Hello, Here at DebCamp, Ian and I are in the midst of the various bugs needed to support pbuilder users much better. Some changes made today and yesterday invalidate some of the discussion on this bug. Here are some notes on how I now think this is going to work, based on reading this thread again, reading some manpages, and some experimention.
1. There should be two dgit subcommands, pbuilder and cowbuilder, which work in exactly the same way except they call pbuilder or cowbuilder respectively. 2. We definitely cannot use pdebuild because dgit now always builds its own source package. 3. AFAICT pbuilder must be run as root, even if your base chroots are stored somewhere in your home directory. pdebuild's default is to use `sudo -E` so presumably dgit should use `sudo -E` too. If someone asks we can make it configurable. 4. The options we will always have to pass are --distribution, --buildresult and `--debbuildopts -b` (for comparison, `dgit sbuild` passes only --distribution and -A in some cases I think). 5. We probably want to forcefully unset SOURCE_ONLY_CHANGES. 6. We expect the user to configure --build-products-dir in their dgit config and --buildresult in their pbuilder config. We do not look at their pbuilder config (see previous discussion). 7. We are going to dump the whole source package, including the .orig, into --build-products-dir/--buildresult, change to that directory and then invoke `pbuilder b foo.dsc`. (Indeed, with changes made yesterday and today, dgit now always does this except when you do e.g. `dgit build -b`.) In the thread Mattia said that pbuilder tries to be clever in copying .origs around a bit, and the assumption seemed to be that you invoke pbuilder from a dir other than --build-products-dir/--buildresult. We might have to patch pbuilder to tolerate what dgit is now doing, but maybe we won't have to -- not sure yet. 8. Ian's wip.pbuilder-config is probably not useful at this point. -- Sean Whitton
signature.asc
Description: PGP signature
