Thursday 28 February 2008

sarge + cowbuilder = love

Update:
A disclaimer/warning is needed: UNLESS you KNOW what you're doing, you probably are better off building packages for each of the supported Debian distros.


I know it sounds cheesy, but I couldn't find a better name.


At my workplace we are packaging software to work for as many distros as possible at one time. As a consequence the .deb packages we produce should work fine on all Debian distros: oldstable, stable, testing, sid.

To make sure we achieve that, we are building them in an oldstable chroot. OTOH, using pbuilder is very demanding on I/O, especially when entering the chroot and when exiting it. That slows down a lot the build system and is time to change that.


One reliable alternative is to use cowbuilder (part of cowdancer package). What is interesting is that cowbuilder relies on the existance of /usr/bin/cow-shell inside the chroot, so the cowdancer package needs to be installed both in the chroot and the host machine.

But cowdancer was not a part of sarge, so, if you try to create a new sarge cow directory with cowbuilder, it will fail at the end saying it didn't found the cowdancer package. Also trying to login via cowbuilder --login will complain it didn't found /usr/bin/cow-shell.


The fix for this is simple, once you figured it out and saw that cowdancer is available in sarge as a backport:
cowbuilder --create --distribution sarge --basepath /var/cache/pbuilder/sarge.cow --othermirror 'deb http://www.backports.org/debian sarge-backports main'
You may prefer to be sure that the only thing from backports is cowdancer and use a more hands-on approach[1]. Also, this will eliminate the need to correct later the apt sources to exclude backports[2].

Write the same command as before, but omit everything starting with --othermirror. Then install by hand the package in the chroot.
twix:/home/eddy# cowbuilder --create --distribution sarge --basepath /var/cache/pbuilder/sarge.cow
twix:/home/eddy# chroot /var/cache/pbuilder/sarge.cow
twix:/# wget http://ftp.de.debian.org/backports.org/pool/main/c/cowdancer/cowdancer_0.23~bpo.1_i386.deb
twix:/# dpkg -i cowdancer_0.23~bpo.1_i386.deb
twix:/# rm cowdancer_0.23~bpo.1_i386.deb
twix:/# exit
twix:/home/eddy# cowbuilder --update


Since after the bootstrap all .deb packages will be in the apt cache in the chroot, you might want to save some disk space:
twix:/home/eddy# cowbuilder --login --save-after-login
twix:/# aptitude clean
twix:/# exit

That's it!


If anything is broken somehow, missing or whatnot, cowbuilder --login --save-after-login is your friend.

[1] the initial bootstrap does not pull from backport anything else but cowdancer
[2] since you want to compile for plain sarge, not sarge+backports ;-)

Postfix - (almost) a satellite system

When installing postfix and configuring it to act as a "Satellite system", all mails, including messages to root, will go to the remote mail delivery system. Also, there might be a local user account that absolutely doesn't exist on the remote MTA and should stay on the local machine, but, by default, this doesn't happen.

This is not good.

What is needed in such a case is virtual_alias_maps and a /etc/postfix/virtual file such as:

root root@localhost
root@fqdn.domain.name root@localhost
eddy eddy@localhost
eddy@fqdn.domain.name eddy@localhost

Don't forget to run postmap /etc/postfix/virtual, after restarting the system.


This tip is particularly useful on laptops.

Monday 18 February 2008

[VCS/SCM] same language, different lingo

When people say "lightweight branches":
  • In git lingo, most likely they will think about the fact that you don't need a different directory to switch to a different branch.
  • In subversion lingo, they will probably mean that the cost of making a branch in the repository does not impose an expensive operation
  • In bzr lingo, it seems they mean a repository with short history (aka shallow copy, for most other people)
  • it seems that mercurial refers to git's meaning as "named branches" (not sure about this); not sure if mercurial documentation refers to anything else as "lightweight branches"

Export means:
  • for subversion: to create a working copy without any meta information
  • for mercurial/hg: "The act of exporting a changeset generates an augmented patch file that describes the change."
  • for git, it seems to be a useless 'cp -r' operation
There are several other differences in the VCS/SCM lingo paralel uses, if you want to look closely, but these are enough for now...

No wonder we have communication problems and people don't grasp the power of a new tool when it changes the meaning of previously used terms.

Tuesday 12 February 2008

wireless prohibition

After setting up wireless via WPA2-PSK with AES and realizing I really need the new firmware, I ended up using the latest wireless-2.6 kernel from branch 'everything'.

Now I am again browsing wireless-ly and I am wondering if the digital divide isn't already here:



Is there any reason why this discrimination is done? do they fear that they won't be able to sell the series?

Friday 8 February 2008

forking privately in Debian

Joachim, the key is probably to implement patch support in srcinst and actually implement something nice which I think has nice applications.

It might be nice to combine srcinst with debian-xcontrol to obtain also customizable dependencies (e.g. no arts support).

I guess you'll be the best person to take over srcinst since:
  1. John Goerzen doesn't want to/can't continue maintaining it
  2. is written in Haskell and you know Haskell
  3. source is available for grabs from http://darcs.complete.org/srcinst/

The basic idea is to have another command[1] wrap aptitude/apt-get/srcinst and make sure it does the proper job, depending on the local configuration.


