Moserei

Rails 3.1: Access compiled assets

2011-09-16T00:00:00+02:00

I use wicked pdf for PDF generation in my rails app foxtrot mike. Wicked pdf has a helper method which copies the content of CSS files into the generated HTML. This does not work in Rails 3.1 because CSS files are generated by Sprockets. Fortunately it is possible to access the compiled asset files. When the asset pipeline is enabled, the application object has an attribute assets which returns a Sprockets::Environment.

The following code includes the contents of application.css in my layout which is used for PDF generation.

  %style{ :type => "text/css" }
    = Rails.application.assets['application.css'].to_s.html_safe

Evince with Tabs - Or rather "in Tabs"

2011-02-10T00:00:00+01:00

While I wrote my bachelors thesis a couple of months ago, I really got angry about Evince and the lack of a tab-enabled PDF viewer in Gnome. While I was (trying to) learn for a university course today, this anger returned. The lecturer felt like it was a good idea to arrange the content in more than 15 PDF files.

I ceased the attempt to learn and again started a search for a tabbed document viewer for Gnome. Several users requested a tab feature in evince on Ubuntu brainstorm and received - in my opinion - annoyingly arrogant answers. The technical aspects are not an argument, look at Gedit. The design-user-experience-blah reasons for not implementing this feature may be justified, but one could still implement it as an option and let the user decide. But whatever…

I did not exactly find a tabbed document viewer. But I found an acceptable workaround:

It is based on whatever browser you like (as long as it is supported by mozplugger, Chromium in my case) and embedding evince there. On Ubuntu mozplugger can be installed like so:

sudo aptitude install mozplugger

After you installed mozplugger and restarted your browser it should show up as a plugin. You can then follow the instructions in the blog post linked to above.

Because Evince has no option to hide the toolbar persistently, I removed it completely. To do this, edit ~/.gnome2/evince/evince_toolbar.xml to look like this:

<?xml version="1.0"?>
<toolbars version="1.0">
</toolbars>

I have furthermore configured Gnome to open PDFs in Chromium.

Switched to Jekyll

2011-01-15T00:00:00+01:00

When I first installed WordPress I really liked it. But keeping it and it’s plugins up to date is a pain for me. Furthermore, running a PHP application feels kind of strange for me who feels disgust when thinking of this language. While my server does not run anything of great value, it is surely more secure not to run a dynamic application for something that is rather static.

So I decided to switch my blog to jekyll. The import of my old posts worked quite well, I just had to reformat them a bit. My layout just needed minor adaptions.

Permalinks of my WP blog were formatted according to the pattern moserei.de/index.php/:id/:slug. I could have generated a similar directory structure with jekyll, but I like the default format (moserei.de/:year/:month/:day/:slug). So I crafted a little rack app which redirects requests to old permalink URLs to the new ones.

class RedirectPermalinks
  Mapping = { '0' => "/",
              '3' => '/2008/09/15/started.html' 
               #... 
            }
  def call(env)
    id = "0"
    if env["QUERY_STRING"] =~ /p=([0-9]+)/ || 
       env["REQUEST_URI"] =~ /\/([0-9]+)\//
      id = $1
    end
    [ 301, 
      { "Content-Type" => "text/plain", 
        "Location" => "http://moserei.de#{id && Mapping[id] || ""}" }, 
      [""] ]
  end
end

It is deployed inside a directory called index.php and redirects requests to the abovementioned permalink URLs and to index.php/?p=:id to the corresponding articles. All other requests are redirected to /.

Comments are now managed by Disqus. Disqus has awesome tools for import and migration. I imported comments from a WXR file exported from WP and migrated the post URLs using a CSV file I extracted from WP’s database.

N-Body fun

2010-12-14T00:00:00+01:00

Update: Replaced the GIF by a video.

We currently have to do a simple two-dimensional n-body simulation of gravitation for a university course in practical parallel programming. It uses MPI and OpenMP for parallelism. For benchmarks we were supplied with a model of a spiral galaxy which (in the original example) was passed by by an massive object. I tweaked the system so that this object now rather goes through the galaxy:

video

Some numbers: The model consists of 402 objects, a step of simulation is 1000 years long, the loop consists of 1.4M steps, a frame was generated every 2500 steps. This took about 25 minutes using 14 MPI processes (no OpenMP).

