Bugzilla Fables: Serving Two Masters

You may know this: spending a few hard days, or even minutes, figuring out a bug and in the end the result is a one line change. What an injustice! So let's expand one such case for the amusement and education of the audience. Fable #501758:

For certain tasks, YaST uses external helpers (agents) communicating via pipes. One of them was causing YaST to hang when cleaning up. Actually no, it was two of them. Either one was not enough, but if two of them had been used, they would both keep on running and YaST would not exit.

It turns out the bug is in YaST not closing unnecessary file descriptors before spawning the agents. So the second child inherits a copy of the pipe that controls the first child. Then YaST closes the first child's commands pipe, expecting it to see EOF and exit. But the pipe is still open in the second child, uselessly. And they all wait. And wait...(*)

Moral: close unnecessary FDs before exec.

If you wonder why we did not see the problem earlier, it is because both agents contained another bug, a banal one of not responding to an explicit Exit message.

*) Actually, only YaST wait(2)s. The other two read(2) ;-)

Reducing Unprioritized Bugs

This Greasemonkey script, Check Priority, will check whether you are modifying a bug assigned to you and leaving its priority at "P5 - None". It will pop up a confirmation box to remind you to set a real priority.
Thanks to Milan Vančura for showing us how b.n.c can be improved (check out also the local time script) and for even implementing this one as I stood behind his back!

Disabling the fetchmsttfonts update

I learned about a new icon of the update applet yesterday, the blue one for optional updates. I wanted to say "I don't want this particular optional update, thank you". But how?
Short: to disable the fetchmsttfonts update, remove pullin-msttf-fonts.

YES # zypper rm pullin-msttf-fonts

Long:
I tried to add a lock on the patch, but it did not seem to have any effect.

NO  # zypper addlock --type=patch fetchmsttfonts     # or al -t patch

After looking in the raw update metadata (updateinfo.xml.gz), I found the solution: remove the package pullin-msttf-fonts, with the first mentioned command.
Curiously enough, also other guesses at adding locks do not work, in a different way:

OK  # zypper al fetchmsttfonts

This adds a lock for the new package that fetches the fonts. The lock actually does what I want, in a different way: it installs the patch but omits the new package. It is also the most obvious command.

NO  # zypper al pullin-msttf-fonts

This is different in that pullin-msttf-fonts is already installed. List-updates still lists the patch as needed but halts on a solver conflict when trying to install it. The locks seem to take effect only when committing the transaction, not when checking for the updates.

But still, I would like to have a clickety way for this. Am I spoiled?

YaST UI table usability

I've just watched my wife delete a repository in the YaST Qt UI: She right-clicked the table item, nothing. She pressed Delete, nothing. Only then she noticed the Delete push-button below the table.
Obviously we can improve the usability: Delete should be handled the same way as a function key so that YCP receives the `delete event. That would be quite easy. Maybe the context menu is doable too.

BbOoOoTt: Parallel Dual Boot

On my laptop I have a work partition with a stable release and a test partition with a development release. Here I want to show you how I set it up to work with both of them at once, using VirtualBox.

The idea is that if virtualization enables you to run an additional machine on an additional (virtual) disk represented by an image file, it should also be able to operate on the actual disk and in this way convert dual booting from serialized to parallelized. And indeed VirtualBox can do it and the documentation is in its manual, chapters 9.9 "Using a raw host hard disk from a guest" and 9.9.2 "Access to individual physical hard disk partitions".

Prerequisites

Caution: accessing one file system from two machines at once is a sure way to corrupt it and lose data. VB gives you some protection but you still have to be careful. One thing, you must not share a swap partition between the two systems. Also, make sure that you do not have the other system's partition mounted, especially as a "quickie" on /mnt.

Setting up

Change this according to your needs:

DISK=/dev/sda
SWAP=5
ROOT=6

Give yourself access rights to the other system's partitions (LOGNAME is preset by bash):

sudo setfacl -m u:$LOGNAME:r $DISK
sudo setfacl -m u:$LOGNAME:rw $DISK$SWAP $DISK$ROOT

Copy the bootloader of the other system to a file:

dd if=$DISK$ROOT of=/tmp/boot.bin bs=512 count=1

Create a virtual disk to refer to the other system's partitions:

VBoxManage internalcommands createrawvmdk -register \
-filename ~/.VirtualBox/VDI/other-system.vmdk \
-rawdisk $DISK -partitions $SWAP,$ROOT \
-relative -mbr /tmp/boot.bin

That's quite a mouthful so let's go over the options:

• register is a convenience setting that attaches the created disk to the VB disk manager.
• filename is the virtual disk being created and it must be an absolute path.
• rawdisk is the device of the whole target disk.
• partitions restricts the access to these partitions. Other partitions will be visible but appear to contain only zeroes.
• relative means that we will be able to drop access rights to the whole disk (see the next command) and retain only those to the used partitions
• mbr specifies bootloader code to be used (partition info is still taken from the real MBR). That is important since in my case the real MBR references GRUB's menu.lst from sda1. That partition will appear as empty and GRUB would fail with "Error 17". So I give VB the boot record of the other system which references menu.lst on its own, visible, partition.
  

