A small, but funny, fail.

No Comments

So, I have this bug where when I shoot a bullet from a turret, the cruiser starts to spin! Not the intended behaviour :)

On Youtube

Making a turret in Unity 3D

1 Comment

As a start to Unity3D I tried to make a turret that consists of 2 parts. One base part and one barrel part. The barrel is connected via “hinges” to the base part.

On this image the base(2) and the barrel(3) is shown whit their intended rotations around their axis. The base rotates on the y axis and the barrel moves up and down on the x axis.
It is important to have the z axis point forward, since that is what Unity assumes to be forward. (I discovered that with trail and error instead of just reading up on the subject)

Hierarchy

When that was done i needed to place it in a proper hierarchy. The Top is an empty GameObject. In there I put an other empty GameObject and named it Sides, and added 2 pillar cubes to it. The barrel is a cube renamed to CenterPiece, the cube inside it is to show where to front is.

Scripting

We have two rotating parts. The rotation is done independently from each other but together they will point at an object.

Got values?

I got the rotation values towards the other object with the neat little Unity function: LookAt. Lookat points the z axis towards the object.
I only need two rotation values. So I saved the values before the lookAt, called lookAt, saved the new rotation values, and set the rotation back to the old values before rotating.

Turning

I made two separate scripts for the base and barrel, while the code is very similar there are slight differences like the axis.
I use a simple transform.Rotate with the help of Mathf.MoveTowardsAngle to smootly turn to only on the required angle. I first tried turning with slerp, but the barrel and base rotated slightly out of sync making the whole thing look terrible. MoveTowardsAngle also gives a nice constant rotation speed, which was exactly what I was looking for.

Starting a game project with Unity 3D

No Comments

Jasper and me are going (at least try) to make a game with Unity 3D. It will be a space opera kind of game. No details yet!

I’ve been playing with Unity 3D for a few days now and the experience so far has been mostly positive.

Starting

I have never done anything with 3D really, so this is completely new for me. And I’m not too good yet at the 3D maths, but that will be a skill to be forcefully improved.

The documentation is decent. The API docs are very brief, and for someone who does not know all the 3D terminology it is hard read and to decide which function to take. It lacks information about what the benefits are and what the function can and cannot do.

I started out doing a tutorial. Unfortunately this tutorial was for an older Unity version. Much is still the same though, just the interface changed. Unfortunately I found this tutorial to be tedious and lacking excitement. While you do get quick visual results with it, it was not what I wanted to know.

Scripting Language choice

We picked JavaScript, or at least the Unity version of it. Just when I thought I was beginning to understand JavaScript in the browser, this dialect comes along. It seems to have better scoping, which is welcome.
Unity comes with a customized version of MonoDevelop. It’s an IDE, with completion for UnityScript. Quite helpful, it had an integrated debugger. So far I’ve only used it to edit the scripts.

Managing Assets

I’ve decided to use Mercurial for this project. I’ve set up a private repository on Bitbucket, and shared it with Jasper. Easy as pie. Just be sure to follow these instructions.
I have not used Mercurial on Windows much but TortoiseHG is manageable, still I would prefer having the command-line tools instead.

We’ll have to see what we want to do with the actual game assets when textures and sounds come along. Mercurial is not the best tool to handle those.

To be continued…

Javascript parent constructor being run when defining inheritance and a solution

No Comments

When using the traditional Javascript methods to do some sort of inheritance, the constructor of the parent will always be run when trying to inherit it. This seems to be a unfortunate side effect of using new a().

The problem

Consider this code. Here b inherits from a

function a() {
    console.log("a");
}
 
function b() {
    a.call(this);
    console.log("b");
}
b.prototype = new a(); // a's constructor will be run here!
b.prototype.constructor = b;
 
new b(); // run once
new b(); // run twice

Here a’s constructor will be run 3 times.

a
a
b
a
b

The solution

There is a simple fix for this, but its only in Ecmascript 5.
Change this:

b.prototype = new a();

Into this:

b.prototype = Object.create(a.prototype);

And the constructor will only be run when actually being called:

a
b
a
b

Using hgtools with mercurial tags to automatically version packages

No Comments

I liked to automate my python package versioning a bit more and did this with hgtools. With hgtools it is quite easy to store the version numbers only in your repository as tags, and not having it again in setup.py.

