emerge -av1 dev-perl/Test-NoWarnings dev-perl/Net-SMTP-SSL These are the packages that would be merged, in order: Calculating dependencies... done! [ebuild R ] dev-perl/Net-SMTP-SSL-1.30.0::gentoo USE="{test}" 0 KiB [ebuild N ] dev-perl/Test-Tester-0.109.0-r1::gentoo 0 KiB [uninstall ] perl-core/Test-Simple-1.1.14-r1::gentoo [ebuild R ] dev-perl/Test-NoWarnings-1.40.0-r2::gentoo 0 KiB [blocks B ] <dev-perl/Test-Tester-0.114.0 ("<dev-perl/Test-Tester-0.114.0" is blocking perl-core/Test-Simple-1.1.14-r1, virtual/perl-Test-Simple-1.1.14-r1) Total: 3 packages (1 new, 2 reinstalls, 1 uninstall), Size of downloads: 0 KiB Conflict: 1 block (1 unsatisfied) * Error: The above package list contains packages which cannot be * installed at the same time on the same system. (virtual/perl-Test-Simple-1.1.14-r1:0/0::gentoo, installed) pulled in by virtual/perl-Test-Simple required by (dev-perl/Net-SMTP-SSL-1.30.0:0/0::gentoo, ebuild scheduled for merge) (dev-perl/Test-Tester-0.109.0-r1:0/0::gentoo, ebuild scheduled for merge) pulled in by >=dev-perl/Test-Tester-0.107 required by (dev-perl/Test-NoWarnings-1.40.0-r2:0/0::gentoo, ebuild scheduled for merge) (You may replace dev-perl/Net-SMTP-SSL-1.30.0 by any other package which depend on virtual/perl-Test-Simple-1.1.14). What is strange about that is that dev-perl/Test-NoWarnings/Test-NoWarnings-1.40.0-r2.ebuild states RDEPEND="|| ( >=virtual/perl-Test-Simple-1.1.10 >=dev-perl/Test-Tester-0.107 )" i.e. Test-Tester can be substituted by perl-Test-Simple. If I mask Test-Tester, then everything works out: [ebuild R ] dev-perl/Test-NoWarnings-1.40.0-r2::gentoo 0 KiB [ebuild R ] dev-perl/Net-SMTP-SSL-1.30.0::gentoo USE="{test}" 0 KiB The question is why does portage insist on installing a (unnecessary and) blocking dependency? Running portage 2.2.28.
I had similar issues with perl-5.24 on two machines. Running emerge with "--backtrack=99" let portage eventually find a workable solution. I'm not sure why portage chose the initial path it did deciding to install a blocker first then work around it. Might turn out to be as simple as the order of possible solutions in one or more virtual. Or something like that, whatever.
Tried it with "--backtrack=99", 200 and 1000. In all three cases portage still tries to merge the blocker first.
Please can you submit a full build.log and per https://bugs.gentoo.org/show_bug.cgi?id=577748#c1 an emerge -vt to aid the perl team? Thanks.
