daily blurbs
29 May 2006 (Mon)
07:39:10
#
Life
tofu, as in emacs.
After dist-upgrading to Xorg 7.0, I noticed many things broke.
Emacs is looking like a tofu display.
This sounds like a bad news.
Another thing is that radeon driver doesn't seem to work,
and I've reverted to vesa for now.
After a reboot, it now works.
I've got radeon driver working, and even with acceleration.
The fonts have come back also. Weird.
27 May 2006 (Sat)
17:55:06
#
Life
cowdancer and its limitations. I've been expecting a few
design problems with cowdancer; and I was hoping that these
problems are going to be minimal that normal package
building isn't affected. One thing that cowdancer cannot
handle is that because it plays tricks with replacing
i-nodes when writing to a file, any application that expects
same i-nodes before and after 'open' syscall will fail.
Apparently, this is a popular behavior to avoid symlink
attacks. So, if an application does lstat, open, fstat, in
sequence, and checks the dev/inode number of the lstat/fstat
result, the application will be confused when running under
cowdancer, which will change the inode number hooking
'open'. For cowdancer with pbuilder, if dpkg/apt and
preinst/postinst works, that's fine, and it seems to be
functioning; which is good.
26 May 2006 (Fri)
06:43:20
#
Life
cowdancer and pbuilder.
I've uploaded a new version of cowdancer and pbuilder, which work with each other.
The two new extensiosn, cowbuilder command,
and --pbuilder option (PDEBUILD_PBUILDER configuration option) to pdebuild,
will allow cowdancing with pbuilder.
Result of hacking at Debconf and in the plane back home.
Enjoy.
22 May 2006 (Mon)
11:45:44
#
Life
debconf7 ends quietly.
I'm in the hacklab, and there are quite a few people still
left in hacklab. Logistically, this debconf had the
conference rooms/hacklab/hotel rooms too far away from each
other, and I found myself staying in one location. Swimming
pool was a pleasant addition, but I couldn't utilise it when
I suffered diarrhoea. Language was a very big problem,
since the Hotel people did not understand a word of English
(well, they sometimes pretended to understand a few words,
but that was not enough to make them do what I wanted.).
Cost of living wasn't too low. The cost of everything
within this Oaxtapec Hotel premises was around 50% higher
than outside. Even in the outside, the cost wasn't as low
as expected, but around 50% of what I would expect from
Japan.
It apparently isn't a common practice to have a price tag
around, so the price of most of the things on store are
subject to negotiation.
Network was down for me most of the time, since there was no
wirefull connection to the hotel rooms, or conference halls,
and my wireless connection had some kind of kernel or
hardware problem which meant I could not be connected for
more than 5 minutes.
I had a few interesting work there, but I realized network
is a very crucial part of hacking.
The following should really be easily accessible offline:
Debian mirror (yeah, I can mirror it, no probs),
Debian BTS (I could rsync the whole beast),
Wiki.debian.org (how?), and
Version Control Systems (using git has solved most problems).
21 May 2006 (Sun)
01:25:18
#
Life
Biella's talk.
Rather hard for me to follow, with difficult terminology
repeated, with hard-to-follow content, with somewhat
US-centric references to whatever Amendment made to whatever
I don't recognise.
- Hacking is jouissance.
-
Humour is an important factor in hacker communication.
Laughter is indisputable success. Pleasure is visible.
-
Humour is practiced in a form, hacking is play for form.
Similarity there?
-
There is tension on meritocracy that it may collapse one
day. Transparency is important. New Maintainer process is
one solution to the problem.
02:14:36
#
Life
Lightning talks.
Coordinated by Joey Hess. Will we fit in?
- Jeroen WolfXXXX( I don't know how to spell his name):
automatic testsuite like piuparts and re-build of all packages, lintian exist, but they are not coordinated.
"Mole" system; to manage test results. A mail-based submission system. If interested, mail to debian-qa@lists.debian.org
-
Jose Parrella: let those people love Debian.
-
Andreas Schuldei: Debian in greater Linux ecosystem.
RedHat is moving to middleware market, and SUSE is not doing that great.
Hopefully one distribution that is supported
becomes LSB in the future.
-
Damog: automating wnpp.
There are 4890 bugs on wnpp, unmanageable.
1-year-old ITPs are going to be closed automatically.
ITA retitled 250 days.
-
Raphael Hertzog: collaborative maintenance.
