use named graphs in my Jena triple store when inserting data. I discovered that the SPARUL update specification requires you to create the graph first (unlike ARC’s SPARQL+) but executing e.g. CREATE GRAPH <http://mygraph/> seems to fail silently as any following INSERT INTO <http://mygraph/> statement fails saying that the graph doesn’t exist. Something to keep investigating. It may be something to do with support for the different types of Jena store (RDB, SDB, TDB, etc) which I don’t fully understand yet (I think my instructions above are using RDB which appears to be old but I couldn’t get TDB or SDB working at all).

So all in all I’m pleased to have worked out how to set this up but I will most certainly continue to use ARC where possible as Jena environments seem unnecessarily complex (although this might simply be because it tends to support the W3 specifications fully!).

This is pretty easy using the Ubuntu restricted drivers tools: Ubuntu Menu -> System -> Hardware Drivers (in the 2nd box of applications). Note that you’ll need an Internet connection to actually install this so hopefully you can make use of a wired connection temporarily! When I initially installed this driver there were actually two to choose from but it was the Broadcom STA driver that worked. Now I only see a single option available.

Creating the indicator
The class used to represent an indicator appears to have changed from IndicatorMessage to Indicator so I threw in the following code to try the new one and fallback to the old one:

try:
  # Ubuntu 9.10 and above
  indicator = indicate.Indicator()
except:
  # Ubuntu 9.04
  indicator = indicate.IndicatorMessage()

Drawing attention
Previously, indicators automatically made the indicator applet draw your attention with a green dot. In Karmic the green dot appears to have been replaced with a change of the envelope colour to black but it is no longer automatic. To make this work you need to set the draw-attention property:

indicator.set_property('draw-attention', 'true');

Note that they’ve also added a count property to display how many notifications are from the same source.

Recently I wanted to build a .deb package for an internal IBM application I was writing so that users could easily install it and also so we could distribute them through some internal repositories. This proved a bit harder than I expected so this is a quick summary of how I ended up doing it. Note that your requirements might be entirely different!

The first thing to do is to create the files required by the packaging process. I discovered that the dh_make command can create a load of sample files that can be used as part of this process. To do this, create a directory in the format [package-name]-[version] (e.g. my-great-app-1.0) and run dh_make from within it (I specified ’s’ for single binary when prompted). This will create a load of sample files in a ‘debian’ subdirectory. Delete any of these you don’t need (which is probably most of them); I kept the following:

• changelog – change history for all versions of the app (keep to the format specified by the Debian Policy Manual)
• compat – no idea why I needed this but things don’t work properly later if I don’t
• control – the details of the package you are creating (see the specification for all configuration options)
• dirs – the list of directories in which your app will install files (e.g. /usr/bin, /usr/share/pyshared/my-great-app, /usr/share/applications)
• README.Debian – the README for your app
• rules – a MakeFile with instructions for how to create the package (for my Python app the important bit here was in the ‘install’ section; here I created a $(CURDIR)/debian/my-great-app subdirectory and copied all files into it as if it were /, e.g. binary to $(CURDIR)/debian/my-great-app/usr/bin/my-great-app)

Once I’d created all those files and put them in my-great-app/packaging/debian and my source in my-great-app/src I created a simple build script my-great-app/bin/build. This looked something like the following:


#!/bin/bash

cd ../build
sudo rm -rf my-great-app*
mkdir -p my-great-app-$VERSION
cp -u ../src/*.py ../src/*.desktop ../src/*.ico ../packaging/my-great-app my-great-app-$VERSION
tar -czf my-great-app-$VERSION.orig.tar.gz my-great-app-$VERSION/
cd my-great-app-$VERSION
mkdir debian
cp -u ../../packaging/debian/* debian/
gksu dpkg-buildpackage


This should create you a my-great-app_1.0-1_all.deb and the my-great-app_1.0-1_i386.changes, my-great-app_1.0-1.dsc and my-great-app_1.0-1.tar.gz files your repository maintainer might want.

