Dependency Hell – and it’s not my package manager to blame

I’m writing a web service for conferences. I’ve been writing it, on and off for 3 years, and I think it would be fair to say the coding reflects my learning over those three years. The script is written in PHP. It has a MySQL backend. It has undergone a lot of changes.

In the beginning, was a concept. The concept was to have a digital timetable. One where you could register your interest in a talk, and the busiest talks got the biggest rooms. It should use any tech the user had to use the system including SMS. It would not expect you to have a user name and password combo (heaven forbid!), but would use OpenID.

The concept was implemented and demo’d at an event 2 years ago, and a friend [1] asked where the API was. “API?” I replied, feeling somewhat foolish, “It doesn’t have an API“.

[1] Lorna Jane Bowman (neé Mitchell aka @lornajane)

I realised the foolishness of my coding when discussing this issue [2]. “It’s dead simple, just create another view which returns just JSON” said someone. “View? I don’t know what you mean“… “Model, View, Controller? MVC – it’s a pretty common pattern”. “Oh no” I replied, “my project has inline PHP. It just seemed simpler that way.” “Well, how about you add a toString() function to your classes, and use that to render the output?” “Classes. Another thing my project doesn’t have. Sorry. Can’t do that either.

[2] With Lorna again, and adding Katherine Reeve (aka @BinaryKitten)

Did you ever get that slightly sinking feeling that maybe you’re making a fool of yourself?

“Well, there are lots of benefits to Classes in PHP, not least of which that you can use PHP CodeSniffer to enforce your coding standards” I start to laugh a little at this point “and you can use PHP Documenter to create all your developer documentation” which shut me right back up again “you can use PDO to create your objects from your database results” wow… mind blown – I’ve jumped on this little trick of a pony… “and of course, you can’t really do unit testing without classes”. Uh oh. “What’s unit testing?” “It’s a set of automated tests you can run against your code every time you commit to your version control software” whew! I’m using that at least! “to make sure that the code you’re committing hasn’t broken anything”.

Fast forward to this week, and I asked on Facebook whether anyone could teach me how to do Unit Testing. See, I’ve managed to cobble together my own MVC – I have classes called “Object_” which are the Models, the controller is my routing script – otherwise known as index.php, and then the Views are templates in Smarty, or just a json_encode((array) $object) [3] for my API. I have my own set of tests – not unit tests, it’s a script called “Test.sh” which runs PHP against the file (to make sure that the pages don’t have brackets missing or similar), then runs PHP Code Sniffer against it, and finally, if all the files are OK, it then runs PHPDoc against the whole mass of it.

[3] As the Apple iPhone/iPad adverts say – some sequences shortened

So, one of the books suggested to me, again by the lovely Lorna Jane, was The Grumpy Programmer’s Guide To Building Testable PHP Applications which mentioned that it’s much easier to do unit testing if you set up dependency injection. Note, I’m still not yet doing Unit testing here.

Woah. What the hell is Dependency Injection? Well, fortunately, there were code examples in that book. Oh boy, were there code examples. So let’s look through this idea. Instead of writing

$stuff = new MyStuff();

class MyStuff()
{
    function construct()
    {
        $this->db = mysql_connect('localhost', 'username', 'password');
        mysql_use_database('production', $this->db);
    }
}

You could instead write this:

$db = mysql_connect('localhost', 'username', 'password');
mysql_use_database('production', $this->db);
$stuff = new MyStuff($db);

class MyStuff()
{
    function construct($db = null)
    {
        $this->db = $db;
    }
}

So, this now means that in your testing framework, you could pass it a non-production database, or a testing database, or, well, anything that relies on something outside your script.

I did a bit of digging around to find some other examples of dependency injection code that might be a bit easier to use, which is to say, something so I don’t need to amend all my constructor functions.

