After returning from vacation I was confronted with a really strange problem on my laptop. Firefox wasn’t any more able to load pages from certain domains. After a reboot it worked, but sooner or later it stopped working.

At first I suspected a network problem, but my Windows PC was still able to load those pages.

I played around with the settings in /etc/nsswitch.conf. This made nslookup work correctly, but Firefox and wget etc still wouldn’t work. Very strange – you’d think that they would resolve a name in the same way as nslookup or host would do.

Eventually I found a solution: Disabling the DNS proxy in the Samba configuration file (setting ‘dns proxy = no‘ in /etc/samba/smb.conf) did the trick.

I had some problems with my fingerprint reader for quite some time, probably since upgrading to Jaunty. It took multiple attempts to read my fingerprint, and often it wouldn’t succeed at all. So I had stopped using it.

Today, while I was researching a different problem I found a solution: upgrading the Thinkfinger libraries from ppa helped. I added

deb http://ppa.launchpad.net/jon-oberheide/ubuntu jaunty main

to my /etc/apt/sources.list file and updated the Thinkfinger libraries, and it started working again!

Recently I updated my laptop to Ubuntu 9.04 Jaunty.

It went pretty smooth; the only drawback was a problem with my graphics card Intel GM965: because there are known problems it is blacklisted, so Compiz doesn’t work anymore. And because of refactorings in the graphics driver some things don’t work so well anymore with the kernel included in Jaunty. But I got it working: I installed a newer version of the graphics driver xserver-xorg-video-intel. I edited the blacklist file and removed my graphics card from there.

A few days later I also changed my xorg.conf to use the UXA acceleration method. With this change Google Earth eventually works really well without displaying only flicker.

Another problem I had (again) was that Xorg was using way too much CPU. I remembered the problems I had after one of the last upgrades and removed/renamed my ~/.gconf directory. That solved the problem. It is annoying that this seems to happen after every upgrade.

I also had to update the Y-viewport positions in the ‘Place Windows Compiz Plugin’. While the X-positions had changed from being 0-based to 1-based with the last upgrade, the previous version was still happy to have 0-based Y-positions. But this changed now…

The good thing is that multi-monitor support really improved with Jaunty. The patches that I had to apply manually previously are all included. Nice.

On Microsoft’s .NET implementation on Windows a Debug.Fail() (which is equal to a Debug.Assert(false)) when run on a debug build displays a dialog box with the options Abort, Retry, Ignore. When run in the debugger it also displays the call stack.

On Mono however no dialog box shows; it behaves the same as when the user pressed ‘Ignore’. What’s worse is that it doesn’t display the call stack by default, even when run in debugger. This doesn’t give any hint if an exception happened.

But: some research showed that there is an environment variable MONO_TRACE_LISTENER that can be set that outputs the call stack to wherever:

export MONO_TRACE_LISTENER=Console.Out

In MonoDevelop you can set the environment variable in Project/Options/Run/General.

Some background information can be found in the Mono bug #317040.

And even better: if you create a .config file for your app and set the assertuienabled attribute to true, you get the same dialog as with .NET (discussion in bug #430477):

File app.config:
<?xml version="1.0" encoding="UTF-8"?>
<configuration>
    <system.diagnostics>
        <assert assertuienabled="true" />
    </system.diagnostics>
</configuration>

But there are still some differences to .NET: if the message box displays we don’t get a call stack in the output. I guess it might be possible to add another TraceListener (e.g. ConsoleTraceListener) that takes care of that.

Quite a while ago I investigated some problems we were having with Chinese IME with one of our applications. Since that was new stuff for me I made some notes but never came around to finish this blog entry.

IME (Input Method Editor) – allows input of complex scripts like Chinese.

TSF (Text Services Framework) – introduced with XP. Will eventually replace the traditional IMEs.

An application needs some extra code in order to make use of TSF. However, XP SP1 introduced an option on the OS level to make use of TSF even in non-TSF-aware applications. Vista extended that even more.

This means if an application allows to enter text with an IME it doesn’t say anything whether or not it supports TSF.

In one of our applications (Flex) Chinese IME behaves differently on Vista than other applications, e.g. it doesn’t show a horizontal menu that gives you choices. It works on XP.

Involved in the Chinese IME is IMM32.dll in Windows\System32 directory. It turns out that on Vista this dll is implemented using the TSF – it references MSCTF.DLL.

TSF has (at least) two sides: probably the most common use of the TSF interfaces is to implement a new text service that might provide an additional input method. There are some samples in MSDN that show how to do that.

