Running emerge -puDN world, I get the following: [ebuild R ] sys-libs/zlib-1.2.3-r1 [ebuild U ] dev-python/sip-4.6 [4.5_pre20061021] [ebuild R ] sys-apps/sed-4.1.5 USERLAND="(GNU%*)" [ebuild N ] dev-lang/tcl-8.4.14 USE="threads -debug" [ebuild U ] dev-python/qscintilla-1.7.1 [1.6] [ebuild N ] dev-scheme/guile-1.6.8 [ebuild R ] sys-apps/dbus-1.0.2-r2 KERNEL="(linux%*) (-FreeBSD)" [ebuild R ] app-admin/gamin-0.1.8 [ebuild R ] media-plugins/alsa-plugins-1.0.14_rc1-r1 [ebuild R ] app-i18n/koffice-i18n-1.6.3 LINGUAS="de* -fr*" [ebuild UD] dev-util/desktop-file-utils-0.12 [0.13] [ebuild R ] kde-base/kalyptus-3.5.4 [ebuild UD] dev-db/xbase-2.0.0 [2.0.0-r1] [ebuild N ] dev-tcltk/expect-5.42.1-r1 USE="doc -X" [ebuild N ] sys-devel/autogen-5.9 [ebuild R ] sys-libs/libkudzu-1.1.62-r1 USE="zlib%*" [ebuild R ] x11-libs/qt-4.3.0 USE="qt3support*" [ebuild R ] sys-apps/module-init-tools-3.2.2-r3 USE="-old-linux%" [ebuild R ] app-text/ghostscript-esp-8.15.3 [ebuild R ] dev-libs/DirectFB-0.9.25.1 [ebuild U ] dev-python/PyQt-3.17.2 [3.16.1_pre20061021] [ebuild N ] dev-util/dejagnu-1.4.4-r1 USE="doc" [ebuild NS ] dev-python/PyQt-4.0.2_pre20061021 USE="doc -debug" [ebuild R ] sys-apps/hwsetup-1.2 USE="zlib%*" [ebuild R ] sys-devel/binutils-2.17 USE="test*" [ebuild R ] sys-devel/gdb-6.6-r2 USE="test*" [ebuild U ] dev-python/pykde-3.16.0-r1 [3.16.0] [ebuild R ] sys-devel/gcc-4.1.2 USE="mudflap* test*" You can ignore most of the crap, but please notice the pointless rebuilds, e.g. zlib or ghostscript. Beside this you see also the test use flag induced rebuilds. Form my POV it would be better, if the test use flag would supercede the feature flag, when set in /etc/portage/package.use. Different bug or do you all strongly disagree?
(In reply to comment #0) > You can ignore most of the crap, but please notice the pointless rebuilds, e.g. > zlib or ghostscript. Apparently those were flag removals ("build" for zlib and "emacs" for ghostscript-esp). I'll fix it so those get displayed. > Beside this you see also the test use flag induced rebuilds. Form my POV it > would be better, if the test use flag would supercede the feature flag, when > set in /etc/portage/package.use. Different bug or do you all strongly disagree? > So, you want the flag in package.use to enable the test phase even when FEATURES=test is disabled, and you want -flag in package.use to disable FEATURES=test even though it is enabled? Maybe we can do something like that. Unfortunately, circular interactions like this often lead to strange logic that is annoying to maintain. FWIW, you can already use /etc/portage/profile/package.use.mask to explicitly disable tests on certain packages.
Created attachment 126192 [details, diff] show --newuse flags that trigger reinstallation even when not in --verbose mode This is fixed in svn r7408.
This has been released in 2.1.3.
> So, you want the flag in package.use to enable the test phase even when > FEATURES=test is disabled, and you want -flag in package.use to disable > FEATURES=test even though it is enabled? Maybe we can do something like that. > Unfortunately, circular interactions like this often lead to strange logic that > is annoying to maintain. Yes, I can follow you here. It's just irritating to have a feature that sort of breaks distinct functionality. > FWIW, you can already use /etc/portage/profile/package.use.mask to explicitly > disable tests on certain packages. Didn't thought about that. Will work for me.
Right now the relationship between FEATURES=test and USE=test is one-way-only, the feature adds (but doesn't remove if disabled) the use flag, so if you mask the use flag the test phase would still run (but probably fail if the additional dependencies are missing). And I'd rather not hack another special case into portage to make it bi-directional, the real solution you're after would be package.features.