A minimal setup.py can look like this:

from setuptools import setup, find_packages
setup(
    name="HelloWorld",
    use_hg_version=True,
    packages=find_packages(),
    setup_requires=["hgtools"],
)

Just give a revision a tag and make an sdist, or bdist and setuptools will use the version from your tag.
Tags have to conform to strict version formatting from distutils to be used by hgtools.

Later I also made it a step in a Jenkins setup, to always build a sdist and copy it to a HTTP accessible directory so it could be installed with pip. This gave me a nice history of available release packages and also, dev versions.

Django class based views and testing them

1 Comment

One of the problems described well by Honza Král’s presentation on Djangocon.eu 2010, is how to test properly and make it easier. In his presentation he touches on using class based views instead of using normal functions (around 14:14 in the link video). This is not as straight forwards as it seems at first. There are some problems with thead-safety for one.

To have fast tests you should avoid hitting the database as much as you can, and after using this method I have to agree.

Making a class based view

We have this simple view:

from django.shortcuts import render_to_response, get_object_or_404
 
from mysite.polls.models import Poll
 
def simple_view(request, poll_id):
    """ A simple view, which takes an id from a url, shows it in a template. """
    poll = get_object_or_404(Poll, pk=poll_id)
    poll.access_counter += 1
    poll.save()
    return render_to_response('polls/detail.html', {'poll': poll})

And now as a Class based view:

from django.template import Context, loader
from django.http import HttpResponse
 
from mysite.polls.models import Poll
 
class simple_view(object):
    def get_objects(self, poll_id):
        """ Return poll object for given poll_id. """
        return get_object_or_404(Poll, pk=poll_id)
 
    def process(self, poll):
        """ Manipulate data and return the manipulated, unsaved object. """
        poll.access_counter += 1
        return poll
 
    def render(self, poll):
        """ Return HTML. """
        template = loader.get_template('polls/detail.html')
        context = Context({'poll': poll})
        return template.render(context)
 
    def __call__(request, poll_id):
        """ Show a Poll, increment the access_counter each time. """
        poll = self.get_objects(poll_id)
        poll = self.process(poll)
        poll.save()
 
        output = self.render(poll)
        return HttpResponse(output)

What did we do here? We separated the functionality into methods. We have four functions now:

• get_objects which gets and returns the data, not more, not less. If you have more then one piece of data to get, return a tuple.
• process which manipulates the data, but does not save it.
• render transform the manipulated data into HTML.
• __call__ the mighty function which ties it all together

Each of these functions do one thing, and try to do it well and isolated. Because this is so isolated, testing becomes quite easy since you can test the output of every step and it does not have side effects of the other parts. You still need to run __call__ at least once to check if it is all nicely tied together.

Testing

Testing a part of this view is simple:

class SimpleviewTest(TestCase):
    def test_simpleview_process_increment(self):
        view = mysite.view.simple_view()
        poll = Poll(question="sample")
 
        poll = view.process(poll)
 
        self.assertEquals(poll.access_counter, 1)

Create an instance of the view, create a poll object and feed it to the function. Check afterwards if the value has increased. This is fast, there is no database interaction!

Wiring it into urls.py

I found a little lambda helper to call the class based view:

# this wraps a class based view in a lambda, so it is thread safe (and I do not need to use __new__)
# yes this returns a lambda
class_view = lambda x: lambda *args, **kwargs: x()(*args, **kwargs)
 
urlpatterns = patterns('',
    (r'showpoll', class_view(mysite.views.simple_view)),
)

Just put the class_view function around the view reference.

This might seem quite a lot of code for this example, and it is, but this will give you some benefits when testing this way. Because this is a lot longer then before, you might not want to use it for every view, just for the more complicated ones.

Vim, python pep8 and pylint.

2 Comments

When you are making your Python code pep8 conformant or want to have a good pylint score and you are using vim, there are a few ways to make it easier and faster.

First you need to know a bit about the vim option -q. -q takes a filename as a argument, and vim then uses this file to jump to locations in files where the used tool has reported something. Vim expects a very simple format here.

