Daily blurbs Jul. 2005

29 Jul 2005 (Fri)

18:54:53 # Life Release management. Since it is a niku-no-hi (the day for meats), I've decided to work with freshmeat and get some of my packages registered on freshmeat. The last time I've done that is more than two years ago...

• wysihtml 0.12
• binfmtc 0.7

17 Jul 2005 (Sun)

01:43:46 # Life I've hacked up on dpatch today, and fixed one of the last pending tasks that I had on my to-do list. testsuite is running 11 tests, and running successfully. Things are looking fairly good. I'm prodding MH to upload the new version for me.

02:11:47 # Life I've met ecasound developer Kai Vehmanen, and discussed a great deal about Debian and audio-related development. I have been maintianing ecasound for the last 5 years and he's the upstream maintainer of ecasound. I found it interesting that he references many things in that direction. He referred to me two programs that are of interest:

• frewheeling
• tutka

16 Jul 2005 (Sat)

00:26:06 # Life Input method BOF, just outside of smökki

• Enrico, tried to set up input methods but never got it working
• kmuto: porting im-switch from Fedora Core. Fedora Core provides a default IM for each language. I only know Japanese, I would like some input on other languages.
• Chenshun? from China. Done some virtual keyboard and writing for Westerns, Chinese input through handwriting is on roadmap.
• Mohammad?. working on western input methods. Contributing to SCIM for Arabic input methods.
• Richard So?. Input method. Forked some Chinese Fonts.
• Masanori Goto; Interest in Font package. Maintain a lot of font packages. Truetype and X fonts.
• Some Kanji is compatible in Chinese and Japanese.
• Junichi: I maintain canna.
• Ukai: Maintainer of SKKinput, XIM engine for SKK style, PM for UIM which is done as a part of Japanese National Project.
• Jaldhar Vyas. Interested in Indian localization, do not know in input method really, but hoping for good support in Indian.
• Simon Richter, learning Japanese for 2 year.
• What works and what does not work with input methods? How many exist? XIM, UIM, SCIM, IIIMF, GTK-immodule(interface for input methods), Qt, kinput2. m17n is a middleware, most of them are interlinked. Are these obsolete? There is no upgrade path which follows closely the behavior.
• For console Japanese use emacs, or uim. For Arabic, akon for console (not in Debian).
• Chinese, XIM. There is no XIM layer. Non-shared, individual developers doing different thing.
• Is SCIM going to be adopted by Xorg as XIM2 ? People want different UI, different UI, this need is not satisfied. XIM is being planned to be dropped; it does not have editing-in-place.
• Current SCIM is outdated in unstable. Roger So uploaded the latest in experimental.
• SCIM is too large and heavy. For Nokia 770, something completely different one is implemented.
• Drawing, is not supported by XIM.
• IM is handled in GTK with IPC.
• Qt will adopt similar framework in Qt4.
• Although there are around 17 input methods that can be chosen in the input method menu, most people will use up to around 2.
• Gnome has an applet for switching keyboard.
• Handwriting is totally missing from open-source, it's only available in closed-source. We usually need to have multiple characters recognized at the same time, rosetta uses one-character-at-a-time. ZI has very good software.
• Can OCR technology be used? They are a bit different in the way, but the basic technology could be used.
• It would be better for Debian if things are concentrated into one. If you are going to SCIM, there must be a migration path from kinput2. Some keyboard shortcuts from kinput2.
• IIIMF is created by Hiura (Sun) who is also original author of XIM. RedHat is going forward in the direction of IIIMF. Architecture of IIIMF is too complex to reasonably support. Fedora already uses IIIMF.
• It would be great if installer could choose the proper input method.
• Locale and IM may not link. They could be different. Current glibc has thread-independent locale support, which means each thread can have their own locale.
• In C++ each file handle has a locale.
• Linux limitation: you cannot change the locale while running the application. But it is not easy to change that.
• Complaint on SCIM: Simon Richter: If the first item is the default item, it's Arabic. How do I set the default IM? How about environmental variables? Policy forbits that, but anyway. There is a scim-configure-agent. Good point about gconf is that it can spread configuration change to all applications.
• SCIM has a configuration with plug-in architecture. Can use any.
• IM Setup howto: We really needs a Howto. But first we really need to decide on what to support.
• Input Method selection should be done in a XDM/GDM/KDM menu, so that the default input method can be chosen for the session.
• Keymaps. Using XML for virtual keyboard keymap.
• Old timers are using SKK, kinput2. UIM is a new input method that is becoming more popular in Japan. SCIM is not something that is evolving in Japan.
• ime-devel.alioth.debian.org as a hosting place for the development.
• TODO for Debian.
  • A howto for setting up IM.
  • Create wishlist items for Display Managers
  • Allow IM selection even if LANG doesn't require
  • How to tell SCIM what is preferred language is

