Created attachment 335600 [details] net-misc:youtube-dl-2013.01.13:20130114-125530.log - example build log Running ebuild(1) without FEATURES=test USE=test set will silently enable them. After this, `ebuild ... test install' runs the test phase twice when the first iteration fails, because apparently the test phase must be run before the install phase.
Do you have FEATURES=test-fail-continue enabled? If so, the reported behavior is as intended. When you pass the "install" argument, it also implies phases on which it depends, which includes the "test" phase. So, it's redundant for you to request both test and install. Do you want it to automatically eliminate the redundant "test" argument on your behalf?
The install phase depends on the test phase by design? If that is what causes rerunning it, what is the rationale for that dependency?
Well, it's just based on the order the phases have always gone in. For autotools based packages, it's typical to do `make ; make check ; make install`, so that's how our phases have always been ordered.
Ah, I momentarily assumed you meant a hard dependency between the two phases. Could this all be because normally you'd see: >>> It appears that 'test' has already executed for 'adobe-flash-10.3.183.50'; skipping. >>> Remove '/home/jer/portage/www-plugins/adobe-flash-10.3.183.50/.tested' to force test. but that now it's omitted since the test phase failed, despite having test-fail-continue? I guess I would expect the test phase to be marked completed when that feature is on, despite its failure.
(In reply to comment #4) > I guess I would expect the test phase to be marked > completed when that feature is on, despite its failure. Okay, done: http://git.overlays.gentoo.org/gitweb/?p=proj/portage.git;a=commit;h=70bfb6b8ebdb74fe5fbe244532e35f86cab9742e
This is fixed in 2.1.11.42 and 2.2.0_alpha153.