* checking ebuild checksums ;-) ... [ ok ] * checking auxfile checksums ;-) ... [ ok ] * checking miscfile checksums ;-) ... [ ok ] * checking porthole-0.4.1.tar.bz2 ;-) ... [ ok ] !!! ERROR: app-portage/porthole-0.4.1 failed. Call stack: ebuild.sh, line 1559: Called dyn_setup ebuild.sh, line 666: Called pkg_setup porthole-0.4.1.ebuild, line 22: Called built_with_use '<dev-python/pygtk-2.8.0-r2' 'gnome' eutils.eclass, line 1592: Called die !!! Unable to determine what USE flags was built with !!! If you need support, post the topmost build error, and the call stack if relevant. This is related to but NOT a duplicate of bug 110961. The problem is that built_with_use was recently changed to error out if the atom cannot be resolved to a vdb entry. To restore the old semantics, porthole needs to use if has_version '<dev-python/pygtk-2.8.0-r2' && ! built_with_use '<dev-python/pygtk-2.8.0-r2' gnome ; then ...
Hm, except that might resolve to a 0.6.11 version. Better: if ! has_version '>=dev-python/pygtk-2.8.0-r2' && ! built_with_use '<dev-python/pygtk-2.8.0-r2' gnome ; then The short-circuiting && will prevent any die()ing.
(In reply to comment #0) > The problem is that built_with_use was recently changed to error out if the > atom cannot be resolved to a vdb entry. vapier, could you comment? Why was built_with_use changed? It always returned successfully so that package.provided could be made to work, and I was relying on that behaviour (even if for a different reason).
(In reply to comment #2) > vapier, could you comment? Why was built_with_use changed? It always returned > successfully so that package.provided could be made to work, and I was relying > on that behaviour (even if for a different reason). Because it doesn't make sense to assume something about stuff that's not even installed or that you can't verify (such as package.provided). See Bug 139842 (also, http://bugs.gentoo.org/attachment.cgi?id=93613&action=view). This change has already uncovered quite a few bugs in ebuilds checking for use flags on non-dependencies as a side effect, seems like a good thing (TM) ;)
(In reply to comment #3) > Because it doesn't make sense to assume something about stuff that's not even > installed or that you can't verify (such as package.provided). See Bug 139842 > (also, http://bugs.gentoo.org/attachment.cgi?id=93613&action=view). Thanks for the pointer. I think the current built_with_use is wrong because of Aug 05 23:52:59 SpanKY if the package is provided, then you have to assume the user didnt fuck it up Aug 05 23:53:09 SpanKY otherwise it'd be impossible for user to install Aug 05 23:53:14 SpanKY they'd have to edit the ebuild > This change has already uncovered quite a few bugs in ebuilds checking for use > flags on non-dependencies as a side effect, seems like a good thing (TM) ;) but I can agree that it would be reasonable to die if a package is neither installed via portage nor in package.provided, so porthole ebuild updated, and I'll ask about the other case later and elsewhere.
i thought about it some more and decided that you cant assume one way or the other (all flags enabled or all flags disabled) in the built_with_use function as that'll restrict the usability of the function Bug 142941 tracks the required functionality for further discussion, there is a thread on the gentoo-portage mailing list