15:17:31 # Life Altivec talk. Summary: There is potential for improving speed of powerpc applications by using altivec, but it is underused and thus undertested under Linux. Several Projects exist to utilize altivec in Linux, and they are starting to make progress.

• Altivec does not require you to place the result on one of the operands like Pentium.
• There are projects like libfreevec; hashing, sorting functions being started.
• Gotom: What is the size of buffer? If the cache is spilled out, we need to bring data from memory. memcpy with altivec 4 times faster will depend on copying size.
• Going outside of cache, the speed will drop.
• Who knows memfrob function?
• Very fast on up to 16kB (L1 cache). 256kB is rather fast as well (L2 cache). But due to cache prefetching, it will be faster than current glibc function.
• Altivec has 128 bit bus to cache, and integer unit has 32 bit bus.
• Insertion sort is a stable algorithm. It is a useful algorithm if you have a almost sorted data. Quicksort is slow on almost sorted data.
• Q. Cost of context switches? A. It should be higher but noone seems to have measured it. Noone uses altivec enough to care about it.
• Gotom: Do you use Altivec for user-kernel copy? Using altivec there in that communication we can achieve more performance.

15 Jul 2005 (Fri)

00:03:21 # Life At Smökki, I was talking to Broonie via IRC. I wanted to meet him. He said he was also in Smökki, and I asked him to stand up. Then I noticed somebody behind me stood up. It was broonie. Heh. That's what debconf is all about, right?

01:37:05 # Life Some people get overenthusiastic when I propose some changes to shared libraries. Enough mail on debian-devel for today; I'll think more about it, the proposal obviously wasn't clear enough. duh.

16:10:11 # Life Debian women talk session.

• 1/3 IT professionals are women, but in Debian it's only 1.5 %
• Debconf participant: 10% are women.
• Users are different, why shouldn't Debian? Diversity.
• Debian-women is interesting in that it's a small-scale Debian. There were trolls, we had to ban a few people. People abusive on mailing list and others. Actively poisonous people do exist, and debian-women has banned such.
• Branden: Debian did have banning by Bruce Perens of Lars, but Banning entrenches hostility; and which is not. Erinn: These people are actively insane. They are harassing people off-list
• Bdale: Vocal minority participants, may create a perception. Vast majority of us may be very nice or very social people, who enjoy working with each other.

19:16:25 # Life Hacking libc locale data, by Denis Barbier.

• Numeral grouping, used by GNU extention printf %'f.
• Money format. Can be checked by money_format %n. In finnish, it should come after money as in 10.00€, but it should come before when it is written as EUR 10.00.
• LC_TIME in POSIX defines week to start with Sunday.
• Really need more hackers to maintain and debug locale information.

19:57:44 # Life debian-installer

• debconf gives i18n, preseeding, etc.
• main-menu dynamically orders udebs by dependencies. Ordering can be done by udeb depenency.
• Trunc for new development. People branches are used for experimental. sarge branch is the stable one.
• IRC for coordination, no central power, joeyh is involved in most parts, so when he his advise would be a good idea.
• English sucks! translate it! Use of gettext (po-debconf system introduced by Denis Barbier). Everything translatable thing goes to debconf template.
• Towards World domination. Translate all the language. Q. There are uncovered regions, is it technical or is it just there are no translators?
• Add more locales, easier framework? (rosetta, pootle, DDTP?)

20:50:59 # Life Marga, are we devoted to our users?

• This is a non-technical talk, social contract states we are to be devoted to users.
• Our users aren't only developers, programmers, or geeks.
• 2 minutes for booting the system, it's too slow!
• Bugs are not fixed, but bugs other than RC bugs do need to get fixed. Advertize them for help. Some bugs are upstream bugs and hard to fix. Providing patches will improve the turnaround time of bugfixing. Publish the name of the people who fixed more bugs; giving more visibility to bug fixing will motivate more non-DDs to help. Too many bugs exist, we can't fix all without some kind of change.
• Teamwork is better. Use Alioth. non-DDs can be incorporated as alioth contributors. Consider NMUs as sign of teamwork!

