Currently, asking for nmu foo_1.0-1 . ALL . -m 'rebuild against bar 2.3' only builds foo_1.0-1+b1 for arch:any packages. No +b1 is built for any possible -doc packages. Often, this is what's expected, but not always. The specific scenario I have in mind is Haskell libraries. They change ABIs often, at least once every time ghc6 gets a new upstream version. That's a lot of packages, needing rebuilding and dep-wait states, with no source changes. Haskell library -doc packages include .haddock files, which are derived from the ABI at build time. They describe the library's API and are used at build time for things like generating correct links in depending libraries' documentation and for generating an index on a user's system. They are itself architecture independent, but still need to be rebuilt along with the libraries themselves. Building and uploading the -doc package corresponding to the binNMU by hand is possible even now (as described in http://lists.debian.org/debian-haskell/2009/05/msg00029.html) but not really any more convenient than doing a sourceful upload, as it stands. It's a hack, too. Plus, it'd be preferable if this was all part of the usual buildd network. I don't know offhand if this has been discussed before. Any reasons why this shouldn't be supported? What's needed to get this working? I guess it'd be more proper if there was some other token that'd trigger building arch:all packages, instead of including it in ALL.
Attachment:
signature.asc
Description: Digital signature