Vim can take the output of the pep8 command:

pep8 --repeat --ignore=E501 *.py  > ~/pep.txt

Then start vim with:

vim -q ~/pepepep.txt

Vim then opens with the first reported issue, to go to the next, use :cn.
Or you can temporarily map z to that for this vim session:

:map z :cn<cr>

Like with pep8, you can do the same with pylint, but you need to use the --format parseable argument:

pylint views.py -f parseable > ~/pep.txt

Some regexps for common formatting issues

Remove trailing whitespace all over the file:
:%s/\s\+$//

Remove whitespace in front of a ,
:s/\s*,/,/

Remove whitespace in front of an equal-sign, if any, add a space.
:s/\s*=/ =/

Configuring the Buffalo AP’s as a Bridge

No Comments

Configuring the Buffalo WHR-HP-54G as a bridge is quite easy, given that you can get Tomato firmware on there. The default Buffalo firmware doesn’t support bridging or, I couldn’t find it. only thing I found was WDS bridging which is not what I want.

Install Tomato

You need to install an alternative firmware, fortunately most WRT derivatives are compatible, most notably Tomato and openWRT. I picked Tomato because it has a nice “out of the box” experience. It has an easy to use Web interface which is easy to use. OpenWRT is more powerful in the end, but harder to set up at first. I also assume we do not need the extra features that are possible.

Get Tomato from its website, its 7zipped so you might need a more recent unarchiver.

Flashing the AP is more trail and error then science, and it did not work for me, but did work for a colleague.

Here are the Tomato install instructions

Configure a Bridge

TURN OFF IT’S DHCP SERVER

Give it an IP address in an other subnet then you main one (this is for performance reasons).

Here are the settings you need to set to be able to bridge:

Tomato WLAN settings

This is a nice addition:

Bridge LED

And its done.

Setting up Suse LTSP with DHCP on another server

No Comments

At work, I have been tasked with setting up a Linux terminal server to lighten the load of our exsisting Windows server.

LTSP (Linux Terminal Server Project) is quite a nice piece of software. It makes good use of PXE netbooting, but it should also work with RDP.

Get Suse

Download the openSUSE-Edu-KIWI-LTSP-live-unstable.i686 iso from http://download.opensuse.org/repositories/Education/images/iso/

Why Suse? It has good support for LTSP and RDP.

Setting up RDP on Suse

Setting up RDP on Suse was very easy (compared to Ubuntu LTSP):
You can skip this if you installed from the live DVD.

zypper ar http://download.opensuse.org/repositories/home:/dreveman:/nomad/openSUSE_11.1
# nomad zypper in rdesktop xrdp xorg-x11-server-dmx xorg-x11-server-rdp


Now you need to edit /etc/sysconfig/kiwi-ltsp and add:

USE_NOMAD=”1″

Run to update the configuration:

kiwi-ltsp-setup -c

Now start the xrdp server with:

rcxrdp start

PXE booting

PXE booting is a nice feature of network cards. It allows the computer to pull the operating system over the network, and boot the pulled operating system. The user can start applications, but they are not run on the client, but on the server. This is done with several services.

DHCP to tell where to get the boot image from.
TFTP this actually delivers the boot image, based on UDP.
Xorg to display apps on the client side which are actually run on the server.
SSH to provide a secure connection between the Xorg server on the client and Xorg on the server (using option -X)

This should work out of the box when this is the only server in your network, and you do not already have a DHCP server on your network.

But we already have DHCP

LTSP assumes that you do not have an DHCP server yet and it runs one for you. This is quite bad if you already have a DHCP server running, which we have. Turning the dhcp server on Suse off is done with this:

insserv -r dhcpd && rcdhcpd stop

To turn it off forever you can remove the dhcp-server package:

rpm -e dhcp-server --nodeps

This should work until the next time you run kiwi-ltsp-setup, so I’ve been told.

You need to configure your original DHCP server (not the ltsp) , by editing the dhcpd.conf:

subnet 192.168.0.0 netmask 255.255.255.0 {
option routers 192.168.0.1;
range 192.168.0.1 192.168.0.254;
#add this and replace with your own LTSP server ip:
next-server 192.168.0.2;
filename "pxelinux.0";
}

If you try to PXE boot it now, it will fail somewhere at the startup. Press F2 to see the init messages.

You also need to edit /srv/tftpboot/pxelinux.cfg/default

The line (its under LABEL kiwi-ltsp):
append initrd=boot/initrd-ltsp vga=791 splash=silent showopts

should become:
append initrd=boot/initrd-ltsp vga=791 splash=silent showopts kiwiserver=192.168.1.231 kiwiservertype=tftp

Try booting it now, and it should work.