14 Jul 2005 (Thu)

03:33:57 # Life Day trip and sauna. It was a very eye-opening event for me. New experience in new part of the world.

Too little hacking done today, but who cares?

03:45:05 # Life With the new su patch, which was active in 4.0.3-36, the following command will fail.

new behavior:
ibookg4:/home/dancer# su -p dancer bash 
/bin/bash: /bin/bash: cannot execute binary file
ibookg4:/home/dancer# su -p dancer -- bash
/bin/bash: /bin/bash: cannot execute binary file

old behavior:
ibookg4:/home/dancer# su -p dancer bash
ibookg4:/home/dancer$ exit
ibookg4:/home/dancer# su -p dancer -- bash
ibookg4:/home/dancer$ exit
	

I know nobody reads the manpage. Just claiming that it was "Not documented", is not a good way to break things, especially if it has been this way for... somewhat 10 years? Surely a transition is necessary?

16:43:26 # Life shared library packaging BOF by myself. Some questions that were raised during the talk.

• There should really be policy on -dev package names, so that we know which is the corresponding -dev package.
• -dbg policy? I'm not quite sure about that, I would like to get it done so that -dev packages have the split debugging symbols, rather than making another 1500 packages just for debugging symbols.
• C++ libraries aren't handled well by d-shlibs. Needs more hacking. Their ABI change too often as well...
• Branden: There is a tool called icheck that checks for ABI changes. It is evolving, and looks promising.
• All in all, there should really be a tool that checks what's changed in the binary package output. icheck needs that, other tools may also.
• libpkg-guide really needs more visibility, and translation.
• gotom: why are you using objdump to determine the NEEDED section; when you can use ldd to obtain the full path of the shared libraries. me: You can find the shared library package from that, but it is difficult to derive the development library package.
• Not many people present knew about how to write pkgconfig scripts. Maybe a tutorial is called for.

19:18:40 # Life Zen and art of Free software, by Enrico Zini

• Use persona, to know needs of users.
• Personal goals are most important. Work goals may come later.
• Work goals of servers maintenance, with personal goal of using mail and reading mail.
• Magic number 7 plus-minus 2, working cache which has a limited size.
• Flanagan Critical Incident Technique is interesting. Social debugging.
• Comment from the field: Flanagan has a missing question: "What did you want to do"
• You are a user also, if you make users happy, you yourself may be happy.

20:02:04 # Life 2 minutes silence in honour of London Bombing victims. Everyone in the room stood up for 2 minutes.

20:02:25 # Life Dannf, and others, on Debian Kernel team.

• Started with herbert.exit(), 2004 May 4;
• Had to go back and see existing patches, not clear why these are there. They were monolithic patches.
• Version sync was a big issue. i386 and other architectures were very different. About 10 different versions, and some architectures not using kernel-source package.
• Quite a few non-DDs working. m68k, arm are not maintained with the kernel team.
• for d-i, kernel modules are split up in different udebs.
• Firmware blobs. 2.6.12 fixed problems. Wiki page on kernel firmware.
• initramfs, 2.6. CPIO archive tugged on to linux kernel image.

12 Jul 2005 (Tue)

00:10:12 # Life BOF talk on debconf organization. memo follows.