-
Matt Taggart:
DSA = elmo neuro joey fil.
debian-admin@debian.org (DSA 4), debian-admin@lists.debian.org (DSA+11 people).
To make things done quickly: make requrest clear, vague requests will need
questions.
- chroot request: check what's installed and missing.
Give a space-separated list.
Which machine, which chroot?
- Joey Hess: learning from gentoo. daemon and config
files. gentoo does not restart the daemon, and new config
file is placed in known place, "dispatch-conf" is used for managing the config file.
-
Jeroen van Wolffelaar: Datamining on Debian packages metadata.
MIA developers tracking is also a delicate subject.
-
Jeroen van Wolffelaar: practice pronoucing name.
Manoj Srivastava (the final A is not pronounced).
Junichi Uekawa.
Thanks for cheering my name!
02:23:36
#
Life
Updated libpkg-guide.
I've made a major re-ordering of chapters. I'm looking at
reducing some text from the document so that it's better
organized.
03:13:42
#
Life
Debugging debacle by AJ and Erinn Clarke.
- originally Debian was part of FSF.
One of Richard's suggestion was
to include debugging information.
-
debugging symbols aren't too useful without source-code.
Where to put the source code?
Downloading source package doesn't actually obtain source-code;
some tarball and patches, which isn't really useful for gdb.
wig and pen format.
05:32:19
#
Life
debconf[67] BOF.
Andreas Schuldei starts by complaining this is not the kind of setting he expected.
It was not a BOF meant to be open to everyone.
- Edinburgh: discussion has been running for the past
half-year on mailing list evaluating some venues within
UK. Well-connected, and not as expensive as London,
Edinburgh turned out to be a good solution.
- Ryanair, Germanwings, and railways are available.
- Glasgow rains, Edinburgh does not.
- Venues allowed access 24 hours
- Andreas required cost of BigMac in UK, and noone really
knows the cost.
- Sarajevo, capital of Bosnia: Partially a Muslim country,
probably vegetarian is not much of a problem.
Apparently war is over, and crime rate is of European Standard.
- Shouldn't accessibilty be a more of a priority?
- Brixen, Italy: A resort place. Hosting school.
Boarding schools in town, which could be cheap accomodation?
Hotels 60 euros/night.
-
Skiing would be obtrusive for hacking?
Swimming is easier to come back to work, but
going back from skiing takes too much time.
-
Summer time is more convenient for many people to attend.
Having debconf in May was not a good idea. July just
before Kernel Summiy and OLS has worked very well for
many, January, February has a lot of problems.
-
Hacklabs should be close to the place the talks take
place. We lost a lot of time. Oslo and Brasil worked
well, but Finland and Mexico, it was too far away.
20 May 2006 (Sat)
00:30:09
#
Life
i18n BOF.
How the translation process should be governed.
- Rosetta reduces the sense of ownership in translation, which is against the current motivation source.
-
Some countries' government fund translation; but they use different tools for us.
-
We currently focus translation on unstable, not on testing.
However, having a statistics about difference on unstable and testing would be good, so that packages can be pushed into testing, at release time.
-
IRC channel is on #debian-i18n. It is on OFTC, not on freenode.
On a different note, I've been analysing the dmesg messages on the wlan card.
It looks like the messages are generated from linux-wlan-ng sources,
something is returning -EIO when it's not really expected.
- src/prism2/driver/hfa384x_usb.c(hfa384x_usbctlx_complete_sync) This results in -EIO: CTLX[3] error: state(Request failed)
- src/prism2/driver/prism2sta.c(prism2sta_commsqual_defer): prism2sta_commsqual_defer: error fetching commsqual
-
src/p80211/p80211conv.c(skb_p80211_to_ether): skb_p80211_to_ether: OTHER frame too large (2934 > 1500)
01:28:38
#
Life
About X and its history.
- V kernel, W window system, X.
X9 never existed.
X1-10 weren't extensible, new protocol required when adding e.g. color.
- X11 added extensibility.
- 1987 collapse of internet. Pre van-jacobsen TCP/IP code.
Lost packet meant re-transmit whole window.
Network slowing down meant re-transmit flooding the network.
Communication between Digital and MIT stopped.
Starting doing Fedex.
-
Give up internet; formed central development, hiring developers.
Consortium required money.
Corporate privilege meant access to source-code in-development.
-
X386 (VGA) code included in X11R5. Much money required for even inclusion of code.
-
XFree86, name play on closed-source X386. Gang-of-four.
-
XFree86 ignored by X Consortium.