A really useful video I found for helping me fill in the contents of the debian control files (and getting me through the whole process) was here. Definitely worth checking out if you need to do this yourself.

Existing stuff

• Good packaging formats
• Depency solvers, downloaders and UIs bolted on
• Can’t have automatic updates (needs password authentication)
• Can’t use fast-user switching (lock out install applications/databases)
• Errors/warnings in English only and really confusing to average user
• Installation is done by package names not application names (many to many relationships)
• Can power down during update – bit dangerous!

PackageKit implementation

• The ‘glue’
• Integrates with existing tools (including dependency mangement etc)
• Improves authentication (uses PolicyKit – fine grained control)
• System activited daemon (only running when you need it)
• Only need to write simple integration between tools and PackageKit (doesn’t even need to be complete and done for most tools already) plus thin UI
• Uses DBUS (two layers – one for full control, one “just do it”)
• Applications can integrate directly (e.g. install clipart from openoffice)
• Installation/update by application not package (users know what it is they’re installing!)
• Doesn’t allow shutdown during installs

PackageKit project

• Easy to contribute (git with anonymous access – merged to release daily)
• Rapid development (roughly one minor release per month)
• Shipped with Fedora 9 (and others)
• Strong interest from OpenMOKO, Ubuntu (and others)

I’ve installed an old-ish release on my Ubuntu machine (straight from the repositories) and it looks pretty good. Definitely gonna pay attention to this project, it looks like a big step in the right direction.

This is quite a long article and you might only be interested in some of it so here are the sections:

• Getting access to request data
• Returning appropriate responses
• Handling different data formats
• Passing all requests to a single PHP script
• Determining the full requested URL
• A generic REST Service class

If you have an suggestions for improvement, please let me know – this was a first attempt!

Some credit for this article should go to lornajane for her PHP REST Server articles (Part 1, Part 2, Part 3) which I used as a good starting point.

Getting access to request data

Getting access to HTTP GET arguments or POST data is well known and easy. With a REST interface you are likely to support other HTTP verbs such as PUT and DELETE. Getting data from these requests is not obvious but luckily lornajane had worked this out for me:

parse_str(file_get_contents('php://input'), $arguments);


Returning appropriate responses

With REST you should really make use of the HTTP Status Codes for all requests (rather that using 200 OK for everything). In PHP, this can be done with the header function in a couple of ways:

// When setting just the status code
header('HTTP/1.1 405 Method Not Allowed');
// When you're setting another header at the same time
header('Allow: GET, HEAD, POST, DELETE', true, 405);


I haven’t found a way of sending just the status code number (e.g. 405) without specifying the exact text (‘Method Not Allowed’) but these are part of a fixed specification so it’s not a huge problem.

The following is a list of status codes that I’ve used and why:

• 200 OK: successful request when data is returned
• 201 Created: Successful request when something is created at another URL (specified by the value returned in the Location header)
• 204 No Content: Successful request when no data is returned
• 400 Bad Request: Incorrect parameters specified on request
• 404 Not Found: No resource at the specified URL
• 405 Method Not Allowed: when a client makes a request using an HTTP verb not supported at the requested URL (supported verbs are returned in the Allow header)
• 406 Not Acceptable: Requested data format not supported
• 500 Internal Server Error: An unexpected error occurred
• 501 Not Implemented: when a client makes a request using an unknown HTTP verb

Handling different data formats

As URLs are supposed to represent resources in REST, it is a feasible requirement that client applications would like to receive responses from the same URL in different formats. Clients can specify the data formats they are capable of understanding using the HTTP Accept header. In Javacript this can be set using the setRequestHeader method on the XMLHttpRequest object. Using jQuery as my current Javascript library of choice, the code for doing this is as follows:

$.ajax({
  type: method, // GET, POST, PUT, DELETE etc.
  url: url, // The actual URL to make the request
  data: data, // Any data/parameters to send to the server
  beforeSend: function(xmlHttpRequest) {
    xmlHttpRequest.setRequestHeader('Accept', format); // MIME Type
  }
});