I found this slideshare from a talk at PHP Barcelona about dependency injection which says you can do dependency injection like this:

$thing = new MyClass($dependency);

OR

$thing = new MyClass();
 $thing->setDependency($dependency);

OR

$thing = new MyClass();
 $thing->dependency = $dependency;

but somewhat weirdly, it also says that you can create a container class which holds your dependencies, and refer to that later – and that this isn’t a singleton. Sadly, I didn’t understand all that code fully (and have gone on to file a bug for the PHP documentation for the functions I didn’t understand to help people who follow behind with it!), but (and I’ve copied this verbatim from the slideshare) essentially, it looks like this:

class Container { protected $values = array(); function __set($id,$value) {  $this->values[$id] = $value;  }  function __get($id) {  if (!isset($this->values[$id])) {  throw new InvalidArgumentException(sprintf('Value "%s" is not defined.', $id));  }  if (is_callable($this->values[$id])) {  return $this->values[$id]($this);  } else { return $this->values[$id];  }  }  function asShared ($callable) {  return function($c) use ($callable) {  static $object; if (is_null($object)) {  $object=$callable($c);  }  return $object; }; } } $container = new Container(); $container->session_name='SESSION_ID'; $container->storage_class='SessionStorage'; $container->user = $container->asShared( function($c) {  return new User($c->storage);  } ); $container->storage = $container->asShared(  function($c) {  return new $c->storage_class($c->session_name); } );

Now, I have to be honest, this confuses the hell out of me. How on earth do I use this in my code? I’ve been doing this in my code thus far:

class Object_User{
  protected $intUserID    = null; // Obtained by the getCurrentUser() function
  protected $strUsername  = null;
  protected $hashPassword = null; // More of these, depending on the SQL

  function getCurrentUser() { // Called as $user = Base_User::getCurrentUser();
 $objCache  = Base_Cache::getHandler(); // A singleton to "cache" any data we've pulled to save getting it repeatedly
    if (
          isset($objCache->arrCache['Object_User']['current'])
          && $objCache->arrCache['Object_User']['current'] != false
        ) {
      return $objCache->arrCache['Object_User']['current'];
    }
    $arrRequest  = Base_Request::getRequest(); // Returns array of parsed request data
    $objDatabase = Base_Database::getConnection(); // Returns a PDO object

    $sql = "SELECT * FROM users WHERE strUsername = ? and hashPassword = ?";
    $query = $db->prepare($sql);
    $query->execute(array($request['username'], $request['password'])); $result = $query->fetchObject('Object_User');
    if ($result != false) {
      $objCache->arrCache['Object_User']['id'][$result->getVal('intUserID')] = $result;
      $objCache->arrCache['Object_User']['current'] = $result;
    }
    return $result;
  }

  function getVal($key) { // Return protected variables
    if (!isset($this->$key)) {
      return false;
    }
    return $this->$key;
  }
}

I know that singleton methods are considered “Bad” because they’re (apparently) difficult to unit test, but I would have thought that it would have been pretty straightforward to create a singleton class which holds all the dependencies (see following)

class Base_Dependencies
{
  protected static $handler = null; protected $arrDependencies = array();   protected function GetHandler() {
    if (self::$handler == null) { self::$handler = new self(); }
    return self::$handler;
  }

  function set($key, $dependency) {
    $handler = self::GetHandler();
    $handler->arrDependencies[$key] = $dependency;
  }

  function get($key) {
    $handler = self::GetHandler();
    if (isset($handler->arrDependencies[$key])) {
      return $handler->arrDependencies[$key];
    } else {
      return false;
    }
  }

  function unset($key) { // Only used for Unit Testing, I would imagine
    $handler = self::GetHandler();
    if (isset($handler->arrDependencies[$key])) {
      unset($handler->arrDependencies[$key]);
    }
  }
}

Doing it this way means I can, from, for example, my database class, which is currently a singleton, say instead:

function GetConnection() {
  $db = Base_Dependencies::get("Database");
  if ($db != false) {
    return $db;
  }
  $config = Base_Config::getAllConfig();
  $db = new PDO($config['DSN'], $config['DB_User'], $config['DB_Pass']);
  Base_Dependencies::set("Database", $db);
  return $db;
}

Is this wrong? Is this just not best practice? Given the above, how can I fix my dependencies in such a way that the poor schmuck who wants to commit a patch can figure out what they’re sending? Or do I just need to fix how my head works with this stuff? If it’s the latter, can someone provide some samples of what I’m doing wrong?

Thanks for reading this mammoth post!

Installing MOTP-AS under Ubuntu 11.10

Please note, I am having issues with localhost authentication. See below

MOTP-AS is a simple installable two-factor authentication system using the mOTP algorythm for generating one-time passwords. MOTP-AS integrates with FreeRadius to provide the same authentication to log in to managed servers in a consistent manner.