Find GnuPG revocation certificate on unmountable partition

2010-07-06T00:00:00+02:00

When you encounter a hard disk crash like I did recently (my fault - dropped my laptop), one of the most important files to restore is your GnuPG private key ring or at least a revocation certificate for your public key(s). I did not have any backup :-(. But at least I had created a revocation certificate, which was now hidden somewhere on my unmountable partition. Scalpel, a file carving tool, helped me out. There is a Ubuntu package (and I believe there should be packages for most distributions):

sudo apt-get install scalpel

It’s configuration is in /etc/scalpel/scalpel.conf. It offers some GPG related rules, but none for a revocation certificate. I used these rules:

revoc	y	100000	-----BEGIN\040PGP\040PUBLIC\040KEY\040BLOCK-----\x0aVersion:\040GnuPG\040v1.4.10\040(GNU/Linux)\x0aComment:\040A\040revocation\040certificate\040should\040follow
revoc	y	100000	-----BEGIN\040PGP\040PUBLIC\040KEY\040BLOCK-----\x0aVersion:\040GnuPG\040v1.4.9\040(GNU/Linux)\x0aComment:\040A\040revocation\040certificate\040should\040follow
revoc	y	100000	-----BEGIN\040PGP\040PUBLIC\040KEY\040BLOCK-----\x0aVersion:\040GnuPG\040v1.4.8\040(GNU/Linux)\x0aComment:\040A\040revocation\040certificate\040should\040follow

I wasn’t quite sure which version of GPG I had installed at the time I created the keys, so I created rules for all possible versions. (See manpage for syntax explanation.)

It took about 2 hours to scan the image of my 250 GB hard drive and scalpel successfully restored my revocation certificate.

I also tried some of the supplied rules but I think they are not suitable for restoring files from an image of a whole hard drive. (Most of them just look for two bytes that mark the beginning of the file.)

slow, slower, activerecord-jdbcsqlite3-adapter

2010-02-09T00:00:00+01:00

I did a little performance test to find out which database to use in a JRuby desktop application. To have a comparison value I also ran the test (insert, find+update, destroy each 500 times) with MRI and the native sqlite3 driver. It revealed that the jdbcsqlite3 adapter is really sloooooooooow:

$ ruby test.rb sqlite3

                 user     system   total     real
insert           0.4500   0.0900   0.5400 (  7.303116)
find and update  0.4900   0.1400   0.6300 (  6.984390)
destroy          0.3300   0.1100   0.4400 (  7.261199)

$ jruby test.rb jdbcsqlite3

                 user     system   total     real
insert          17.2500   0.0000  17.2500 ( 17.250000)
find and update 16.3270   0.0000  16.3270 ( 16.327000)
destroy         15.0900   0.0000  15.0900 ( 15.089000)

$ jruby test.rb h2

                 user     system   total     real
insert           1.9320   0.0000   1.9320 (  1.932000)
find and update  0.7800   0.0000   0.7800 (  0.780000)
destroy          0.3180   0.0000   0.3180 (  0.318000)

That’s what I call an easy decision. My setup can be downloaded here.

Difference between do-end and {} blocks

2010-02-07T00:00:00+01:00

There is a important difference between the two block syntaxes in Ruby: Their precedence. Consider following code:

def method1(*args)
  puts "method1 got a block" if block_given?
end

def method2(*args)
  puts "method2 got a block" if block_given?
end

method1 method2 do
end

method1 method2 {
}

You would expect both method calls to produce the same result, wouldn’t you? Output:

method1 got a block
method2 got a block

Obviously the do-end block is passed to the first method in the expression while the {} block is passed to the method called directly before it. Another example shows that the assignment does not count as a method call here:

#[...]
class Foo
  def bar=(o)
    puts "Foo#bar= got a block" if block_given?
  end
end
foo = Foo.new

foo.bar = method1 method2 do
end 
#method1 got a block

foo.bar = method1 method2 {
} 
#method2 got a block

Accessing DeviceKit with DBus and Python

2010-01-08T00:00:00+01:00

Under the impression of loosing some really important data because of a damaged partition table on a USB flash drive I am developing a backup strategy for my system. I know myself and so I decided that I need to automate this. While trying to figure out how one can run a script as soon as a specific drive is mounted, I came across DeviceKit. It’s the planned replacement of HAL and is used in Ubuntu Karmic. Udev is not an option for me, because I don’t want to mess around with mounting myself (and I hate running stuff as root…). So here is what I found out about using DBus and DeviceKit in Python:

import dbus

bus = dbus.SystemBus()

proxy = bus.get_object("org.freedesktop.DeviceKit.Disks", 
                       "/org/freedesktop/DeviceKit/Disks")
iface = dbus.Interface(proxy, "org.freedesktop.DeviceKit.Disks")
#enumerates all devices
print iface.EnumerateDevices()

#gets the device kit path of a specific device
path = iface.FindDeviceByDeviceFile("/dev/sdc1") 
#= "/org/freedesktop/DeviceKit/Disks/devices/sdc"

#gets an object representing the device specified by the path
device = bus.get_object("org.freedesktop.DeviceKit.Disks", path) 

#prints some XML that shows you the available methods, signals and properties
print device.Introspect()

#gets a proxy for getting properties
device_prop = dbus.Interface(device, "org.freedesktop.DBus.Properties")

#you need to specify an interface (properties could be ambiguous)
print device_prop.Get("org.freedesktop.DeviceKit.Disks.Device", "device-mount-paths")

#gets a proxy you can call methods on
device_iface = dbus.Interface(device, "org.freedesktop.DeviceKit.Disks.Device")

#unmounts the partition
device_iface.FilesystemUnmount(dbus.Array([], 's'))

The DBus API of DeviceKit is documented here. Now we want to be notified when a drive is mounted:

import dbus
import gobject
from dbus.mainloop.glib import DBusGMainLoop

def device_added_callback(device):
    print 'Device %s was added' % (device)

def device_changed_callback(device):
    print 'Device %s was changed' % (device)

#must be done before connecting to DBus
DBusGMainLoop(set_as_default=True)

bus = dbus.SystemBus()

proxy = bus.get_object("org.freedesktop.DeviceKit.Disks", 
                       "/org/freedesktop/DeviceKit/Disks")
iface = dbus.Interface(proxy, "org.freedesktop.DeviceKit.Disks")

#addes two signal listeners
iface.connect_to_signal('DeviceAdded', device_added_callback)
iface.connect_to_signal('DeviceChanged', device_changed_callback)

#start the main loop
mainloop = gobject.MainLoop()
mainloop.run()

A typical output when a flash drive is plugged in looks like this:

Device /org/freedesktop/DeviceKit/Disks/devices/sdc was added
Device /org/freedesktop/DeviceKit/Disks/devices/sdc1 was added
Device /org/freedesktop/DeviceKit/Disks/devices/sdc1 was changed

With this knowledge I’m currently working on a little python script that runs in background and executes a shell script when a file system is mounted. I’ll post it, when it’s finished :-)

Sequel single table inheritance

2009-10-20T00:00:00+02:00

Sequel’s STI is not so well documented as ActiveRecord’s is. Because it cost me about 20 minutes to find out about it’s syntax, I post a little example.

class A < Sequel::Model
  plugin :single_table_inheritance, :object_type
end

class B < A
end

The second argument is the name of the column to be used to store the class name in DB. Don’t use ‘type’ here, it collides with Ruby’s Object#type. This works with version 3.5.0 and should work with any version >= 2.12.0.