• Would like to sponsor speakers. However, it is more important to getting as much people to Debconf, rather than sponsoring someone who may be wealthy enough to come by himself.
• Prioritization: travel is large chunk. housing is more or less fixed cost.
• big-mac index is almost best estimate for the cost of Debconf. Oslo surpassed Japan for being the most expensive.
• We moved out from a low-level hacker thing to a more civilized thing. We tried to keep couples to a room. Last year one fixed couple, this year tons of them. People scream when asked to sleep on the floor.
• Finland, summer, noone is there. Registration was lot of work. Brazil, people waited. This year we had volunteers. We had to handle volunteers and develop processes. People good at people-handling, making procedures, is required. It is harder to get people in finland to help, since they are gone.
• Money is handled through the Linux aa...
• Past experience with using SPI for transfer of money was troublesome, but this year it is functional.
• Core group of people with separate responsibility is required. Money, housing, food, reservation. People who don't change every year, is ideal. There is a learning curve.
• There is a lot of documentation, we have quite a lot in Wiki.
• Make the right guess; you need to keep track of what is being spent, and make a projection, or you will have problem at the end of the event if expenses are more than what is gathered.
• Some people needed list of speakers before they were sponsored or allowed to attend by their employees. Invitations were sent for such people.
• Toronto, 120 people said they would come, 80 people came; several didn't say they would come.
• 230 people registered by 6 June. Around 300 people are here now. hotel; 42, living in area 47, HUT 170, 29 latecomers who stay in the cellar.
• It is important to get website coordinate with the organizers. debconf.org and comas having different login was confusing, and some people actually did something wrong.
• Preparation for bags with few days. Bags were personalized.
• Video team, 20 people. Video itself is private material at a dear cost.
• Q. debconf was 3 days originally, why was the decision to make it 10 days? A. People were tired after debcamp, so debconf didn't work very well. We tried to mix both. 4-5 days required for prep, and debcamp this year was good for the preparation period, refining process.
• Experimented with optimization for talk, for parallelisation of talk. Person-based might be better, as it was done in OSLO.
• Which talk is a BOF isn't clear. Schedule doesn't have the details, which is unclear.
• Q. Were you successful in making load distributed? A. tried to push tasks to people, people come to help with debconf, but before things become real, people don't really relize the need for help. Monthly meeting on every first monday, people didn't really realize. We met only once physically. However, electronically, things get longer to happen. Telephone was used extensively.
• when debconf is over, at 12 o'clock, it needs to be evacuated and cleaned.
• Whether mattresses are on the floor or on the bed is the least of the worry; the food is more important, and other arrangements.
• Photo guide was very good, it was effective. Only problem was that it was rather difficult to print out.
• Hosting, will be simpler if an organization could back things up, that would be good. Or using a hotel that can handle the hosting.
• HUT was empty for this period just before some sports event.
• We expect and hope for a more automatic system for conference. Meal ordering, food pool, etc.
• 1. Planning is important. 2. Volunteers do what is required. 3. Figure out what the volunteer is capable of.
• Size changes, 500 people, we have different logistics. For example, sauna schedules. Packaged deals look more interesting, which include day trips etc.

15:19:40 # Life Thoughts on apt with gpg support for pbuilder. If the overhead of running gpg inside the chroot is not prohibiting, an implementation might be useful to install gpg inside the chroot, and bind-mount some keyring into the chroot. This will require two things: requirement for bind-mount read-only, and apt using keyring from a directory that it will not try to write to (i.e. not /usr/share/apt/).

15:41:58 # Life Reimplementing autobuilders, by Simon Richter

• Tried to use apt for autobuilding; but apt had its limitations. Dependency cycle and other things.
• Comment from the room: Couldn't it be possible to let apt to unpack files that are downloaded while downloading others? It waits until everything is downloaded before unpacking.

16:04:14 # Life debbugs tips and tricks by Anthony Towns.

• debbugs static web site took more than 24hours to regenerate, which was a problem.
• Audience: debbugs web interface is crap! Somewhat limited links. pkg....cgi takes too much CPU cycle, it takes 10 seconds.
• index files were meant to be generated to speed up, but it wasn't updated for a year...
• email interface needs to be really easy to use. You can even use just telnet to file bugs (by talking smtp). At one time in past, it took 24 hours reflected for the webpage.
• debbugs is working fine for Debian, splitting bugs with packages. This didn't work for gnome/kde, they wanted finer split.
• Finding interesting bugreports is hard.
• We will not hide problems ... we will not track security issues (in BTS).
• google was DoS-ing master, and BTS is currently not google-able.
• Mail interface was down for a few hours for yesterday, but not many people noticed it. Web interfaces are different, people expect instant response.
• debbugs has 15 minute delay in sending email; avoids loops becoming DOS.
• 50,000 active bugs, 300,000 bugs in total. Debbugs is a big thing. Scalability problems. ext3 filesystem in 2.4 much better.
• summary files: new way of storing status for bugs. Now rfc822, colin watson implemented it.
• log files: run away! Has control characters to split data.
• status, report: just ignore them.
• merkel contains a mirror of BTS, look into it for examples. If you are a Debian Developer, you should have access to the machine.
• Whenever you are planning doing any mass-filing of bugs, you will look into the mirror, and talk to your inner-soul, and then discuss on IRC, or mailing list, and add a lintian/linda test.
• Wishlists: Tracking a bug that is automatically filed, closing when the automatic detector no longer has problems. Tracking of regression tests.
• How many people have a regression test that is ran every time a package is built? ( few hands up ). This should be everyone.
• 3,18,33,48 minute of hour, "processall" is ran.
• errorlib is scary perl4 code.
• Version tracking is probably a needed feature
• Hacking debbugs, you will be accepted as a member of developer, since nobody really maintains it right now.