Then give up rights we no longer need:

sudo setfacl -x u:$LOGNAME $DISK

Now create a new machine in VB and attach the vmdk to it.

(there is a CLI way but I just clicked in the GUI)

Running it

Make sure that the other partitions are not used: (Hmm, VB should do that by itself. Must file a bug for that...)

sudo swapoff $DISK$SWAP
sudo umount $DISK$ROOT

Click Play and have a lot of fun.

If you used the defaults when partitioning, you will not get very far because the initrd will not find the disk. The bootloader refers to it by its ID but it is no longer /dev/disk/by-id/ata-TOSHIBA_MK8032GAX_Y6EAT0PKT-part6 but rather .../ata-VBOX_HARDDISK_VB22889591-0c34ac94-part6. The workaround is to edit the GRUB menu on the fly (ESC, Enter, (E)dit, ..., (B)oot) and to replace the long strings by the short names like /dev/hda6. The fix is to use paths from /dev/disk/by-uuid.

Fork!

Osmarender Tweaks

I have made a couple of simple tweaks to the Osmarender rules and styles: Red Arrows for Oneway Streets, Residential Buildings, Hiking Trails. Images included!

TODO: make it easy to render it yourself.

Coordinates:

50.1375; 14.5120

OpenStreetMap Diaries via a GeoRSS Box Filter

OpenStreetMap has this nice feed of its users' diaries, so that you can get all excited about the progress that fellow mappers have done near you. Now if you actually try to follow that feed, you may notice, as did I, that the problem is in the word near, which is the missing piece. It does not help to learn about new developments on another continent.
Fortunately (and unsurprisingly) the feed is GeoRSS so we can filter its contents based on the coordinates supplied with most diary entries. I did not find such service on the Web so I wrote it myself: a GeoRSS Box Filter. You say geofilter.php?url=...&minlat=...&maxlat=...&minlong=...&maxlong=... and it filters the given feed according to the given bounding box. If you omit the url, it defaults to the above mentioned OSM diary feed, and the bounding box defaults to Czechoslovakia.
More examples? How about the United Kingdom? or Georgia? (The Export tab is handy to get the numbers, BTW) To watch for uploaded GPS traces, use geofilter.php?url=http://openstreetmap.org/traces/rss. I wonder what other interesting feeds you can find to experiment with (but I did not try anything else than RSS 2.0).
Oh, and I worked on this all on the company time because it's Hack Week₃ now, yay!

cnetworkmanager 0.7.1

Cnetworkmanager is a command-line client for NetworkManager, intended to supplement and replace the GUI applets. So far it is a single python script. What is new in version 0.7.1:

• it does not need a configuration file anymore:
  cnetworkmanager -C publicnet
cnetworkmanager -C myessid --wep-hex 112234445566778899aabbccdd
cnetworkmanager -C another --wpa-psk-hex \
 112233445566778899aabbccddeeff00112233445566778899aabbccddeeff00
• it works with NM 0.6 (tested on Ubuntu 8.04 Hardy and a pre-release OLPC) in addition to the older support for NM 0.7pre (tested on openSUSE 11.0)
  

What is still left to do:

• sooner:
• specifying the key as a non-hex passphrase
• reading the configuration stored by the GNOME nm-applet
• possibility to quit after a connection is established
• later:
• more encryption schemes (WPA2?)
• more connection types (dial-up, VPN)

Cultural Exchange 4: First YaST

Managed to compile and run YaST up to nfs-client and ntp-client. Of course it does not work properly, but makes for a nice Potemkin hamlet.

Fresh libzypp needs cmake 2.6 but there is only 2.4 in .debs. Fortunately the cmake authors provide a binary tarball with 2.6, only one has to know where to put the contained directories.

Notes for clean-up: Libzypp should say it needs hal-storage and boost-filesystem too. Python-bindings should check for python-dev, should use python-config. Perl does not seem to respect vendor_arch for ycp.pm?

Cultural Exchange 3: First .deb

Started to build YaST. Managed to compile yast2-core. Now going to make it repeatable by properly packaging the dependencies. The first one is blocxx. Bleeding edge stuff is in the Build Service: home:mvidner:debian. The post How to make debian standard debs from scratch was much helpful.

Tried to print and failed miserably. The setup tool would not work with a remote CUPS printer, kept asking for my password. Switching AppArmor from enforcement to complain mode did not help, nor did disabling it.

The development package for openssl is libssl-dev (openSUSE: libopenssl-devel). http://packages.debian.org/ helps with that.

Perl packages, named perl-Foo-Bar on openSUSE, are usually named libfoo-bar-perl on Debian.