Created a coorporation to join X Consortium.
However, still ignored.
-
XFree86 weren't interested in protocol and toolkits, but the drivers for cards.
-
X Consortium shuts down, The Open Group obtains
control. TOG wanted revenue stream for support, doesn't
really want free license. However, XFree86 wins control.
But that meant XFree86 meets the burden of full X.
-
X Consortium had some developers during 1986-1996.
TOG also had some developers.
-
TOG subgroup for X, hire consultants for bugfixes.
Most bugfixes actually come from XFree86
-
Market move over from proprietary to XFree86-related hardwares.
Commercial Unix vendors start relying on XFree86 code.
-
Most people except one person lost interest in XFree86,
but Cooporation is left as it is. The only person
remaining becomes the dictator.
-
Developers were ejected from project in fear of loss of
control, and a new license. XFree86 falls apart.
-
Xorg resulted in a flat community, rather than layered cabal.
-
Lesson: Governance matters; quick fixes last a long time;
stay involved, stay open.
05:31:18
#
Life
shared library packaging by Josselin Mouette.
-
API breakage, source level change. ABI, binaries need rebuild. SONAME signifies ABI.
-
API breakage, bump -dev package name.
- ABI breakage, change SONAME
-
Library transitions: ask the release team before upload!
- upload to unstable, and rebuild
- upload to experimental and work
- upload several versions of sources
- case study: gconf. /usr/lib/gconf2/gconf should be
located in /usr/lib/libgconf2-4/gconf to solve problems
with circular dependency.
-
Unstable ABI: HDF5, always changing interface, with same
SONAME. Mozilla, always changing interfaces. Handle
SONAMEs specifically for Debian, although it will break
binary compatibility with other distribution. (But what is
compatibility, really?)
- exporting all symbols is bad.
-
libtool options to allow only the symbols: -export-symbols-regex Mix_.*
-
Use -Wl,-Bsymbollic.
-
mysql; Debian symbol versioning differs from upstream,
since upstream took different name to symbol the version.
-
Relibtoolize with Debian libtool, so that indirect shared
library dependencies are removed.
-
Removing .la files is dangerous unless everyone provides pkgconfig data.
pkgconfig now allows specifying private libraries.
-
GNU ld, --as-needed, removes unneeded dependencies.
However, libtool recently broke this option.
This option is not very widely supported by upstream,
and many people consider it a hack.
-
Q. libpkg-guide, what's outdated?
A. libfoo3-dev, shouldn't be necessary.
Required updates to libpkg-guide
- done: -DEV package name recommendation change
- .pc favoured over .la
- done: libXcursor, libXrender .la files missing from the dependency chain will break.
-DEV dependency.
e.g. 363239.
.la files were removed by mass-bug-reporting, such as
363057, or with binNMUs.
Removing .la files had to start with leaf libraries.
- done: Removing indirect shared library dependency
- done: ld --as-needed
- done: -export-symbols-regex
- Documenting API/ABI/SONAME clearly
- done: pkgconfig section is required
- document pkgconfig
- removing indirect dependencies introduced with libtool
- cautions of libtool .la removal
17 May 2006 (Wed)
05:18:32
#
Life
Still fighting with my wireless setup.
I've upgraded the kernel to 2.6.16.16, and tried again.
However, it seems to pull the machine to a hard halt after a while.
Hmm..
This is USB wlan device (GW-US11H) on iBook G4.
May 17 05:14:14 ibookg4 kernel: hfa384x_usbctlx_complete_sync: CTLX[3] error: state(Request failed)
May 17 05:14:14 ibookg4 kernel: prism2sta_commsqual_defer: error fetching commsqual
May 17 05:14:14 ibookg4 kernel: skb_p80211_to_ether: OTHER frame too large (2940 > 1500)
May 17 05:14:16 ibookg4 last message repeated 2 times
May 17 05:14:36 ibookg4 kernel: skb_p80211_to_ether: OTHER frame too large (-30 > 1500)
May 17 05:14:37 ibookg4 kernel: skb_p80211_to_ether: OTHER frame too large (2940 > 1500)
May 17 05:14:37 ibookg4 kernel: skb_p80211_to_ether: OTHER frame too large (-30 > 1500)
May 17 05:14:37 ibookg4 kernel: skb_p80211_to_ether: OTHER frame too large (2940 > 1500)
May 17 05:14:38 ibookg4 kernel: skb_p80211_to_ether: OTHER frame too large (-6 > 1500)
On the other hand, I found something is weird about my cpuinfo.
bogomips of 36 doesn't sound quite right.
processor : 0
cpu : 7447A, altivec supported
clock : 1066.666664MHz
revision : 0.1 (pvr 8003 0101)
bogomips : 36.73
timebase : 18432000
machine : PowerBook6,5
motherboard : PowerBook6,5 MacRISC3 Power Macintosh
detected as : 287 (iBook G4)
pmac flags : 0000001b
L2 cache : 512K unified
pmac-generation : NewWorld
I asked in the mailing list, and got answer from BenH and
others that ppc moved on to using timebase instead of
bogomips, and bogomips is no longer calculated.