Monkeybars - Little bug

2009-09-10T00:00:00+02:00

For a little side project I am currently evaluating different ways of GUI programming using Ruby. Monkeybars is one of the most interesting candidates. There is a really good article about it, if you want to learn more. Version 1.0.4 has a little, but annoying bug: If you generate a new application skeleton it won’t compile/run:

manifest.rb:32:in `require': no such file to load -- monkeybars (LoadError)
    from manifest.rb:32
    from manifest.rb:21:in `require'
    from main.rb:21

To fix it, you got to change line 21 of manifest.rb:

add_to_classpath '../lib/java/monkeybars-1.0.2.jar'
#to
add_to_classpath '../lib/java/monkeybars-1.0.4.jar'

First time > 500 km

2009-08-13T00:00:00+02:00

On last Wednesday I had to write my last test for this semester. As the long term soaring forecast looked quite good I decided to skip the party on the evening and to drive home (i.e. the airfield) instead. The weather report for Thursday promised us good but blue (no clouds that mark the upper ends of thermals) conditions. I don’t really like flying on blue days as you can never be sure if there is still thermal activity where you are heading, but I was quite motivated because I did not fly cross country for nearly two months. On Thursday morning Daniel and I assembled the Ventus 2cx (Thanks btw) and I put in 58 liters of water ballast. At 11 AM me and some of our guests from Hamburg were ready for tow. Fortunately the weather forecast was not entirely correct: We already saw some clouds develop.

At 11.47 AM (local) I released the tow rope a little north western of Cham in about 700 m AGL (above ground level). My first thermal wasn’t good (or I just did not get it the right way). So I proceeded a little further north, lost some more height and found myself in weak, narrow thermal. I decided to drop my water ballast which probably was wrong. Afterwards the thermal improved and I could fly on to the north. Within the first hour I seldom was higher than 1000 m AGL. When I passed Weiden the altitude of the cloud base and the thermals’ strength improved and I got faster. The Fichtelgebirge provided me with some strong lift and soon I passed Kulmbach.

Sonneberg

Overhead Sonneberg I entered the Thuringian Forest. Some kilometers before Suhl I turned back southward. All in all I had to thermal too much for my taste on the first leg.

Bridge construction site on the Nuremberg–Erfurt high-speed railway

On the way back south I did a better job and did not thermal much until I was south of Tirschenreuth. A Janus joined me in close formation for some 40 Km. In order to stretch the flight a little I went on to the south east of Cham and turned near Bischofsmais.

From there I planned my final glide which was quite tight and got tighter due to some strong downdrafts in the area where I turned. But I was lucky and hit some thermals on the glide home. The second one was quite strong and I decided to gain any height I could get in order to make some extra kilometers. I again passed Cham and flew on to the north west. I could use some calm evening thermals, turned at Neunburg vorm Wald for the last time and went home fast.

With a distance of 529 Km and a duration of 6 hours and 37 minutes this was my farthest and longest flight. Because of the high amount of traffic and the sometimes rough thermals I only took a few pictures today.

Here is the flight on the OLC.

xkcd rocks

2009-07-20T00:00:00+02:00

massage passing as means of coordination

2009-07-01T00:00:00+02:00

A nice quote found on the slides of our distributed systems lecture:

massage passing as means of coordination

One should try massaging each other to improve a team’s coordination. ;-)

ActiveRecord: write_attribute and UTC conversion

2009-06-23T00:00:00+02:00

A little gotcha with custom setters for datetime attributes is, that when setting an attribute through ‘write_attribute’ it is not converted to UTC (or whatever else your default time zone is). This problem can easily be reproduced:

./script/generate model Thing a:datetime b:datetime

My model:

class Thing < ActiveRecord::Base
  def b=(d)
    write_attribute(:b, d)
  end
end

My tests:

require 'test_helper'

class ThingTest < ActiveSupport::TestCase
  test "a and b" do
    d = DateTime.now
    t = Thing.new
    t.a = d
    t.b = d
    t.save
    assert_equal t.a, t.b
    t.reload
    assert_equal t.a, t.b
  end
end