20:02:38 # Life linda by StevenK

• Detects shared library problems, A binary links with libA and libB where libB depends on libC, with libC being the same library libA with different SONAME. Example: bookmarkbridge package which is maintained by Masami Ichikawa, who has already made a c++ transition.
• 0.3, introduces a testsuite.
• Help, with translations.
• Patches would help. Bug with only the symptom without a way to reproduce.
• Adding testsuite.
• junichi: i18n is not really good. Tags are used as msgid's for gettext. Now English text is added as the comment. However, this does not allow translators to notice.
• liw: Is it possible to change the policy to be written in a machine-readable format, so that it can be used as a checker.

20:48:10 # Life Securing the testing distribution, Joeyh.

• Actually unstable doesn't have security
• Most people in the audience used testing
• 5 years of testing, and no security. It required that time to gain experience. Now with team of 10-12 people on mailinglist.
• Have a list maintained automatically. CVS. CAN etc. Claimed by whoever. Check what's happening. check ITP bugs to see if that package is going to be added to Debian. White bugs are security holes in theory, which don't get often fixed, Red bugs are remote security holes.
• CAN-2005-XXXX, why is it XXXX? Found in other place than CV US, so it's not shown up in CV. mdz: CV ID can be assigned if vulnerability is forwarded to matt from testing-security team.
• Is testing-proposed-updates working? I doubt it is.
• Add the CVE in the Changelog, so that we know it's closed.

11 Jul 2005 (Mon)

01:03:36 # Life I tried to debug the problem with pbuilder and su interaction: however, I couldn't get sid to debootstrap on my powerpc machine today. This is strange. aptitude doesn't seem to depend on this package.

The following packages have unmet dependencies:
  aptitude: Depends: libsigc++-1.2-5c102 but it is not installable
E: Unmet dependencies. Try using -f.
	

01:23:53 # Life I've tried to debug problems with 'su' and pbuilder. I'm not quite sure I've grasped the whole problem and how it affects pbuilder and how the real fix should be; but changing a behavior of a very core package because a behavior is "more correct" is rather disturbing.

15:15:02 # Life Debian Developers in Latin America, Gwolf.

• Debian has 900 Developers, 1 Debian developer in 6,666,666 population. Some countries are more densely populated.
• Many geek people who don't speak English at all. Not everybody speaks English. People in Latin America, Spanish or Potuguese find it more. People who visited Porto Allegro should notice.
• Argentina, and in India; Government funded 'cheap computer' with Windows were given out, but that was a method to close in customer. Half-price for Windows license. Linux machines with AMD chips were more inexpensive than government funded.
• It will be better to communicate in their tongue.
• Brazil; had dio Debian largest, as a direct result of having Debconf in that area. Manaus.

16:05:04 # Life Derived distributions round-table.

• Dannf; HP OSLO, HP Debian Enablement. Support for new hardware. We want sarge and support for latest ia64, we have d-i which satisfies both. Predictability and release frequency.
• Branden's question: is 18 month a too long release cycle? Dannf: We have to support woody for the next 10, 15, 20 years, and our 3-year release cycle is almost too short. Bdale: Embedded and telco systems won't be touched for a long time since they are turned on for the first time. Question: You are supporting it 10 years, is it available to the general public? Dannf: hpde.linux.hp.com has the stuff. We have customers mapped to specific version. We support a subset of woody.
• Wookey: Embedded Debian is a collection of tools, rather than a distribution.
• jlicquia, Progeny.
• Petter Reinholdtsen: Some maintainers are easier to approach. Trying to put back changes made for the derived distribution to Debian itself, the difficulty depends on how cooperative the maintainer is.
• Andreas Tille, Debian med:
• Matt Zimmerman, ubuntu. 18 month support, time-based release. Visibility in desktop region, but try to be generic. Independent package archive. Have a build infrastructure. Question: can you change the name of Ubuntu, so that rather than call is it is linux for humans, debian for humans? Ubuntu combined with Debian is a strong partner. Debian should be the link between upstream and distributors, Ubuntu can be the link between users and Debian.
• Roger So. Debian was not usable for users.
• Simon VA Linux J. Enterprise targets. Security issues are important.
• Pere: Past experience tells that people who make the most post in the mailing list are the ones who don't actually do much work in the project.

and /me was busy writing testsuites for dpatch, and adding versioned symbol example to a example shared library implementation.