This then needs to be handled by the server. In PHP this data can be retrieved using $_SERVER['HTTP_ACCEPT']. I also found a handy library for parsing this header according to the specification (not trivial), cleverly named ‘HTTP_ACCEPT‘. I then wanted a way of knowing the most appropriate format from a list of formats I would support so I threw the following code together:

$accept = new HTTP_Accept($_SERVER['HTTP_ACCEPT']);
foreach ($supportedFormats as $supportedFormat) {
  $supportedFormatQuality = $accept->getQuality($supportedFormat);
  if ((!isset($bestFormat) && $supportedFormatQuality > 0) ||
       $bestQuality < $supportedFormatQuality) {
    $bestFormat = $supportedFormat;
    $bestQuality = $supportedFormatQuality;
  }
}


The server also needs to specify the type of content it is returning; this can be done using the Content-Type header as follows:

header("Content-Type: $format");


Passing all requests to a single PHP script

I needed all requests with a certain URL prefix to be handled by a single script. This was easily solved using the following URL re-writing rule in a .htaccess file:

Options +FollowSymLinks
RewriteEngine on
RewriteRule ^.*$ index.php


Note that I had to enable the mod_rewrite module in Apache and ensure that directives could be overridden with the following:

<Directory /var/www/>
  AllowOverride All
</Directory>


Determining the full requested URL

Another requirement of my application complicated by URL rewriting was that I needed access to the full URL. I achieved this with the following bit of code:

$protocol = $_SERVER['HTTPS'] == 'on' ? 'https' : 'http';
$location = $_SERVER['REQUEST_URI'];
if ($_SERVER['QUERY_STRING']) {
  $location = substr($location, 0, strrpos($location, $_SERVER['QUERY_STRING']) - 1);
}
$url = $protocol.'://'.$_SERVER['HTTP_HOST'].$location;


A generic REST Service class

To allow me to reuse some of the REST code I'd created, I put it into a generic REST Service class which I subclassed and replaced different methods for different means:

class RestService {

  private $supportedMethods;

  public function __construct($supportedMethods) {
    $this->supportedMethods = $supportedMethods;
  }

  public function handleRawRequest($_SERVER, $_GET, $_POST) {
    $url = $this->getFullUrl($_SERVER);
    $method = $_SERVER['REQUEST_METHOD'];
    switch ($method) {
      case 'GET':
      case 'HEAD':
        $arguments = $_GET;
        break;
      case 'POST':
        $arguments = $_POST;
        break;
      case 'PUT':
      case 'DELETE':
        parse_str(file_get_contents('php://input'), $arguments);
        break;
    }
    $accept = $_SERVER['HTTP_ACCEPT'];
    $this->handleRequest($url, $method, $arguments, $accept);
  }

  protected function getFullUrl($_SERVER) {
    $protocol = $_SERVER['HTTPS'] == 'on' ? 'https' : 'http';
    $location = $_SERVER['REQUEST_URI'];
    if ($_SERVER['QUERY_STRING']) {
      $location = substr($location, 0, strrpos($location, $_SERVER['QUERY_STRING']) - 1);
    }
    return $protocol.'://'.$_SERVER['HTTP_HOST'].$location;
  }

  public function handleRequest($url, $method, $arguments, $accept) {
    switch($method) {
      case 'GET':
        $this->performGet($url, $arguments, $accept);
        break;
      case 'HEAD':
        $this->performHead($url, $arguments, $accept);
        break;
      case 'POST':
        $this->performPost($url, $arguments, $accept);
        break;
      case 'PUT':
        $this->performPut($url, $arguments, $accept);
        break;
      case 'DELETE':
        $this->performDelete($url, $arguments, $accept);
        break;
      default:
        /* 501 (Not Implemented) for any unknown methods */
        header('Allow: ' . $this->supportedMethods, true, 501);
    }
  }