Michael Everitt (IRC: veremit) wrote on 2016-06-04 00:26:11 UTC: > Please can you submit a full build.log and per https://bugs.gentoo.org/show_bug.cgi?id=577748#c1 an emerge -vt to aid the perl team? Thanks. Sure. The output of emerge -vt is quoted below. A build.log cannot be supplied because nothing is build. ---- # emerge -av1t dev-perl/Test-NoWarnings dev-perl/Net-SMTP-SSL These are the packages that would be merged, in reverse order: Calculating dependencies... done! [ebuild R ] dev-perl/Test-NoWarnings-1.40.0-r2::gentoo 0 KiB [nomerge ] dev-perl/Net-SMTP-SSL-1.30.0::gentoo USE="{test}" [nomerge ] virtual/perl-Test-Simple-1.1.14-r1::gentoo [uninstall ] perl-core/Test-Simple-1.1.14-r1::gentoo [nomerge ] dev-perl/Test-NoWarnings-1.40.0-r2::gentoo [ebuild N ] dev-perl/Test-Tester-0.109.0-r1::gentoo 0 KiB [ebuild R ] dev-perl/Net-SMTP-SSL-1.30.0::gentoo USE="{test}" 0 KiB [blocks B ] <dev-perl/Test-Tester-0.114.0 ("<dev-perl/Test-Tester-0.114.0" is blocking perl-core/Test-Simple-1.1.14-r1, virtual/perl-Test-Simple-1.1.14-r1) [blocks B ] <dev-perl/Test-Tester-0.114.0 ("<dev-perl/Test-Tester-0.114.0" is blocking perl-core/Test-Simple-1.1.14-r1, virtual/perl-Test-Simple-1.1.14-r1) Total: 3 packages (1 new, 2 reinstalls, 1 uninstall), Size of downloads: 0 KiB Conflict: 1 block (1 unsatisfied) * Error: The above package list contains packages which cannot be * installed at the same time on the same system. (virtual/perl-Test-Simple-1.1.14-r1:0/0::gentoo, installed) pulled in by virtual/perl-Test-Simple required by (dev-perl/Net-SMTP-SSL-1.30.0:0/0::gentoo, ebuild scheduled for merge) (dev-perl/Test-Tester-0.109.0-r1:0/0::gentoo, ebuild scheduled for merge) pulled in by >=dev-perl/Test-Tester-0.107 required by (dev-perl/Test-NoWarnings-1.40.0-r2:0/0::gentoo, ebuild scheduled for merge)
CCing dev-portage as per dilfridge request. Previously upstream, dev-perl/Test-Tester was a standalone CPAN module as was dev-perl/Test-use-ok However, the upstream dist Test-Simple (which is also shipped in perl itself) incorporated both into version 1.1.10 This means there's a problem in portage where things that previously required Test-Tester now have to depend on virtual/perl-Test-Simple, but only if they need a newer version than 0.114.0 But it also means that if somebody upgrades to Perl 5.24, they should be upgrading to using Test-Simple instead of using Test-Tester. Subsequently, dev-perl/Test-Tester-1.1.14 exists for the first case, which pulls virtual/perl-Test-Simple-1.1.14 and doesn't install any files of its own. ( And it must not, because any perl-core/Test-Simple-1.1.14 would conflict with files ) And subsequently for the second case, virtual/perl-Test-Simple blocks against older versions of Test-Tester than 0.114.0, because installing that virtual without forcing an upgrade of Test-Tester breaks assumptions that hold true on CPAN about what happens here. Under this condition, for some bizarre reason, this construct: > || ( >=virtual/perl-Test-Simple-1.1.10 >=dev-perl/Test-Tester-0.107 ) ( Which is intended for a replacement construct for "Just use Test-Simple" until such time as we can exorcise Test-Tester from tree ), results in portage picking Test-Tester..... Which is weird in multiple ways. The preferred behaviour here is of course that if you *can* pick Test-Simple instead of Test-Tester, you *should*, but portage seems to do it backwards, pulling Test-Tester into the graph, and then blocking against it. Which seems surely like a resolver error if it can be so trivially fixed by masking the second package.
(In reply to Kent Fredric (IRC: kent\n) from comment #5) > Which seems surely like a resolver error if it can be so trivially fixed by > masking the second package. Unfortunately, there are also some cases where an || ( foo bar ) dependency should prefer bar if bar happens to be installed, and there's just no easy way for the dependency resolver to automatically distinguish between these cases. Therefore, for confusing cases like this, it's better to choose some other dependency construct than || ( foo bar ). For example, a combination of USE conditional dependencies and use.stable.mask might be appropriate in some situations. For the current issue at hand, I would suggest to introduce a virtual/perl-Test-Tester package, where there's a stable version that pulls in exclusively >=dev-perl/Test-Tester-0.107, and and unstable version that pulls in exclusively >=virtual/perl-Test-Simple-1.1.10. Then the unstable version of virtual/perl-Test-Tester should be stabilized in lockstep with >=dev-lang/perl-5.24.0. Problem solved.