Packaging system started to complain that packages cannot be verified.

(Cultural Exchange is my long term project of working with Ubuntu instead of openSUSE. TODO: make a permanent index page.)

YaST Workshop 2008

We had the annual YaST workshop last week. The people in the team boldly went to explore new APIs and one of the results is experimental integration of PolicyKit into YaST. If you really want to break your system, you can try the current code.

Cultural Exchange 2

Gnome would crash before the panel would show up. Also, the text console did not work (graphic artifacts instead). Failsafe Gnome did work, which does not have Compiz, so I took a hint to change the graphics driver. System / Management / Hardware drivers offered me to install the non-free xorg-driver-fglrx and now it works.

command-not-found is enabled out of the box and it rules. openSUSE should have it by default too.

dpkg phrasebook:

rpm -ql foo

dpkg -L foo

rpm -qa "foo*bar"

dpkg -l "foo*bar" (and it also shows packages that are not installed)

Cultural Exchange: First Blood

I decided to expand my horizons and install Ubuntu on my main work computer. So today I offer you the first notes from my observations.

mutt somehow stopped tracking new mails. Fortunately :set check_mbox_size on works around that.

My subversion working copy from openSUSE was created with svn 1.5 but Hardy has only 1.4. So I manually got debs from the coming Intrepid release, and luckily 3 of them were enough: dpkg -i {subversion,libsvn1,libneon27-gnutls}*.deb from here and here.

LinDevDoc Got Tagging

Bluebear has improved LinDevDoc (original announcement) so that it can do tags now.

Addressing Lines in Plain Text Files

Jirka suggested a possibility to enhance YCP documentation with real-life code examples. For that, we'd need to point to numbered lines within the source code. I thought that ViewVC can do it but apparently not (maybe annotate has it, but it is broken at our site). But there is a new generic way: RFC 5147: URI Fragment Identifiers for the text/plain Media Type. Only none of the browsers seems to support it today...

This points to line 13, with the prototype of main:

http://svn.opensuse.org/svn/yast/trunk/libyui/examples/HelloWorld.cc#line=,13

Medoosa in OBS

Medoosa is a documentation tool for C++ that can produce UML class diagrams including generalizations and associations. Corrections can be made interactively in a diagram editor (Dia) and are fed back into the source as Javadoc-style comments. At this time, the layout must still be done by hand. Some code is already present to produce the layout with Graphviz.

I wrote it for my master's thesis at the Charles University back in 2001 and it has been since then quietly sitting at SourceForge. Today I have rebuilt it in the openSUSE Build Service: Medoosa, CcDoc-0.7a (required).

cnetworkmanager 0.4

I pulled together 0.4 yesterday:

• New: changed license to GPLv2 or later, to match other parts of NM
• New: distinguish WEP, WPA, WPA2 for -n.
• New: -o, -w control online and WiFi status.
• New: basic device info for NM 0.6
• Fix: recognize Notification Messages in knetwormanagerrc.
• Fix: -u: better error message when applet not running.
• Fix: standardized option parsing, both -Cfoo and -C foo work.

The software is now hosted in a Git repository.

KNotify Client

Korshak asked how to do emerge -uND --world && knotify "Done!". I thougt it would be a piece of cake, but apparently not.
The short answer, zypper in libnotify; notify-send Done, has a disadvantage of requiring a fair amount of GNOME software. But it does use a proposed freedesktop.org standard.
Then I find the method org.kde.KNotify.event, but its nice signature (ssavsayasx) means dbus-send (dbus-1-1.2.1-12) cannot invoke it (it cannot pass empty arrays, and it cannot wrap variants in arrays). No problem, let's use Python. But then we are blocked by knotify4 (kdebase4-runtime-4.0.4-19) reporting a wrong signature via the introspection interface.

# killall knotify4; cp /usr/bin/knotify4{,.bak}
# sed 's/type="a(ss)"/ type = "av"/g' /usr/bin/knotify4.bak > /usr/bin/knotify4

Now which event should we use so that the user actually sees the message? There are many of them, with user configurable actions. warning/kde seems to work out of the box.

#! /usr/bin/python
import sys
import dbus
m = sys.argv[1]
kn = dbus.SessionBus().get_object("org.kde.knotify", "/Notify")
i = kn.event("warning", "kde", [], m, [0,0,0,0], [], 0,
    dbus_interface="org.kde.KNotify")


That did it. But I still hope that there is a simple solution for KDE4 that I have missed.

D-Bus Spy

Like dbus-monitor, it monitors the system or session bus. Additionally, it can write the traffic to a file and read it later (-w, -r). It can dump all the gory details (-d) or pretty print all the gory details (-p). Like DBusMessageBox which inspired it, it can make message sequence charts (-c). It has some simple filtering abilities (-f, -F).

Try 0.1 if that sounds interesting.