The second assertion will fail. Is this intended or a bug? Quick fix:

  def b=(d)
    d = d.utc
    write_attribute(:b, d)
  end

jQuery, jRails and the Accept header

2009-06-18T00:00:00+02:00

I included jQuery by installing the jRails plugin, which seems nice because you can continue using rails’ ajax helpers. But when I tried to implement a more special functionality by using the $.ajax function it proved impossible to set the Accept header for my request. Neither using the dataType option nor by setting it directly in a beforeSend function. The Accept header always read:

text/javascript, text/html, application/xml, text/xml, */*

I was about to uninstall and hate jQuery for the rest of my life. By coincidence I took a look at jrails.js which is a part of the jRails plugin:

    (function($)
    {
      $().ajaxSend(
        function(a,xhr,s){
          xhr.setRequestHeader("Accept","text/javascript, text/html, application/xml, text/xml, */*")
        }
      )
    }
    )(jQuery); 
    [...]

WTF? Is that documented anywhere? This hard coded ~~shit~~ stuff breaks rails’ respond_to functionality, doesn’t it?t To be fair, it works if you use extensions to determine what datatype you expect. (Like /things/1.js) But I don’t do that when I build a custom ajax request where I can set the Accept header directly.

Update: Some research on the topic “accept header vs. extension” showed that in terms of cross browser compatibility I should favor the extension approach. And I’m not surprised it’s Microsoft’s fault :-)

ActiveRecord: Overwrite attribute setter

2009-06-17T00:00:00+02:00

It’s somehow trivial, but I found out about ’write_attribute’ only just: (another case of rtfm)

class Thing < ActiveRecord::Base
  def name=(str)
    str.upcase! unless str.nil?
    write_attribute(:name, str)
  end
end

Update: Be careful if you use this with Datetime attributes.

Force a USB device to use OHCI

2009-04-21T00:00:00+02:00

As I described in my previous post I built a cheap USB wireless adapter into my NSLU2. It finds my network and connects without problems. But when I transfer bigger loads of data, it sometimes stops working. The log showed that it disconnects from the USB hub. Other people experiencing similar problems indicated that this was a problem with the ehci_hcd kernel module and “solved” by removing this module. But thats definitely not a good idea when your only mass storage device is connected via USB too.

So I needed to disable the EHCI for the wireless adapter only. That should have no big impact because I don’t think my network really offers speeds above 12 Mbit/s. Fortunately there is a script by Ernestas Vaiciukevičius that does exactly this.

Update: Seems like the link is dead. You can find the script here. Usage: Create a udev rule that runs it:

SUBSYSTEM=="usb", ENV{PRODUCT}=="xxxx/yyyy/*" RUN+="/usr/local/bin/force_ohci.sh"

xxxx and yyyy can be found in lsusb’s output:

$ lsusb
Bus 001 Device 001: ID xxxx:yyyy Device name
[...]

For testing reasons you can also run it from the command line:

$ ./force_ohci.sh "xxxx:yyyy"

Wireless NSLU2

2009-04-20T00:00:00+02:00

I bought a NSLU2 some weeks ago. I will use it as printer and scanner server and as an ever running torrent client. As I don’t want to run a ethernet cable through the walls of our flat, I need wireless network on it. There is a nice HowTo on NSLU2-Linux. I got my hands on a wireless usb stick for 15 €, warmed up the soldering iron and got started: You really need a steady hand for soldering onto these tiny contacts.

But to my own surprise I managed to:

Inside the case of the NSLU2 there is enough space, so i just taped the new part on top of some capacitors:

5 minutes later I was really glad to see the device in lsusb.

Photos online

2009-04-14T00:00:00+02:00

I just finished uploading some photos taken during my recent flights. See them here. And I got a new hobby.

Multi-line comments in ruby

2009-04-07T00:00:00+02:00

I keep forgetting how to do them, so I post myself a little reminder here.

Just put

=begin
...
=end

around the stuff to be commented out. There must not be whitespaces before these keywords.

dm-optlock now on github

2009-03-03T00:00:00+01:00

dm-optlock is on github now. It can still be installed as gem:

gem install moser-dm-optlock

If you haven’t done before, you need to add github to your gem sources:

gem sources -a http://gems.github.com

dm-optlock v 0.1.4

2009-01-10T00:00:00+01:00

Update: dm-optlock now on github

I released a new version of dm-optlock. It is NOT compatible with v 0.1.2’s code, because replaced the set_locking_column method by a method called add_locking_column. Your DB should be compatible.

class Model
  include DataMapper::Resource
  ...
  add_locking_column :my_version_col
end

It takes a options hash just like DataMapper’s property method does. I added it to enforce some options on the locking column.