[1] let's say, the debtoo command - srcinst should remain the tool for clear 'from source installs'

wpa2-psk with aes on a broadcom wlan0 (2.6.24)

Update: I managed to find out why the wpa_action stuff was needed. Please ignore the lines written like this; they are there just for reference.

One more update: it seems that the firmware needs to be in sync. I ended up using the wireless-2.6 kernel from the everything branch.



I just managed to get my wlan from my laptop to work with WPA2-PSK with AES with the free Broadcom driver (now named b43, formerly bcm43xx).

0b:00.0 Network controller: Broadcom Corporation BCM94311MCG wlan mini-PCI (rev 01)

0b:00.0 0280: 14e4:4311 (rev 01)

In order to do this I needed linux 2.6.24-1 from unstable and the b43 driver.
bounty:/home/eddy# lsmod | grep b43
b43 119976 0
rfkill 12816 3 rfkill_input,b43
mac80211 132236 1 b43
led_class 10120 1 b43
input_polldev 9872 1 b43
ssb 39428 2 b43,b44
pcmcia 45720 2 b43,ssb
pcmcia_core 46500 2 b43,pcmcia
firmware_class 15232 2 b43,pcmcia

The final trick was to convince wpasupplicat to reload the config with:
bounty:/home/eddy# wpa_action wlan0 reload
wpa_action: reloading wpa_supplicant configuration file via HUP signal

This is the wpasupplicant.conf file that I used:
bounty:/home/eddy# cat /etc/wpa_supplicant/wpasupplicant.conf | grep -v '^\s*#' | sed 's/psk=.*/psk=aaabbb___ENCRIPTED_SEE_wpa_password___cccddd/'
ctrl_interface=/var/run/wpa_supplicant
ap_scan=1
network={
ssid="toblerone"
scan_ssid=1
proto=RSN
key_mgmt=WPA-PSK
pairwise=CCMP
group=CCMP
psk=aaabbb___ENCRIPTED_SEE_wpa_password___cccddd
}

and this is the relevant interfaces area:
auto wlan0
allow-hotplug wlan0
iface wlan0 inet dhcp
wpa-driver wext
wpa-conf /etc/wpa_supplicant/wpasupplicant.conf
wpa-ap-scan 2


This is what it looks like when is working:

bounty:/home/eddy# iwconfig wlan0
wlan0 IEEE 802.11g ESSID:"toblerone"
Mode:Managed Frequency:2.412 GHz Access Point: 00:1B:FC:45:33:70
Bit Rate=48 Mb/s Tx-Power=27 dBm
Retry min limit:7 RTS thr:off Fragment thr=2346 B
Encryption key:1337-0000-C121-d73D-0207-RE41-0000-3210 [2]
Link Quality=98/100 Signal level=-36 dBm Noise level=-68 dBm
Rx invalid nwid:0 Rx invalid crypt:0 Rx invalid frag:0
Tx excessive retries:0 Invalid misc:0 Missed beacon:0
And here's the result of a scan (relevant section):
  Cell 02 - Address: 00:1B:FC:45:33:70
ESSID:"toblerone"
Mode:Master
Channel:1
Frequency:2.412 GHz (Channel 1)
Quality=93/100 Signal level=-42 dBm Noise level=-68 dBm
Encryption key:on
IE: IEEE 802.11i/WPA2 Version 1
Group Cipher : CCMP
Pairwise Ciphers (1) : CCMP
Authentication Suites (1) : PSK
Bit Rates:1 Mb/s; 2 Mb/s; 5.5 Mb/s; 11 Mb/s; 18 Mb/s
24 Mb/s; 36 Mb/s; 54 Mb/s; 6 Mb/s; 9 Mb/s
12 Mb/s; 48 Mb/s
Extra:tsf=0000000072ff542d
and here is proof it works:
bounty:/home/eddy# route -n
Kernel IP routing table
Destination Gateway Genmask Flags Metric Ref Use Iface
192.168.77.0 0.0.0.0 255.255.255.0 U 0 0 0 wlan0
0.0.0.0 192.168.77.254 0.0.0.0 UG 0 0 0 wlan0

bounty:/home/eddy# ping debian.org
PING debian.org (192.25.206.10) 56(84) bytes of data.
64 bytes from gluck.debian.org (192.25.206.10): icmp_seq=1 ttl=36 time=201 ms
64 bytes from gluck.debian.org (192.25.206.10): icmp_seq=2 ttl=36 time=199 ms
64 bytes from gluck.debian.org (192.25.206.10): icmp_seq=3 ttl=36 time=200 ms

--- debian.org ping statistics ---
3 packets transmitted, 3 received, 0% packet loss, time 1999ms
rtt min/avg/max/mdev = 199.691/200.519/201.046/0.786 ms

Woooohooo! :-)

Thanks to all people involved in b43 development and all the ones made this possible (Debian developers).

Posted from bed, via wlan.




Update: You need the firmware blob, which can be extracted from a Windows driver with bcm43xx-fwcutter (now called b43-fwcutter); I already had it from my previous attempts to configure wlan with bcm43xx. I am not sure if I should use the new tool. You really need the firmware and driver to be in sync.

Update: it seems the b43 driver page (the entire linuxwireless.org site) went down sometime yesterday evening, since yesterday afternoon I was browsing through the site without any issues. (Note: I live in Europe, for reference)