I’ve recently installed this on my Ubuntu 11.10 laptop and on my Ubuntu 12.04 Beta server, and the installation instructions worked on both, so I thought I’d share them with you.

Installing appropriate packages

sudo apt-get install libpam-radius-auth freeradius mysql-server phpmyadmin

Alternatively, use tasksel to install the LAMP server task, then

sudo apt-get install libpam-radius-auth freeradius

Download the latest version of motp-as from http://motp-as.network-cube.de/index.php/download/current-version

Unpack it.

tar xfz ~/Downloads/motp-as*

Setting up the database

Go into the Setup/MySQL directory of the MOTP-AS directory. Edit motp_schema.sql at the line “CREATE USER”. Change the password from motp to something more secure.

mysql -u root -p < motp_schema.sql

Now update Setup/config.php with the new password you just created.

Setting up the web site

Copy the HTML directory to /var/www/motp (or somewhere else in your web root). You may need to do this either as root, or as a user with permissions to write to /var/www

cp -Rf ~/MOTP-AS_*/HTML /var/www/motp

Note this must be done after you’ve made your changes to Setup/config.php

Setting up FreeRadius

Stop the FreeRadius service

sudo /etc/init.d/freeradius stop

Users

Backup the users file

sudo mv /etc/freeradius/users /etc/freeradius/users.dist

Edit the users file you’re about to copy in

nano ~/MOTP-AS_*/Setup/Freeradius/users

Find the part where it says “/var/www/htdocs/radius-auth.php” and change that to “/var/www/motp/radius-auth.php

Copy in the new users file

sudo cp ~/MOTP-AS_*/Setup/Freeradius/users /etc/freeradius/users

Dynamic Clients

Backup the dynamic-clients file

sudo mv /etc/freeradius/sites-available/dynamic-clients /etc/freeradius/sites-available/dynamic-clients.dist

Edit the new dynamic-clients file

nano ~/MOTP-AS_*/Setup/Freeradius/dynamic-clients

Find the three lines saying “/var/www/htdocs” and replace that string with “/var/www/motp” (I use Ctrl+W, Ctrl+R in nano to do a replace-all.)

Copy in the new dynamic-clients file

sudo cp ~/MOTP-AS_*/Setup/Freeradius/dynamic-clients /etc/freeradius/sites-available/dynamic-clients

Then make that function available

sudo ln -s /etc/freeradius/sites-available/dynamic-clients /etc/freeradius/sites-enabled/dynamic-clients

Accounting

Amend the default script to enable accounting

sudo cp /etc/freeradius/sites-available/default /etc/freeradius/sites-available/default.dist

Then edit it to use the MOTP accounting functions

sudo nano /etc/freeradius/sites-available/default

Search for the line “accounting {” then comment that whole block out with the hash/pound sign “#“. Fortunately in the distribution supplied default file, this only means commenting out a few lines, which are “detail“, “unix“, “radutmp“, “exec“, “attr_filter.accounting_response“, and then the closing “}” for that block.

If you’re using nano, press the insert key (or Ctrl+R if you can’t find that easily) and enter /home/MyUserName/MOTP-AS_v0.7.2/Setup/Freeradius/accounting (amend the path as appropriate). Replace the section “/var/www/htdocs” with “/var/www/motp“.

Save and exit

Finishing off FreeRadius

sudo /etc/init.d/freeradius start

Install your client

Personally, I have an Android device, and I chose to install the Mobile-OTP app from the Android Marketplace. I also, through work, have a Nokia 6303i Classic, on which I installed the MOTP application from the MOTP site.

I’ve heard good things about iOTP for iPhone, although I personally don’t have one.

Configuring MOTP

Go to http://localhost/motp (or https://yourdomain.com/motp)

Login with the username admin and password of motp.

Securing the admin account

Click on the red text in “First time configuration

Click on “Change password of User ‘admin’

Enter a new password. Do not set the time or uses section of this page. Click “Set“. Ignore the warning.

Click on “Home

Setting up your first user

Click on “Quick Add” (under “Wizards”)

Enter a username. It should be the username for your Ubuntu 11.10 device.

On the client, create a profile for the device. Most of them create a profile by asking for a seed, rather than a secret, so those will likely be more than 16 characters long – maybe even 20 (Mobile-OTP for Android) or 25 (MOTP Java app).