Yahoo! or better Oh-no!

2008-12-21T00:00:00+01:00

I forgot my Yahoo!-ID for my flickR account. I know it’s nothing to be proud of… But what really annoys me is what Yahoo calls their customer support. I sent in their contact form with a description of my problem which I though to be easy to solve. One day later (!) they replied by a standard email describing how to get your lost password (!) and what mistakes one can make when typing one’s ID.

That was on Wednesday. Since then I got the same email four times signed by four different employees. Each time I replied with a description of my problem and ~24 hours later the same dumb email reached me.

In the meantime I already found out my ID using the english help pages (the german copy doesn’t seem to be complete), but I think I will continue playing mail ping pong with Yahoo for some time.

shared git repo + ssh = umask problems

2008-12-19T00:00:00+01:00

When you want to share a little git repository on your own server the first thing that comes to your mind is SSH. It’s installed virtually everywhere, secure and easy to configure. In the beginning there are no problems, because everyone can clone your repo and if they don’t change files concurrently even pushing works. About one day latter you get umask problems and start to search the web for solutions. To make it short, here it is:

git repo-config core.sharedRepository true

This option makes git care about the permission stuff. I wasted about 10 hours of my lifetime for this one and found an odd thing about ~/.ssh/rc:

umask 002
date > ~/sshrc-executed

I added the second line, because I wasn’t sure if it was executed. It was (look at the file permissions):

...
-rw-rw-r--  1 moser git     29 Dec 19 18:41 sshrc-executed

But my umask didn’t last:

moser@HellAgent:~$ ssh xxx.net -n 'umask'
Password: 
0022

Don’t know why, if anyone knows tell me please.

Waste Separation 2.0

2008-11-03T00:00:00+01:00

I just went through the pictures I took during our holiday in Barcelona and found this one:

To be fair I got to say that all the other bins had 3 bags in them :-)

How to use IGC data to geotag your photos

2008-10-07T00:00:00+02:00

When taking photos on a cross-country flight it’s not always easy to remember where they were taken. And wouldn’t it be cool to look at your photos on flickr's map? Either you buy a camera with an integrated GPS receiver which automatically stores a geotag in your picture file’s meta data or you use GPS data that’s gathered anyway: Your IGC files. What do you need?

Both programs are available for both Linux and Windows. My article is focused on Linux but the commands on windows will be quite similar. If you are running Debian or Ubuntu you can install these via

apt-get install gpscorrelate gpscorrelate-gui gpsbabel

The GPS Photo Correlator accepts GPS data in the GPX format. So you will need to convert your data into that format:

gpsbabel -i igc -o gpx <infile>.igc <outfile>.gpx

Now lets geotag your files:

gpscorrelate -g <gpxfile> -z +2:00 --show file(s)

“-z +/-XX:XX” is needed because the timestamp on your photos is probably not UTC. file(s) is either a list of files (“DSC2131.JPG DSC2322.JPG”) or a wildcard expression (”” for all photos in the directory or something like “DSC12.JPG”). If you add ”–no-write” no EXIF data will be written and you can check if your photos would have been tagged right.

If you want flickr to use your photos’ EXIF geotags you have to change your account settings.

Optimistic Locking for Datamapper

2008-10-01T00:00:00+02:00

Update(2009-3-3): dm-optlock now on github

Update(2009-1-10): New version

I should be learning for my test on next tuesday, but coding helps keeping me in a good mood. Here is a little gem I will need in a future project for which I’ve choosen merb + datamapper. Datamapper doesn’t support any row level locking. My code checks if the record you want to save was changed since you loaded it. If it was changed it raises an exception (DataMapper::StaleObjectError). It works quite like ActiveRecord’s optimistic locking. You need to define this column in your model you want to be locked:

property :lock_version, Integer, :default => 0

Anywhere in your code where you update your objects you should be prepared to handle the exception:

  if @obj.update_attributes({:dummy => 123})
    #successful
  else
    #validation errors?
  end
rescue DataMapper::StaleObjectError
  #tell the user that he was too slow
  #or even better: show him the conflicts and let him merge the changes
end

The gem can be downloaded here: dm-optlock-0.1.0.gem Please let me know what you think of it. I’ll try and put it on RubyForge, too.