The other side is the client side. An article in MSDN magazine suggests that you get automatic TSF support if you use plain edit controls, RichEdit controls, HTML editor controls, or the app is based on Windows Presentation Foundation. This means that most people don’t have to implement this side. The mentioned MSDN magazine article shows an example how to do it.

The TSF aware blog has some information and explains that you have to implement at least the ITextStoreACP interface.

One of the problems we’re having on an XP machine was that the Chinese IME didn’t work anymore after we went to certain dialogs. The user had to restart the app. I found a workaround: Add a DWORD value to the registry entry HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\CTF\KnownClasses\<RealClassName> = 1. You can find the real class name e.g. with Spy++ by clicking in the Note edit box. On my machine it’s “WindowsForms10.Window.8.app.0.2004eee”. Unfortunately it seems that the class name changes either with different builds or on different machines.

However, this workaround didn’t seem to work in all cases – don’t know why.

.NET has built in support for IME and supports it for all controls that are built in to .NET. However, ContainerControl (and derived classes) can’t receive IME input by default, so IME support is disabled for them. Before .NET 2.0 SP1 the Control.CanEnableIme property used to be internal, so there was no chance to make a control derived from UserControl (which derives from ContainerControl) IME-aware.

With .NET 2.0 SP1 Microsoft fixed this and made CanEnableIme protected so that it can be overridden.

Miguel Lacouture describes some of the things that are going on in a forum message. Microsoft describes this problem in a KB article (934197). However, the hot fix they are announcing seems a little dated…

After overriding CanEnableIme things work much better, but there are still the following issues open:

• On XP: With the QuanPin IME, if you go to a different field, it switches to English Input Mode (which you can see by the A in the QuanPin toolbar in the lower left corner of your screen).
• On Vista: With Pinyin it switches to English Input Mode when you go to a different field.
• On Vista: With QuanPin it displays the candidate window, but it doesn’t write any text to the field when you select a candidate.

Vista introduced a new kind of IME, Table Driven Text Service, that QuanPin 6.0 is based on. Michael S. Kaplan explains how it works in his blog.

Yesterday I managed to build MonoDevelop from SVN source. I had done it before but had forgotten how to do it, and there are no fool-proof instructions on what you need to build.

I compiled mono from SVN and installed in parallel to Ubuntu’s 1.9.1 version, following the instructions from http://www.mono-project.com/Parallel_Mono_Environments and http://www.mono-project.com/Compiling_Mono_From_SVN.

MonoDevelop has some dependencies that I also had to build: debugger, mono-addins, gtk-sharp, gtksourceview-sharp.

To build everything I did the following steps:

Mono debugger

cd ~/mono-svn
svn co svn://anonsvn.mono-project.com/source/trunk/debugger
cd debugger
./autogen.sh --prefix=$MONO_PREFIX
make
sudo make install

Mono addins

cd ~/mono-svn
svn co svn://anonsvn.mono-project.com/source/trunk/mono-addins
cd mono-addins
./autogen.sh --prefix=$MONO_PREFIX
make
sudo make install

GTK#

cd ~/mono-svn
svn co svn://anonsvn.mono-project.com/source/trunk/gtk-sharp
cd gtk-sharp
./bootstrap-2.14 --prefix=$MONO_PREFIX
make
sudo make install

gtksourceview-sharp

cd ~/mono-svn
svn co svn://anonsvn.mono-project.com/source/branches/gtksourceview-sharp-0.x
cd gtksourceview-sharp-0.x
./autogen.sh --prefix=$MONO_PREFIX
make
sudo make install

Monodevelop

cd ~/mono-svn
svn co svn://anonsvn.mono-project.com/source/trunk/monodevelop
cd monodevelop
./configure --prefix=$MONO_PREFIX
make
sudo make install

This puts monodevelop in $MONO_PREFIX/bin/monodevelop
where I can run it. However, when I start MonoDevelop I get a message box saying that an add-in with an empty path could not be started (and in the details: Could not load type ‘MonoDevelop.Platform.GnomePlatform’ from assembly ‘GnomePlatform’…). I haven’t figured out why I get this error, but if I click Yes to continue everything seems to work.

The first time configure is run in the monodevelop directory, it gives the option to select the desired plugins. Subsequent runs of configure simply re-use that choice. To pick different plugins you can run

./configure --prefix=$MONO_PREFIX --select

Today I upgraded my Dell laptop to Intrepid. After downloading 1.3 GB of data it started the installation and asked about several configuration files it thought I had manually changed. Besides that everything went uneventful, and a reboot also worked.

