Bootstrap/Crossbuild Sprint
Location, Date
- when: March 27 - 30, 2025
- where: 97082 Würzburg, angestoepselt e.V., Zeller Straße, 29/31
- local contact: josch
Participants
Please add yourself to the list below if you wish to attend. There will be space for 9 participants.
No |
Name |
confirmed |
booked accom. |
Affiliation |
Country |
Arrival at |
Departure at |
|
1 |
Johannes Schauer Marin Rodrigues |
Yes |
local |
Debian |
DE |
|
|
|
2 |
Jochen Sprickerhof |
yes |
yes |
Debian |
DE |
27 |
30 |
|
3 |
Helmut Grohne |
yes |
yes |
Debian |
DE |
27 |
30 |
|
4 |
Gioele Barabucci |
yes |
no |
Debian |
DE |
|
|
|
5 |
Holger Levsen |
yes |
yes |
Debian |
DE |
26 |
29 |
Agenda
(tentative/final agenda here before/after the sprint)
Discuss 1065416 (linux-libc-dev arch:all/arch:any + linux-libc-dev-$arch-cross) and try to figure out previously unexplored approaches to the various requirements
- None of us has a good understanding of why arch:all would be beneficial. The downsides of that approach only really matter for new architectures, so the trade-off may still be sensible.
1069065 moving -for-host packages to rtlibs stage
798955 multiarchified glibc headers: Is it worth resuming work on this?
Can we drop libcrypt-dev from build-essential?
- TODO: Approach Sanvila about an archive rebuild and mbf packages, then drop.
- Can we reduce perl's cross configs from 1200 lines per architecture to something manageable?
How to deal with packages expecting bare toolchains when built natively?
- Before fixing this issue, we should ensure that triplet-gcc actually uses triplet-prefixed binutils. This can be achieved by symlinking the triplet-prefixed tools into the gcc installation. Additionally, gcc-triplet needs to depend on binutils-triplet without :any to ensure that linker plugins have matching architecture. Once these are fixed, solutions involving native-architecture-is may be worth exploring.
Issues related to mmdebstrap that need to be looked at:
dpkg still reads configuration from the host: 808203
interaction between dpkg-preconfigure and chrootless: 1091442
dpkg will use its own architecture as the native one in chrootless mode: 825385 and 1020533
information should not be hardcoded in mmdebstrap but move to distro-info-data: 993589 993590 (will allow the removal of 800+ LOC)
What's missing to M-A:same build-essential? 815172
- Issues related to dpkg-root-demo salsa CI pipeline testing chrootless mode:
cherry-pick https://git.hadrons.org/cgit/debian/dpkg/dpkg.git/commit/?h=next/main&id=605cdfcbaec0e)--
talk about integration of GNU/Hurd
unreproducible /var/cache/debconf/config.dat-old and /var/cache/debconf/templates.dat-old
unreproducible /etc/ld.so.cache
- sbuild improvements for crossbuilding
- Discuss sustainability of cross build support (requires continuous effort, close to no progress for years, what should be prioritized?)
- What does it take to have QA/CI for build profiles?
- Possible goals:
Check for FTBFS with build profiles enabled (though some e.g. upstream-cargo intentionally FTBFS without internet, nocheck FTBFS is rc though)
- Check for reproducibility (which profiles do we expect to be reproducible?)
Check for missing profile annotations (suggest <!nocheck> annotations)
- implementation / maintenance / computing resources
- crossbuilding as salsa CI job
- Possible goals:
- write a small blurp about what bootstrapping and crossbuilding is and why it is useful and why we are doing this which can then be used for publicity purposes
What is preventing us from replacing gcc's DEB_BUILD_OPTIONS=nolang=... with build profiles?
1099587 build profiles in policy?
- fix bootstrap.d.n
- Can we use busybox without subverting the dependency system somehow?
- Getting rid of multilib
Reports
(pointers to communication about the sprint)
- announcement: (e.g. link to lists.d.o archives)
- report: (e.g. link to lists.d.o archives)
- other:
- other:
- ...
Acknowledgements
the sprint has been possible thanks to:
donations to the Debian project
- sponsor ...
- venue ...
- etc ...