16 May 2006 (Tue)
00:22:45
#
Life
wookey, BOF on embedded Debian.
scratchbox 2, crossbuild.
- scratchbox2 is large
- crosstools: we currently lack standard set of cross compilers in Debian,
some maintained in emdebian.org. Apparently, cross tools which run on reasonable architectures will go into mainstream.
-
slind, uses dpkg-cross. postinst is ran natively over NFS, to workaround some postinst problems.
-
slind/ directory in source overriding debian/ directory in source for customization.
-
ARM EABI transition, there is a separate BOF. New arm architecture is required.
-
Extramadura meeting result: we will upload crosstools in Debian main.
-
How to start working on embedded Debian.
Getting an evaluation board from a vendor,
asking the vendor if they support linux.
Retail embedded devices tend to be undocumented and
actually harder to work with.
-
Official answers from semiconductor companies will say
they would say no, but there are many engineers.
03:09:28
#
Life
translation BOF.
- Problems with rosetta. Multiple packages are translated.
Web-based tools open to users, but doing QA over such
openness is one problem.
-
Concern about rosetta is that it's non-free.
-
Flow for translation: translation, review, discussion.
-
The same strings sometimes need to be translated
differently depending on context.
-
DDTP has been broken for a while.
There's something new which should now work.
-
We probably really need a New Translator Process.
05:40:07
#
Life
My wlan is dying quite often.
It's a prism2 card.
There seems to be some kind of error condition, and once it enters that condition, the number of error packets will just increase.
hfa384x_usbctlx_complete_sync: CTLX[3] error: state(Request failed)
prism2sta_commsqual_defer: error fetching commsqual
skb_p80211_to_ether: OTHER frame too large (2940 > 1500)
skb_p80211_to_ether: OTHER frame too large (2940 > 1500)
skb_p80211_to_ether: OTHER frame too large (2940 > 1500)
skb_p80211_to_ether: OTHER frame too large (2940 > 1500)
14 May 2006 (Sun)
21:56:08
#
Life
Fighting with wlan card.
I'm trying to figure out why this Prism2 USB
card is locking my system hard.
Maybe I'll give up and use the wirefull network for the timebeing.
3 May 2006 (Wed)
01:20:37
#
Life
Wondering how to remove tempfiles.
Since I'm so much annoyed by the current behaviour of cowdancer,
I don't want to have .ilist file in the current directory.
Instead I want to put it somewhere under /tmp.
Current model of cowdancer execution is that it creates a file
called .ilist in the current directory, and
each session of application under cowdancer control will mmap that file.
It might be better if I just dump something in /dev/shm, or /tmp
which might even be memory-backed to alleviate the need for disk access.
cow-shell execs the child session, and thus it's no longer
available to remove the .ilist file when it's time to free it.
.ilist file isn't that big, so leaving it in /tmp for
later automatic removal by other means sounds feasible; I'm not too sure.
SysV IPC shm routines with Linux extentions look a bit
attractive, although I really dislike the arcane interface.
Gives me the same sense of fear as when I look at X Toolkit programming.
I wonder what's the current recommended way to go forward.
|
Pros |
Cons |
A known filename in current directory |
Same file for multiple invocation from the same
directory |
Hard to automatically detect the files no longer
used, to mark them for deletion.
|
/dev/shm or /tmp |
Eventually automatically deleted by tmpreaper et al.
Could be memory-backed.
|
Multiple files can be generated for same directory
under different names. |
shmget/shmat |
Could be deleted when all attaching processes die
when shm segment is marked for deletion.
(Linux extention) |
Uses an arbitrary number as shm ID.
Arcane interface with Linux-specific extensions,
something to be left for good along with the fond memories of
the 20th century? |
Junichi Uekawa
$Id: 200605.html.en,v 1.22 2006/05/29 14:06:56 dancer Exp $