However, not all things are working:

• Compiz is disabled
• Fingerprint reader no longer working
• Cursor keys not working; up-arrow key opens screen shot app

1. Cursor keys not working

I googled for the problem and found several bug reports related to this. However, the workarounds suggested in the reports didn’t work for me.

Eventually I found the reason: I had mapped some keys and created my own xmodmap.conf. This seems to conflict with Intrepid.

I renamed my ~/xmodmap.conf file to xmodmap.conf.old and logged out and in again. A dialog came up asking if I want to load xmodmap.conf and telling me that a file xmodmap.conf.old is available. I removed the xmodmap.conf file from the list of files to load – and then my cursor keys worked again!

2. Compiz not working

I played around with it and tried several things and last night I suddenly was able to turn it on again (selected “Normal” in Visual Effects in System/Preference/Appearance). It might have to do with the fact that my laptop was undocked and no external screen connected.

In the docking station with the external monitor connected the background doesn’t draw properly on the second screen – I had to reapply the mesa patch described earlier. xserver-xorg-video-intel is included in a version in Intrepid that already contains this change. And my modifications in xorg.conf were also still there.

After installing the newly generated packages the background worked correctly.

But then it seems like some .gconf files were corrupt – when I tried to log in it brought me immediately back to the log-in screen. So I renamed my ~/.gconf directory (which creates a new one if I log in again) and then selectively copied back some of the settings directories.

3. Fingerprint reader no longer working

It turns out that it is working. However, it requires to press the Enter key after swiping the finger (see bug 256429). After installing the patched packages linked from the bug report it works now.

But the problem mentioned there with this patch is also true: if I enter my password manually instead of swiping my finger it still sends two carriage returns.

4. Other

I noticed that the viewport positions in the Place Windows Compiz Plugin are now 1-based instead of 0-based. All of my windows I had generated rules for to position on a specific workspace came up on the wrong workspace. After I modified the viewport positions it worked again.

Today I was looking for a screen recording application on Linux. I came across recordmydesktop (and it’s GUI gtk-recordmydesktop) which looked very promising. It is included in the Ubuntu repositories and it is easy to use: just select the window you want to capture and click on the record button. However, when I looked at the generated video the characters in the recorded window looked unacceptably distorted.

I thought that it’s probably related to the size of my window, so I tried different sizes but it didn’t get any better. Eventually I tried the video in a different movie player – and voilà, in VLC it looked perfect regardless of the size of the recorded window. So I had a look at Totem again and found: if I turn off deinterlacing (View/Deinterlace) it also works there.

After I upgraded to the current kernel version (2.6.24-21) the sound suddenly stopped working on my Dell D830 laptop running Ubuntu Hardy 8.04.

Today I tried to solve it and found the solution on this page (http://linux.dell.com/wiki/index.php/Ubuntu_8.04/Issues/No_Sound_After_Distribution_Or_Kernel_Upgrade). Although it speaks of kernel 2.6.24-17 it also applies to 2.6.24-21.

Here are the steps I had to do (copied from the mentioned site):

1. Restore name of sound driver that old modem driver renamed:

   $ cd /lib/modules/2.6.24-21-generic/ubuntu/sound/alsa-driver/pci/hda
   $ sudo mv snd-hda-intel.ko.REPLACEDBYhsfmodem snd-hda-intel.ko

2. Remove old sound drivers from kernel module directory:

   $ cd /lib/modules/2.6.24-21-generic/updates
   $ sudo rm snd-hda-intel.ko
   $ sudo rm snd-hda-codec.ko

3. Rebuild initrd:

   $ sudo depmod -a
   $ sudo update-initramfs -u

4. Reboot

I’m running Vista in a VirtualBox virtual machine that I have installed on Ubuntu 8.04. This all worked great until I recently enabled the support to enter complex characters in order to be able to test how some software behaves when using an IME. This morning when I started my virtual machine, it didn’t respond to any keyboard input.

Enabling the complex character support installed SCIM. A quick Google search revealed a nice solution for the problem: unsetting the variables QT_IM_MODULE, XMODIFIERS, and GTK_IM_MODULE before you start VirtualBox prevents VirtualBox from communicating with Scim (which is what I want anyways since the VM itself might have its own IMEs).

Since I usually start VirtualBox from a launcher in the panel, I opened the properties of the VirtualBox launcher and replaced the command with

/bin/bash -c "unset QT_IM_MODULE && unset XMODIFIERS && unset GTK_IM_MODULE && virtualbox"

Now keyboard input again works!