Once you’ve got your secret (on Mobile-OTP, by pushing-and-holding on the profile name and selecting “Show Secret“, on MOTP Java app, once you’ve put 0000 as the PIN for the first time to initialize it, you get a string “Init-Secret:“), put that into the “Secret” field, and then ask the user to set their pin here – I suggest 1234 initially, as the user can change it to something they want after.

Click OK, then click “Logout” and test authentication. If it all goes OK, they should be presented with “Welcome to the Mobile OTP Authentication Server“.

Under “Settings” they can change their own PIN.

Testing radius authentication works OK

Run the radius testing program, like this, as a user:

radtest username passcode localhost 0 testing123

(This assumes the default localhost password hasn’t changed)

If you get anything like “rad_recv: Access-Reject packet from host“, then you’ve failed to configure something properly, or you’ve entered the PIN or code wrong.

Restart FreeRadius in debugging mode by doing the following:

/etc/init.d/freeradius stop
/usr/sbin/freeradius -X

This will produce a large quantity of logs on-screen, so I’d suggest running the test itself from a separate window. Run the radtest command (listed above) again. Look for your error messages. In my case, I forgot to update the line in users, so I saw this error message: Could not open input file: /var/www/htdocs/radius-auth.php

To find where this fault was, I did (as root, in /etc/freeradius)

find -R 'htdocs' /etc/freeradius

And got back: users: Exec-Program-Wait = “/usr/bin/php /var/www/htdocs/radius-auth.php %{User-Name} %{User-Password} %{Client-Shortname}”

That told me the fault was in the users file.

Fix the issue, check it again, and when you get this message “rad_recv: Access-Accept packet from host” press Ctrl+C to cancel the test mode of FreeRadius, and then run:

sudo /etc/init.d/freeradius start

Configuring pam_radius_auth.conf

Edit /etc/pam_radius_auth.conf

sudo nano /etc/pam_radius_auth.conf

Find the line which says “127.0.0.1” and replace the shared secret with something you want your server to use. You will also need to amend /etc/freeradius/clients.conf and replace the “secret” in the localhost client there (by default, it’s “testing123” in freeradius).

If you want to use your OTP for all authentication credentials, edit /etc/pam.d/common-auth, or if you just want to use it with specific access protocols, edit the relevant file in /etc/pam.d for the authentication systems you want to use OTP for.

You need to add the following line – either on the line before “@include common-auth” (for non common-auth files) or after the primary comment block for common-auth.

auth sufficient pam_radius_auth.so

Open a separate terminal session to your box (especially! if you’re remote) and ensure you can still login with your regular credentials.

Then try a connection with your radius credentials. It should just work! If not, stop the freeradius server and re-run it using /usr/sbin/freeradius -X and see whether you’re getting a different error message.

** UPDATE **

I have noticed that I’m getting locked out when using my non-radius credentials. This is probably due to the placement of the line in the /etc/pam.d/common-auth – it should probably come after the pam_unix.so line, but I’ve not tested that yet. I’m also going to try to suggest that there be an optional time-out period on locked accounts to the developers of MOTP-AS.

The second issue I’m struggling with is that I’m getting errors when using the LightDM. I’m getting the following error message in /var/log/auth.log:

pam_succeed_if(lightdm:auth): requirement "user ingroup nopasswdlogin" not met by user "spriggsj"

I don’t know if this is because I’m using ecryptfs as well, or because there’s something wonky going on with the common-auth structure I’m using.

CampFireManager Workshop

NOTE: Images have been removed from this post 2017-05-02

What is CampFireManager?

CampFireManager is a tool to schedule talks at a Conference or Unconference (such as a barcamp). It is written in PHP and uses a MySQL backend to store the data about the event. It optionally can use SMS messaging and Microblogging services (like identi.ca or Twitter) to perform routine tasks on the system.

A modified version (CampFireManager-Lite) of CampFireManager was recently used at OggCamp ’10, in combination with tools created by Xibo. Here’s a look at what you would have seen if you were there:

CampFireManager-Lite

This is the main timetable. Note, it only shows two talks before “now”, the talk on “now” and the next 6 slot’s worth of talks. This caused us problems on the Saturday afternoon when we tried to display the timetable for the next day. The only way we could do that was to change the system time on the server.