Created attachment 436494 [details] Dependency Plan.png Hopefully I understand this plan correctly. BR Green => Stable, BR Yellow => ~arch, Gold => Decision, Redarrow => "X -> blocked by -> Y"
Yeah, that looks good to me.
Created attachment 436558 [details] Graph reoriented with things routed to exclude dev-perl/* fluff Actually, given there's a stable pathway that satisfies both Test::Tester and Test::use::ok now via perl-core/Test-Simple-1.1.14 .... it seems it would be easier to just kill Test-Tester and Test-use-ok now entirely and just push everyone to using Test-Simple to provide these things. After all, creating ~6 virtuals, going through the tree, mapping everything to use those virtuals, waiting for the fallout, then waiting another 12 months before finally weaning ourselves off Test-Tester/Test-use-ok before mapping everything to Test-Simple and nuking that stuff.... Why don't we just step to the end game and do what we're going to end up doing anyway, and just pull Test-Simple exclusively? The only downside I see is that means everything using Test::Tester or Test::use::ok will pull perl-core/Test-Simple on stable, which is a *potential* risk for migration to 5.24, ... As opposed to this actual horrible thing we're doing now. ( Red squares = Delete/Hardmask )
(In reply to Kent Fredric (IRC: kent\n) from comment #9) > The only downside I see is that means everything using Test::Tester or > Test::use::ok will pull perl-core/Test-Simple on stable, which is a > *potential* risk for migration to 5.24, ... How about if we introduce an unstable virtual/perl-Test-Simple-5.22 that drops perl-core/Test-Simple from the choices, like: RDEPEND="|| ( =dev-lang/perl-5.24* =dev-lang/perl-5.22* )" Which would be analogous to the virtual/perl-Test-Tester idea suggested earlier.
(In reply to Zac Medico from comment #10) > > How about if we introduce an unstable virtual/perl-Test-Simple-5.22 that > drops perl-core/Test-Simple from the choices, like: > > RDEPEND="|| ( =dev-lang/perl-5.24* =dev-lang/perl-5.22* )" > > Which would be analogous to the virtual/perl-Test-Tester idea suggested > earlier. PV=5.22 is a non-player here because it too much restricts the range of dependency restrctions that can be used by things that need min-max versions of Test-Simple. However, as a compromise, I can see: virtual/perl-Test-Simple-1.1.14 -> arch, maps only to perl-core/ virtual/perl-Test-Simple-1.1.14_p5.22 -> ~arch, Maps only to 5.24 and 5.22 Because that will still be logically smaller than 1.1.15 and logically larger than 1.1.13
Created attachment 440442 [details] Plan 3 As per my understanding. Going to glue together a branch doing just this for review.
After hacking on it I realised my model was wrong. They should *both* be arch. virtual/perl-Test-Simple-1.1.14 -> ~arch, maps only to perl-core/ virtual/perl-Test-Simple-1.1.14_p522 -> ~arch, Maps only to 5.24 and 5.22 However, only the *second* of these should be stabilized, with Perl 5.22 And the first should be stabilized whenever something needing Test-Simple-1.1.14 needs it. The idea being as soon as anyone sees both of them in ~arch, the second will "win", and cause a cascade of removing perl-core/* Very rough initial implementation for review: https://github.com/gentoo-perl/gentoo/compare/master...gentoo-perl:bgo-584238 The only *real* pain I see here is people foolishly selectively unmasking for ~arch. If they ACCEPT_KEYWORDS="~arch" for virtual/perl-Test-Simple, *both* will come into scope, and it will force an upgrade to perl-5.22 and force removing perl-core/* But that seems reasonable. And then when _p522 is fully stable, its non_p* version can be nuked.