  protected function methodNotAllowedResponse() {
    /* 405 (Method Not Allowed) */
    header('Allow: ' . $this->supportedMethods, true, 405);
  }

  public function performGet($url, $arguments, $accept) {
    $this->methodNotAllowedResponse();
  }

  public function performHead($url, $arguments, $accept) {
    $this->methodNotAllowedResponse();
  }

  public function performPost($url, $arguments, $accept) {
    $this->methodNotAllowedResponse();
  }

  public function performPut($url, $arguments, $accept) {
    $this->methodNotAllowedResponse();
  }

  public function performDelete($url, $arguments, $accept) {
    $this->methodNotAllowedResponse();
  }

}


# convert to pdf
unoconv -f pdf myfile1.odt myfile2.odt ...
# merge pdfs
pdftk myfile1.pdf myfile2.pdf ...
# remove individual pdf documents
rm myfile1.pdf myfile2.pdf ...


A handy little hack to generate my entire portfolio.

For anyone interested, it’s a Perl script which watches for the D-Bus signals from gnome-screensaver. Code is as follows:

my $cmd = "dbus-monitor --session \"type='signal',interface='org.gnome.ScreenSaver',member='SessionIdleChanged'\"";
open (IN, "$cmd |");
while (<in>) {
  if (m/^\s+boolean true/) {
    system('/etc/acpi/screenblank.sh')
  }
}


Note that it relies you running Linux, using GNOME and there being a screen blanking script ‘/etc/acpi/screenblank.sh‘ (which in turn executes ‘/usr/share/acpi-support/screenblank‘ for every display). Not too much to ask I’m sure

– Edit 8th October 2008:
Since upgrading to Ubuntu Karmic, this script no longer seems to work for a couple of reasons. I’ve not investigated in any detail but I do have a new working script using a slightly different D-Bus signal and a different command to blank the screen. Try this if the one above doesn’t work!

#!/usr/bin/perl
my $cmd = "dbus-monitor --session \"type='signal',interface='org.gnome.ScreenSaver',member='ActiveChanged'\"";
open (IN, "$cmd |");
while (<IN>) {
  if (m/^\s+boolean true/) {
    system('xset dpms force off')
  }
}


The One Account is one of those ‘offset everything you own against your mortgage’ type accounts. This means that you merge your mortgage, savings and current accounts in a single pot so any money you have is constantly offset against your mortgage. This means the actual amount of money you’re borrowing at any one time is less that it otherwise would be and hence you pay less in interest. Ok… sounds interesting but what does that actually mean? Well, it gives you a variety of options but the most important one for me is that you can pay of your mortgage in a much shorter time period.

We signed up for the One Account at the beginning of this year and set the ambitious aim of paying our (not insignificant) mortgage off in 4 years instead of the usual 20-25. What?! 4 years? Are you freakin’ crazy I hear you say? [Well you might not be saying that but it makes this post far more interesting for me if I imagine you are so I'll pretend.] We didn’t pluck the 4 year figure out of thin air; the One Account website has a handy Mortgage Shrinker you can use as a guide. The following graph gives you an idea how this works:

As you can see, salary comes in every now and then which reduces how me we owe to the bank and then this amount slowly increases as we spend money before the next salary payment. More importantly, the amount it increases to each time is lower than the previous amount, i.e. the mortgage is getting smaller.

So half a year in, how’s it looking? Pretty darn good actually. Amazingly, that ability to offset your salary payments against your mortgage until you spend them, plus the ability to effectively pay off bits of your mortgage constantly has left us in pretty good shape. We are entirely on track to pay the thing off in 4 years, if not sooner.

P.S. Worth noting that just because you can pay off loads of your mortgage quickly, doesn’t mean you can’t spend it too. Effectively you could end up only having paid off interest after 20 years and no capital, or pay off £20k and then blow it on a new car or something. Only a good idea if you have some self-control over your finances!