Buildd/Wanna-build BOF, debconf13 2013-08-15 enabling multiarch in the buildds ---------------------------------- - allows mingw to specify build-dep on wine:i386 - allows cross-compiler builds directly against libc:host - colinW: has sent patches to sbuild for :any and :native fixes. Not clear what is needed to be sure that the version actually in use has those fixes. built-using auto-update ----------------------- - mingw has 'built-using' fields for binutils, gcc. Would be nice if something caused rebuilds to occur when those packages have new uploads. - some concern about churn from uploads. wanna-build docs/setup ---------------------- - complaint that everything has changed and newer stuff not documented - Sylvestre: GSOC project last year docuemnted WB setup - General issue of WB setup being hard (and overkill for many things) - Desire to make ports.net infrastructure ------------------------ - Do we need to make changes? aurel overloaded - pkern can fix things too - machine is quite full (due to edos-debcheck checks) - is currently bytemark VM provided by DSA, but not administered by them - We should aks for more hardware. Currently recalc takes 10 mins of 15-min update period. DSA have more hardware, apparently - we should ask. - Otherwise things are working OK and no obvious reason to change. - Integrating debian-ports.net into debian proper first steps: 1. Support mixed archive with unreleased suite, maybe ppaext helps. 2. Merge wanna-build databases. 3. Uploaders/Porters might not be DDs saving debug data on failed builds ---------------------------------- - core dumps (maybe create dh_coredump) - keep preprocesed C files, when compiler ICE - config.log (some packages, GNOME, dump config.log upon failure) - snapshot of whole build dir - size issues (maybe keep for a week to give developers a chance to look at) sbuild has some functionality to store things on failure already. for core dumps they need to be enabled on the machine. apport may be useful to ensure something sensible is done with the dumps a 'failed:' target in the debian rules was sugested as a good way to define what should be saved/done on falure.