This is the now and next screen for the main stage, and a screen showing similar information for each of the stages. By default, each screen shows, in turn, the timetable, the “now and next” screen for all stages and then the “now and next” screen for each individual screen. This is entered by the database, but should *really* have been configured at the screen.

Administration was performed by organisers. Data was collected on sheets, and then entered into the system. Using an in-built version of the timetable, these were placed into the appropriate slots in the timetable, which then updated all the other screens.

This, however, isn’t how I originally envisaged CampFireManager. The full version of CampFireManager is designed to operate with minimal administrative overhead from on-site staff.

CampFireManager

Users (as well as support staff and administrators) would log into the website using OpenID. Several common providers are pre-populated and available using icons to select them.

On logging in, all users initially see a timetable, with no definitions of rooms, but with the complete timetable for this day. At the top, under the “Slot” times, is a hyperlink showing “New Talk”. Clicking on this brings you to a page where you can enter your talk proposal. Clicking on “Go” inserts that talk into your timetable.

If you wanted to enter some contact details for your talk (so that your peers at the event can reach you), this is done by clicking on “Amend Contact Details”, and entering your details. This will retrospectively correct your details on all the talks you’re giving.

By clicking on “Add other access method” – you can add an “authorization code” (which is a 5 character string of case sensitive, non-similarly rendered letters and numbers) to allow you to update your talks, your contact details or the talks you’ll be attending from other communication methods – SMS or XMPP being the two primary examples.

However, I realised that I could use this access method logic to allow administrators the ability to grant administrative access to other organisers, or to promote staff to “Support” staff. In the same way you’d enter a 5 character string to give the same rights to your phone as your web access, you could enter a 9 character string (again, non-similarly rendered, case sensitive letters and numbers) to become an administrator or support staff member. Here you see an Administrator’s view of the timetable, which adds two extra options at the top box – Provide support to attendees and modify config values.

So, let’s support an attendee. We enter some details, either their Auth Code (if they know it), or some details about them they entered previously – a blog address, e-mail, some other contact method, or in extreme cases, their name. If we’ve had to use any form of wildcard in the search (that is, I typed “Jon” into the name box), then we get prompted to confirm it’s them, and can give them their unique Auth Code. If they’ve not registered on the site, then click on Create New Auth Code to create a new user. In either case, we get a page showing “This is with an AuthCode: ". From here we can amend their contact details, propose a new talk on their behalf, or if there's a talk in the future, show their attendance (which, to be fair, here I couldn't do, as I'd not created a talk for them to attend).

Here’s the administrative screen, where you see options that have configured the screens and access methods.

There’s a mobile interface to the site, which is quite similar to the support staff’s view of the site, and again it uses OpenID to authenticate. It gives the user access to a slimmed down portal – showing all talks which are yet to occur (allowing the user to show or remove their attendance), just this user’s talks yet to occur, and allowing the user to add a new talk and amend their contact details.

That is all the interactive screens shown, so here’s the two most regularly seen, non-interactive screens, both with a scrolling box between the event title and the content saying (in this instance):

Phones: 07 88 24 64 91 8 on the Three Network (with 18% signal)
Website: https://home.north.sprig.gs/CampFireManager/
Mobile site: https://home.north.sprig.gs/CampFireManager/m/
Event Hashtag: #oc10
Identify with this service by sending
I <your name> [email:your@email.address] [http://your.web.site]
(there are more options for identification by going to the website)
Propose a talk by sending P <Time Slot> <Slots Used> <Talk Title>
Cancel a talk by sending C <Talk Number> <Time Slot> [Reason]
Rename a talk by sending E <Talk Number> <Time Slot> <New Talk Title>
Attend a talk by sending A <Talk Number>
Decline the attendance of a talk by sending R <Talk Number>
Note: You can combine multiple A and R commands in one message.
Statements surrounded with <> are mandatory options, those statements surrounded with [] are optional.
These commands should be sent to your preferred mobile service listed above.

The Timetable:

And the “Direction” screens (like the now and next screens on the CampFireManager-Lite):

The above URLs do work, and the system is available for experimentation. The code and ticket tracker (if you want to run your own, local, instance) is at http://code.google.com/p/campfiremanager. Please contact me if you want to get access to my instance of CampFireManager from an administrative or support perspective. Please note, the SMS engine is not running.