Monthly Archives: January 2008

The kernel evolution speed

Apc quotes Jonathan Corbet on his “The Kernel Report” talk in LCA 2008:

“We are adding about 2000 lines of code to the kernel every single day of the year, without exception,” Corbet said. “Nobody can really keep up with this [on their own] any more. It’s an amazing process, and it seems to be working.”

Although I didn’t (and probably can’t) contribute a single line to the kernel, reading this made me a prouder Linux user… you too ?


Filed under Proud to use free software

tar.bz2 sources in debian

Reading Raphael’s Misc development news (#2), I thought I can now upload packages with orig.tar.bz2 sources. Dak quickly replied an error, and the ftp-masters verified that this feature isn’t available (yet).

So, for people like me, who didn’t read the text carefully or comprehended it wrongly, the new feature refers to the comprssion of the .deb file itself and not the the one of the upstream tarball.

I just wonder when will it be possible to upload .orig.tar.bz2 files to the archive, as it will save me repackaging some of my upstream tarballs…


Filed under Debian GNU/Linux

My thoughts on RHEL 5.1 installation

I decided to test RHEL 5.1 to check what changed since RHEL 4. As I usually do a network installation, I had a few surprises.

The installation was done with qemu (with kqemu module) on an AMD Athlon(tm) XP 1800+ processor. The virtual ram was set to 256MB. The installation was done an an LVM LV, and took 25 minutes (pretty good I believe).

I tried to do a minimal installation, so I deselected all the available packages from the list. I ended up with 366 rpms and 935 MB on the disk (excluding /boot).

Look back in the anaconda-ks.cfg seems like that not the minimal installation as this is the %packages section:


I couldn’t find on the CDs a file to explain each of the groups here. The manuals talk about comps.xml for this info, but there is no such file on the 5 CDs. Such files are available for the extra software on the CDs like the cluster / virtualization software.

I also was surprised to see that I needed 3 CDs for the minimal installation. Although cd #2 was needed only for:

  • libgcj
  • java-1.4.2-gcj-compat
  • gjdoc
  • jpackage-utils

and cd #3 only for:

  • wvdial
  • isdn4k
  • minicom
  • lrzsz
  • libwvstreams
  • antlr

I guess these could easily be moved to the first CD and save people the CD changes which wastes time while babysitting the installation instead of doing it completely unattended. Also, seems like by removed the @dialup and @java groups we can be satisfied with only the #1 CD.

This also brought to my attention the method red hat choose on which CD each RPM will be placed. I’m sure doing a popularity contest like Debian is hard for a commercial distribution, but still knowing the method behind the CDs will be useful.

Generaly speaking, the installation went fine. I can start testing the system itself. about testing the


Filed under Red Hat Enterprise Linux

How technical should error messages be ?

When I opened pidgin today, I got an error message regarding my buddy list file. Since the program couldn’t use the current list, it has loaded an old file with about half my buddies.

Pidgin blist.xml error message

The problem with this error message, is that it doesn’t give you any clue to the root cause of the error. This of course makes it harder for me to fix the problem and get on with using the program.

I’ve opened the file with vi, and it generally seemed OK with the all the major starting/ending XML tags. So I opened it with Firefox, to see if it parses the XML, and only then I could see the real problem:

XML Parsing Error: reference to invalid character number
Location: file:///home/kaplan/.purple/blist.xml~
Line Number 416, Column 47:

Looks like one of my buddies has a weird Unicode character in his nickname, which wasn’t correctly escaped by pidgin when it had been written to the XML file. Removing the problematic character fixed the problem and pidgin started with my full buddy list.

For me, as a technical person, seeing a problem with a XML file, usually raises the flag regarding parsing errors. I don’t expect common users to have this information. But giving a more informative error message could help such a user to get help and directions to the solution in forums or IRC.

So to all the programmers reading this, I’d like to ask you to try and give a clue to the problem in your error messages. It can simplify usres’ life…


Filed under Free software applications

Why does Oracle’s applications needs xscreensaver in order to run ?

Working with some of Oracle’s products, made me wonder about their installation requirements as appears in the documentation.

Let start with Oracle® Database 10g Release 2. The installation guide for linux x86 says under “Checking the Software Requirements” it needs the following RPMs for RHEL 4.0:

  • binutils
  • compat-db
  • compat-libstdc++-296
  • control-center
  • gcc
  • gcc-c++
  • glibc
  • glibc-common
  • gnome-libs
  • libstdc++
  • libstdc++-devel
  • make
  • pdksh
  • sysstat
  • xscreensaver
  • setarch

Most of them are valid requirements when needing to compile C/C++ code (as it may be done inside the database). But why do I need to install gnome’s libraries, gnome’s control-center and xscreensaver?

I fail to see how are these 3 are related to running a database. It is reasonable to require these packages for running a graphical installation, but that’s not a runtime requirement! I even tested that with checking the database opened files while it’s been running.

I really don’t want to have these packages installed on servers unless I must, as I believe server should contain the minimal software required on top of the distribution default installation.

Another funny (and weird) fact is that for SELS 9.0 there isn’t any requirement for gnome’s control-center. For RHEL 3.0 there aren’t any requirements for the 3 packages…

Mentioning optional requirements are probably the best why to solve this, unless there’s some real requirement that eludes me. John Smiley’s “Installing Oracle Database 10g Release 2 on Linux x86″ has this for openmotif21 which is mentioned as needed only for Oracle demos.

The same happens for Oracle® Enterprise Manager in installation guide for linux x86 10g Release 2 (10.2), but this time the package requirements for RHEL 3.0 and SELS 9.0 don’t mention any of the 3 packages, whilst they are required for RHEL 4.0. Very weird.

The status for Oracle Application Server 10g Release 3 is similar.

I would be happy if someone could shed some light on the issue.


Filed under Proprietary software, Red Hat Enterprise Linux

Who moved my PEs (or why does pvmove sucks on linux) ?

Yesterday I wanted to extend an existing LVM VG with some central storage disks. In the process, I notice the VG is built upon local & external disk, which isn’t according to my standards.

So I decided to extend the VG with the external disks, and then use pvmove command to move all the PE from the internal disk before reducing the VG. The problem is, that on RHEL 4.4, the pvmove command hangs during it’s work, never returns to the prompt, and locks all the other LVM commands.

Seems that for little amounts of PEs the command works, but for 10G disks it hangs. As nothing else helped, I was forced to hard boot the server using to power button. Running the command in single user also doesn’t effect the result, although non other user/services are running.

RHEL 4.4 has LVM2 version 2.02.06, RHEL 4.5 has version 2.02.21 and RHEL 4.6 has version 2.02.27. Reading upstream’s changelog seems there isn’t any change regarding pvmove, although it’s still worth testing if upgrade helps.

Version 2.02.29 has “Refactor pvmove allocation code” in its changelog, but it will take time before this version will be officially available for RHEL (fedora 8 only has 2.02.28). Hoepfully it will solve the problem, as my HP-UX colleagues are laughing on my expense… And that I have to invest more time in mimicking pvmove work manually.


Filed under Red Hat Enterprise Linux

usertag support for mass-bug script

While working on mass bug filling in Debconf, I noticed problems with specifying user & usertags in mass-bug’s templates. Which resulted with some manual work to tag all the reported bugs after getting the bugs’ numbers.

Thanks to Mohammed Adnène Trojette, it seems to be resolved:

* mass-bug: add –user and –usertags options to avoid their wrapping. (Closes: #429479)

So I guess that’s a small improvement for the QA people… Enjoy.

Leave a comment

Filed under Debian GNU/Linux