30 Jun 2004
Finally sat down for keysigning.
There are so many keys I'd like to sign, and
I'm rather exhausted after reviewing so much.
Much left to do. It's going to take several days...
26 Jun 2004
python2.3-dev doesn't depend on python-dev, since when has that done that?
Is it since 6 June upload?
That's breaking a few things which I don't really want to be bothered about.
20 Jun 2004
Chatted about pbuilder on IRC.
I've not been on IRC often recently, due to my lifestyle change.
I have, however, relied on IRC while I was attending debconf, and
scheduling with people.
#debian-devel is one of the channels I feel most at home with,
and I like that. Time and connectivity is a big constraint now.
I used to be able to IRC all the time with Libretto and AirH"
connected to IRC channel all the time while I was travelling,
but I no longer tend to do that.
I probably need to get my Zaurus fixed up to get connectivity
while I'm moving around.
I have really almost forgot about it, but my blog entries have
a big problem that they don't have a 'title' for syndication.
joshk pointed it out to me; maybe I'll have a hack at adding a
title for dancer-diary.el.
I used to think titles are really unnecessary for blogging,
but Since I no longer really use 'sections', and
I would consider titles describing an entry to be more useful.
16 Jun 2004
Trying to do bug maintenance.
Bug reports are usually clueless.
'This doesn't work' kind of bug report is usually difficult to
analyze and dissect.
15 Jun 2004
I've wrote a little test to test the library version symbols.
It is available for download.
make clean; make test
Will start a test. It has a README file for what it does.
10 Jun 2004
Being back at real life, and getting back to work after debconf
has been an interesting experience.
There's so much work to do as usual, and I am having
problems getting up to speed with everything.
Uploading a few packages, reading mail backlogs,
I'll probably need another month to get back to normal operation.
6 Jun 2004
Now I'm back at home.
Uploaded dsh packages to Debian archive, to reflect my
I'm too tired to do anything more at the moment,
maybe I'll recollect myself later.
2 Jun 2004
Updated libpkg-guide from Stephen Frost's comments.
1 Jun 2004
New month starts in Japan time, it's still end of May in Brasil.
Memo on lamont's talk.
- Upload; queue daemon; archive maintenance tools: basic checks, "accepted" queue, wanna-build shows as "Needs-Build"
- Packages-arch-specific, used now to restrict packages to specific architectures, for quinn-diif et al. cvs.debian.org:/cvs/dak.
Do not try to spit out infinite output to stdout.
- order of build in sections
- once-clean and hopefully minimal chroot, unlike pbuilder.
Some buildd maintainers chose to leave build-depends there to
avoid the FTBFS.
buildds do not get build-dep-indep.
arch-all packages are not quite handled.
Installs build-depends in chroot
dpkg-buildpackage -B: fakeroot debian/rules clean , debian/rules build, fakeroot debian/rules binary-arch
Tries to restore the chroot by removing packages.
Some packages fail to remove cleanly, which shows up.
Result is sent as mail: few hundred emails a day.
FTBFS: missing build-depends, term coined by
How did you get your package build on your machine?
- Build-depends missing
- root activities in build target cannot be done as policy says
- installing in /
- depending on existence of a user
- locales package doesn't install, because
of arch-all and arch-any problem.
- chroot maintenance: it's a bad idea to upgrade while a package is building.
until around 2003, packages built today was installed the next day; we introduced "accepted buildd"s.
pimp graphs: 2002-2003 percentage was counted against.
At 37%, bdale was satisfied with the available packages.
"pimp graph" to motivate buildd admins.
Please shrink build logs, if it goes over 10Mb.
Re-uploading on build failure and experimenting until it builds
is not a good way of using the buildd resources.
kov talk on Brazil free software.
I've missed most of the talk because I overslept, gah.
- One-man teams. One person doing debconf templates, one person doing package description. 90% of work is done by single work.
- Territorialism. translator gave a dotted list instead of a comma-separated list, and maintainer disagreed on it.
Translators are given lower status.
Translators are not seen as first-class development resources.
Not seen as part of Debian, lack of identity.
Debian CONTRIBUTOR status would be good.
13 developers in Brazil.
No signed keys in Argentina until now?
Would be nice to have a user-keyring like developer-keyring?
Argentina, there is no debian developer in there, but we are trying to use it.
vorlon: BTS is restricted in English, is there a solution in Brasil?
kov: We have such a project, we just do not have the manpower.
vorlon: samba package supports english, spanish, xxx,xxx.
kov: But that's a developer feature.
Users finding problems and reporting to local user groups,
there is a barrier.
Many tools were created by non-hackers.
democracy etc are tools created by non-hackers.
We are hackers; act like hackers, and create tools.
Code is a political instrument. Stop acting like politicians and start hacking!
Bug report could get some automatic translation?
Bug reports have low information content.
It would be nice to have some KDE BTS.
Most of bug reports don't have a good quality.
Openoffice bug reports are only forwarded to developers after
a sequence to reproduce the problem are known.
Implementing such two-tier method;
with first layer in the local language should be possible.
Should Debian keep secrecy ?
It should really be public, because it's embarassing when
some processes are not transparent and we discover that
something has not been done for a long time.
I've updated the libpkg-guide to include notes on
methods to add versioned symbols to a shared library.
Multiarch proposal by Matt Taggart and Tollef Hog Heen.
Matt explains upstream changes.
- i386/x86_64, ppc/ppc64, s390, s390x, sparc/sparc64, mips/mips64, sh/sh64, ia64 etc.
i386/x86_64, RH and SuSE use /lib/ and /lib64/.
ia64 uses /lib for 64-bit.
- Mixed instruction sets. i386/ia64 (hwemu), i386/any qemu,
s390/any (hercules), ia64/i386(ski) etc.
Potential emulation cases.
- ia64 Debian, /emul/ia32-linux,
David Mosberger designed, hardcoded in kernel?
- qemu advocate /usr/local/qemnu-arch/
- Mixes OS environment.
- Current practice;
solaris on linux: /usr/gnemul/sunos/,
others use pseudoroots or chroots (wine etc.)
Mixed endian systems. arm, hppa, ppc, ia64, mips/mipsel.
Mixed ABIs; ABI events?
packages with architectures-based directory: glibc, libstdc++, libssl, octave, emacs etc.
We have a big mess! can we solve it?
FHS changes, prefix/lib/arch-os
/lib64 is a symlink to /lib/arch-os
FHS2.3 added them.
/lib/ld.so.version moves to /lib/i386-linux/ld.so.version
Tollef Hog Heen
- Follow upstream changes, will need change to all binary-arch.
tools such as pkgconfig will help the migration,
packages using autoconf etc should be better.
Conflicts? /usr/share/doc/libfoo moved to -common or -dev.
Will require a symlink in /usr/share/doc/libfoo.
ld.so will search LD_LIBRARY_PATH; plugins on dlopen? Depends?
substvars don't work with package name in debian/control.
ajt's proposal requires versioned provides implemented.
sed magic should work.
tfheen's proposal requires change in dpkg/apt/etc.
Master plan, all library goes multi-arch, and
sarge+1 releases 2010,
by 2038 noone uses x86 so that we can remove compat.
ldl was designed to be able to move around.
It's not impossible.
$Id: 200406.html.en,v 1.18 2004/06/29 23:30:49 dancer Exp $