(00:27) Little update: I added a ‘set_locking_column’ method. dm-optlock-0.1.2.gem

Update: [dm-optlock is on Rubyforge](http://rubyforge.org/projects/dm-optlock/). Now you can get it like this: gem install dm-optlock

Stable Array Sorting in Ruby

2008-09-18T00:00:00+02:00

Ruby uses some kind of Quicksort to sort its arrays. Unfortunately Quicksort is not stable. When you get sorted data from your database, which almost certainly sorts faster than anything in Ruby, you sometimes may want the data to be resorted in a stable way.

I found following snippet:

class Array
  def stable_sort
    n = 0
    sort_by {|x| n+= 1; [x, n]}
  end
end

I extended it to accept a block like the original sort method does:

class Array
  def stable_sort
    n = 0
    c = lambda { |x| n+= 1; [x, n]}
    if block_given?
      sort { |a, b|
        yield(c.call(a), c.call(b))        
      }
    else
      sort_by &c
    end
  end
end

Mergesort in Ruby

2008-09-18T00:00:00+02:00

While playing around with stable sorting I implemented Mergesort.

class Array
  def mergesort(&cmp)
    if cmp == nil
      cmp = lambda { |a, b| a <=> b }
    end
    if size <= 1
      self.dup
    else
       halves = split.map{ |half|
        half.mergesort(&cmp)
      }
      merge(*halves, &cmp)
    end
  end

 
  protected
  def split
    n = (length / 2).floor - 1
    [self[0..n], self[n+1..-1]]
  end

  def merge(first, second, &predicate)
    result = []
    until first.empty? || second.empty?
     if predicate.call(first.first, second.first) <= 0
        result << first.shift
      else
        result << second.shift
      end 
    end
    result.concat(first).concat(second)
  end
end

It is not suitable for productive use as it’s quite slow. It’s about ten times slower than the stable_sort method from my another post. A little benchmark sorting an array of 2,000,000 random integers:

                user     system      total        real
Original Sort 0.630000   0.010000   0.640000 (  0.639860)
Stable Sort  12.780000   0.700000  13.480000 ( 13.477982)
Mergesort   109.170000  13.420000 122.590000 (122.617323)

Advanced Array Sorting in Ruby

2008-09-16T00:00:00+02:00

While developing a member management application for my soaring club I found out quite a little about sorting arrays in Ruby. Rubys sort uses the operator <=>:

irb(main):001:0> 1 <=> 2
=> -1
irb(main):002:0> 1 <=> 0
=> 1
irb(main):003:0> 1 <=> 1
=> 0

By implementing it on your class you can specify a way in which an array of instances should be sorted. Here is a fragment of my Member class:

def <=>(b)
  name <=> b.name
end

My members will be sorted by their name. But sometimes you need a special sorting. Thats why the sort method accepts a block to specify a way of sorting

arr.sort! do |a, b|
  a.birthdate.yday <=> b.birthdate.yday
end

This is the sorting I needed for a list of birthdays. Sure enough you can do anything inside the block as long as it returns -1, 0 or 1. It’s no problem to sort by 2 attributes:

arr.sort! do |a, b|
  r = a.birthdate.yday <=> b.birtdate.yday
  r = a.name <=> b.name if r == 0
  r
end

Started!

2008-09-15T00:00:00+02:00

The rainy weather should actually encourage me to study analysis, but it rather made me go through old email. I found one reminding me of this domain I registered some time ago, which I never really used. So I thought I could finally join the community of bloggers and kill some time by installing Wordpress and playing around.