Current matrix of things that are blocking "ubiquitously use Test-Simple to provide Test-Tester and Test-use-ok" That is, the following stable flags need to be provided on >=virtual/perl-Test-Simple-1.1.10 before we can actionably remove Test-use-ok and Test-Tester ( These are all possibly making our dependency problems more complicated ) Key: "foo - bar, baz" => "Test-Simple needs foo because bar and baz want foo" arm64 - Test-NoWarnings, Test-Output hppa - Test-CleanNamespaces, Test-NoWarnings, Test-Output ia64 - Test-Deep, Test-NoWarnings, Test-Output m68k - Test-Output ppc - Test-Deep, Test-NoWarnings, Test-Output ppc64 - Test-Deep, Test-NoWarnings, Test-Output s390 - Test-NoWarnings, Test-Output sh - Test-NoWarnings, Test-Output sparc - Test-Deep, Test-NoWarnings, Test-Output
Other changes yet to land, but this is the one that needs the most testing: commit d4b59936cf7ef7183dd40a89969388c06fff6830 Author: Kent Fredric <kentnl@gentoo.org> Date: Wed Jul 13 01:15:52 2016 +1200 virtual/perl-Test-Simple: remove dev-lang/perl fallback for perl-5.20 This is part of the fix of bug #584238. This bump removes dev-lang/perl-5.22 and dev-lang/perl-5.24 from the alternatives list for people using this virtual. If you're upgrading to perl-5.22, you should upgrade to virtual/perl-Test-Simple-1.1.14_p522 at the same time, which will excise perl-core/* instead of pulling perl-core/* Keywords DROPPED: ~arm ~alpha , because they were previously only stabilized because dev-lang/perl was stable for those keywords. However, perl-core/Test-Simple-1.1.14 is not stable, so this can't be an alternative and stay stable. But virtual/perl-Test-Simple-1.1.14_p522 is already "stable" for ~arm and ~alpha, so people on stable will move directly there. virtual/perl-Test-Simple/perl-Test-Simple-1.1.14-r1.ebuild | 22 ---------------------- virtual/perl-Test-Simple/perl-Test-Simple-1.1.14-r2.ebuild | 20 ++++++++++++++++++++ 2 files changed, 20 insertions(+), 22 deletions(-) commit 0467ae3501424ead51d86db396c1bd045ab1c3bf Author: Kent Fredric <kentnl@gentoo.org> Date: Tue Jul 12 22:16:09 2016 +1200 virtual/perl-Test-Simple: Add virtual 1.1.14_p522 for 5.22 stabilization This is an attempted solution at solving bug #584238 This virtual maps only to dev-lang/perl and forces removal of: - dev-perl/Test-Tester - dev-perl/Test-use-ok - perl-core/Test-Simple Keywords all dropped to ~arch except for keywords that were already marked stable for dev-lang/perl-5.22.2 in bug #567482 ( arm , alpha ) _THIS_ package should become the new stabilization target for 5.22, and the non _p522 version should *not* be further stabilized. virtual/perl-Test-Simple/perl-Test-Simple-1.1.14_p522.ebuild | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+)
Thx guys. Just wanted to inform you that with virtual/perl-Test-Simple-1.1.14-r2 the block is gone. So from my point of view this report can be considered resolved.
(In reply to gentoo_usr from comment #16) > Thx guys. > Just wanted to inform you that with virtual/perl-Test-Simple-1.1.14-r2 the > block is gone. So from my point of view this report can be considered > resolved. For future reference, *please* report emerge --info as instructed. It now occurs to me that you had some keywording context that *may* have been relevant ( its still not clear based on what you've said ). Lack of information leads to confusion and wrong conclusions.