19:02:21 # Life wookey, on Embedded Debian

• cross building doesn't really work currently, many things are broken.
• Emdebsys; kernel configuration is extended to describe the userland.
• toolchain-source. But we can build cross-compilers from the same source. Too many variations, so not all will be in Debian.
• Embedded used to mean something like 32k, but it's something like 32M. However, it's not as large as desktop machines with 40GB HDD.
• scratchbox has done a lot of good.
• dpkg-stag, stag-get, wrappers for dpkg and apt.

20:34:09 # Life I got hacking on dpatch testsuites. tla tells me arch_commit: unable to acquire revision lock (internal error in archive-pfs.c(pfs_lock_revision)) which seems like a bad news. Apparently, this is caused whenever signing is not properly set up. Locking needs to be unlocked using lock-revision -b command. Confusing.

23:05:05 # Life Since I'm using so much of gworldclock in this side of the World, I decided to hack on l10n. After about an hour of hacking, I got it working, displaying Japanese. 317778 for POTFILES.in, 317784 for ja.po, 317790 for menu fix.

10 Jul 2005 (Sun)

16:12:31 # Life Welcome to debconf.

• Talk by Andrews Schuldei
  • Jesus is great, he buys me milk
  • Video group is working hard

17:05:41 # Life Anthony Towns talks of debootstrap and Debian.

• The debootstrap talk, by Anthony Towns.
• Death of Debian predicted, news at 11:00
• What attracted me to Debian at first, and what happened to them afterwards
  • Runs everywhere, technical, freedom
  • It's not Branden's fault, it's not a Branden's fault
  • 1990's, Debian was about the only community-based distribution
  • Policy is good. Initial design from policy is not a good idea. Policy stuff is not going well now, discussion in debian-policy is not quite fixed.
  • Nowadays in unstable, apt is going into ubuntu before Debian; we don't live a good unstable.
  • Bugs are not resolved
  • Bunch of architectures are supported badly; leading to vancouver prospectus.
• New features in debootstrap;
  • resolvedeps is half-documented. Base is in two parts. Section: base, required, the essential parts. Section: Base. You cannot pre-depend on anything in base, it just depends.
  • Packages file. Section: base sucks. Priority: important is not used for other. Is it useful to have priority(bdale)? required and important didn't have any distinction, but now used by debootstrap. Standard is dselect.
  • cross-debootstrap. Hurd was using i386-linux bootfloppies, and it was important for them.
  • slashdeb? devices. udev, would be better since debootstrap doesn't have to deal with the probelms?
  • gpg support?
  • debootstrap argument format really sucks; maybe use a configuration file; maybe windows.ini format. The exact format should be discussed. It would be difficult to be in XML, since a XML parser in shell would be difficult.

19:16:24 # Life RTFM, by Branden Robinson

• learn to use an macro package with groff. Also, mdoc is used for BSD docs.
• man pages between terse user documentation, and manuscripts
• Sentences should end with a newline, to help groff with space between sentences.
• Don't use the font escapes, \fB; rather use the .RB.
• - is hyphenation dash (which should render narrow on UTF-8 terminal). \(en en-dash, \(em is em-dash, \- is literal hyphen-minus.
• Do use \- for .SH NAME section.
• use .URL macro
• GFDL and GPL is incompatible, be careful in selecting license for manpage.
• Alternatives handling; set manpages as slave links. (what happens if you have 30 translations?)
• The slides will be on the web soon...

20:07:09 # Life Tollef's talk on multiarch.

• DEBIAN/control has dependency on system$ features. system$. Features: mechanisms do have their limitations.
• Branden: rpm feature of being able to depend on a file was more probelmatic in practice.
• The advantage of this approach is that this will eventually give us ability to transition to another architecture. But main problem is complexity.

