On Mon, Aug 18, 2003 at 10:02:32AM +0100, Andrew Suffield wrote:
> On Sun, Aug 17, 2003 at 09:38:54PM -0500, Joe Wreschnig wrote:
> >
> > What causes FTBFS bugs now:
> > 1. Bad Build-Depends.
>
> These are handled by the current system - they're trapped on the buildd.
Only if it fires - which it doesn't for Arch: all packages.
> > 2. Broken code that doesn't port (either to another architecture, to
> > something that's not the maintainer's home directory, whatever).
>
> And so are these.
See above (barring 'to another architecture').
> > All of 1 and many of 2 would be solved by binaryless uploads. A new
> > type, "maintainer never tried to build package ever (and so debian/rules
> > has a syntax error or debian/control has a bad version or whatever)"
> > would be introduced - but such packages would be blocked from entering
> > the archive anyway.
>
> But none of this is new. These things are already handled by the
> system we currently have in place.
Only if it's other than Arch: all; otherwise, it never enters the system.
Which is the gap that causes the problems which keep starting this thread
over and over.
> What you are proposing to do is essentially to require that a package
> has built on all architectures before installing it into the
> archive. We have that already: it's called "testing".
It isn't the 90% of packages this helps, it's the 10% of packages which
*don't* go through it that trigger the debate.
So, in the interest of proposing a (hopefully) useful alternative, which
doesn't cause 11 buildd failures because of a flat-out binary-less upload
policy:
[ Proposal - capitalized words should be considered to have RFC-style ]
[ meanings, because I want to be clear and I'm sure this would be ]
[ rewritten before going into any actual policy. ]
Uploads to the archive MAY contain compiled package files for any of the
currently supported architectures; they SHOULD (MUST?) contain at least one
full set of architecture-specific package files.
Uploads MUST NOT contain architecture-independant package files; for
each upload, one build daemon will be selected to build and upload the
architecture-independant packages, in the same general manner as building
the architecture-dependant packages for architectures which were not
uploaded (this is not guaranteed to occur on the same run as a build of any
architecture-dependant packages; it is not even guaranteed to occur on the
same build daemon).
[ End ]
True, it's possible that someone really dumb might do a binary-arch build
and upload that, without ever testing the binary-indep; but under this
proposal, we'd only waste *one* build-run worth of time, not 11. And since
the results are (theoretically) Arch: all, it can run on *any* buildd with
spare cycles at the time (or there could be a dedicated one).
Note that recovering from a bad upload (one with Arch: all packages
included) is fairly trivial: just discard them and send a nastygram to the
uploader reminding them that it violates policy.
--
Joel Baker <fenton@debian.org> ,''`.
Debian GNU NetBSD/i386 porter : :' :
`. `'
`-
Attachment:
pgp81j58p1VPL.pgp
Description: PGP signature