Aha, after some digging, I realized that Launchpad uses a forked copy
of sbuild
(https://code.launchpad.net/~canonical-launchpad-branches/launchpad-buildd/trunk).
So it's not clear if this affects Debian's sbuild. Feel free to close.
-mt
On Mon 30 Jul 2012 12:38:16 EDT, Roger Leigh wrote:
On Mon, Jul 30, 2012 at 10:50:35AM -0400, Michael Terry wrote:
Consider the case of libio-socket-ip-perl in quantal as of this writing. It
has a dependency like "libsocket-perl (>= 1.97) | perl (>= 5.15.6)".
libsocket-perl's current version is 2.002-1 (which is fine). Perl's current
version is 5.14.2-12 (which does not satisfy the check).
So sbuild correctly reports:
libsocket-perl: missing
perl: non-matching version installed (5.14.2-12 ! >= 5.15.6)
And then does not try to install libsocket-perl. Which I think is a clear
bug? The build fails because it is dep waiting on libsocket-perl, despite it
actually being available.
It seems that sbuild's dependency checker stops if part of an or-clause is
already installed, regardless of whether it actually satisfies the
requirements. I'd argue it should take that into account and attempt to
install another part of the or-clause.
Which dependency solver were you using? I'm assuming it wasn't
internal, which should be removed in this version.
I think you're probably running afoul of the RESOLVE_ALTERNATIVES
setting, which for the apt resolver default to 0. That is, all
alternative build dependencies will be stripped out. You should
be able to confirm this by looking at the dependencies in the
dependency package--they should be printed as part of the build
log. You should see the filtered dependencies as having all the
alternatives removed.
On the Debian buildds, RESOLVE_ALTERNATIVES is always 0, except
for the experimental and backports buildds which use aptitude and
are out of necessity more flexible, but less reproducible. For
all stable/testing/unstable builds, alternative build dependencies
are not allowed. I'm not sure what Ubuntu buildds do.
I hope that helps explain things. If my assuption is wrong, I'll
be happy to look at it further.
Regards,
Roger
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