22:24:34 # Life To configure ssh timeout with your dsh session. The syntax of dsh (dancer's shell) command-line option is sometimes a bit confusing. The -o option may be a bit misleading. ssh itself has an -o option, to confuse matters, and make it look less beautiful. If you want to configure the SSH timeout in dsh, you can use the command-line option -o -o -o ConnectionTimeout=10. I know, the three -o look a bit silly.

9 Jul 2005 (Sat)

20:01:26 # Life On my way to Helsinki; I decided to venture into the World of shared library and how they are relocated. A correct understanding of the mechanics of the art would be required to fully excercise the features of libtool et al to provide satisfactory results, for Debian as a distribution.

8 Jul 2005 (Fri)

07:02:55 # Life su apparently has changed behavior and no longer accepts -- option. It used to take the -- option, and pass it as the command-line; but it no longer does. 317264 317361

7 Jul 2005 (Thu)

08:50:20 # Life I've received an email wrt dsh from a user. The guy pointed me to the IBM dsh manpage, and asked if a specific feature was present in IBM dsh, and why it's not in dancer's shell. Although I would very much like a good dsh in Debian, IBM's dsh is proprietary stuff, and my dsh is developed in my free time according to my needs. Things are kind of different. If you have something specific and determined to get it included in dancer's shell, you are welcome to write a patch.

4 Jul 2005 (Mon)

09:19:14 # Today's hack Hacked on dpatch, and made one more test pass. Good. There are quite a few feature requests and legitimate bug reports. Some of them sound good, some of them sound like maintenance nightmare. I discovered there was unfuzz option to dpep, but it's apparently unimplemented from the look of it. hmmm..

Make sure that I don't forget, here is a copy of my mail, requesting the new key to enter Debian Keyring.

3 Jul 2005 (Sun)

10:50:01 # Life I was discussing with gotom last night, and he said that DMA transfer bugs do exist when cache lines aren't flushed properly. This bug 277309 on xserver nv driver feels like one of those bugs, since it's forever waiting.

After refreshing some memory, I've remembered that the upstream bug tracking system does mention something similar, and the source code seems to mention something similar.

void NVSync(ScrnInfoPtr pScrn)
{
    NVPtr pNv = NVPTR(pScrn);

    if(pNv->DMAKickoffCallback)
       (*pNv->DMAKickoffCallback)(pScrn);

    while(READ_GET(pNv) != pNv->dmaPut);

    while(pNv->PGRAPH[0x0700/4]);
}

	

2 Jul 2005 (Sat)

13:13:14 # Life pbuilder hacking. Since apt in experimental is now in sid, I'm kind of forced to release the new version of pbuilder. Looking for a sponsor since I'm not quite sure if my new key has entered Debian keyring. I wonder when it will, if it ever will.

On the other hand, Matt Kraai has found the problem with debootstrap 314858. This is a pleasant thing to know. a few people are reporting find | xargs rm is giving 'argument list too long' error. But that shouldn't really happen, and that needs to be pursued, as a problem on findutils.

I was a bit overloaded a few days without doing much hacking. It didn't help a bit when my travel agent called me and told me they failed to reserve the flight from Tokyo to Paris; and they needed to show me an alternative route. It so happens now that I'll be staying at Charles-de-Gaule for the first night of the trip, and my trip is longer by 1 day. Bleh.

I've fixed pbuilder to get to the point of working. It's pending 315914 fix, which is required to get debootstrap functional again. I worked around with manually adding libslang2 in /usr/lib/debootstrap/scripts/sid:work_out_debs(). required="$(get_debs Priority: required) libslang2"

After all this, I've got pbuilder testsuite result to look like it's passing 9/10 tests. The remaining 1/10 requires a fixed pbuilder actually installed in sid. Not bad.

[OK]   create-sid
[OK]   build-sid-dsh
[OK]   pdebuild-sid-dsh
[FAIL] pdebuild-internal-sid-dsh
[OK]   execute-sid
[OK]   create-sarge
[OK]   build-sarge-dsh
[OK]   pdebuild-sarge-dsh
[OK]   pdebuild-internal-sarge-dsh
[OK]   execute-sarge
	

1 Jul 2005 (Fri)

06:31:28 # Life It's scary that it's already time to look at going to Finland. I've not been making much progress with things I would like to have been hacking on. I haven't done enough research into dpatch; I haven't yet fixed debootstrap. pbuilder is broken now that secure apt has entered sid. Hmmm..