@jer, re: https://bugs.gentoo.org/show_bug.cgi?id=567482#c10 ( Talking here to avoid spamming stabilizers ) Yes, that is an oversight of sorts, but it was more specifically an oversight that "~" would not match against _p<XXX>. Test-Tester-1.1.14 should be /removed/, but ascertaining what dependency conditions that cause portage to elect *not* to remove it is hard, as is how to fix it. But I don't have enough information ( or concentration capacity ) based on what you've said there to actually know what to do to make the right path activate. Replacing the "~" with a ">=" /could/ work, but I'm still not sure how that works. Because the stabilization target should be saying "remove Test-Tester entirely", so the "~" or ">=" flag should be immaterial. ( Additionally, if you can make a tar of /var/db/pkg, /var/lib/portage, /etc/portage and any "local-only" overlays that can be used to replicate this error and put it in a public place somewhere for testing one day when we work out how, that might be useful )
(In reply to Kent Fredric (IRC: kent\n) from comment #18) > Yes, that is an oversight of sorts, but it was more specifically an > oversight that "~" would not match against _p<XXX>. =perl-core/Test-Simple-1.1.14* will give precisely the effect that you are looking for.
Also, I assumed ( because you were on the stabilzation ticket ) you were testing to stabilize. But if you weren't, that changes a *lot*, so I need to know what your keywords are so I can work out what path is causing the problem. ( There is way too much dependency complexity here to reason about easily )
Also, this is nonsense: > perl-core/Test-Simple-1.1.14-r1::gentoo (masked by: package.mask) Where is that mask comming from? The only masked perl-core/ should be the 1.300x version. perl-core/Test-Simple-1.1.14 is presently the only way to satisfy >=virtual/perl-Test-Simple-1.1.14 on stable until dev-lang/perl-5.22.2* and virtual/perl-Test-Simple-1.1.14_p522 are stable. ( Currently, only stable on arm and alpha ) So it looks like you've got local configuration blocking it working.
(In reply to Kent Fredric (IRC: kent\n) from comment #21) > Also, this is nonsense: > > > perl-core/Test-Simple-1.1.14-r1::gentoo (masked by: package.mask) Note: You're responding to bug #567482 comment #10 here. > Where is that mask comming from? I am trying to set up things locally to prevent emerge from trying a downgrade to dev-lang/perl-5.20. Set up FEATURES=test USE=test and you will find that things do not upgrade smoothly. On the other bug you noted that this cannot block stabilisation, but it does.
(In reply to Jeroen Roovers from comment #22) > (In reply to Kent Fredric (IRC: kent\n) from comment #21) > > Also, this is nonsense: > > > > > perl-core/Test-Simple-1.1.14-r1::gentoo (masked by: package.mask) > On the other bug you noted that this cannot block stabilisation, but it does. This is really a horrible problem, because this problem is portage picking the wrong path, where it chooses to install Test-Tester instead of removing it. The problem is the package that triggers "This problem" for you, Test-NoWarnings, is itself, stable. In order to "Fix" this problem properly, Test-NoWarnings must be modified to exclusively pull >=virtual/perl-Test-Simple-1.1.10, and never pull dev-perl/Test-Tester ( and that would permit Test-Tester to be uninstalled ) However, *that* commit *cant* be done, because no versions of >=perl-Test-Simple-1.10 are yet "stable". So stabilization is essentially blocking itself.
As a sidenote, I recently tweaked the dep spec to resolve the superficial ~ issue: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=9716505b5f5519e31b1f4629aafda292530ca830 But that won't solve portage being convinced it aught to conflict with itself. The "Correct" option if you are a stabilizer trying to stabilize 5.22 is to locally pmask and uninstall dev-perl/Test-Tester entirely and then report any bugs that produces. Because that is the "right" solution for 5.22, and as soon as we can do that, we will. ( But 5.22 must happen before we can )
@jer, ok, stabilization should no longer be actually blocked by this problem. All the remaining offenders pulling dev-perl/Test-Tester have been fixed by dependency nuking and disabling tests when not satisfied. Essentially, instead of FEATURES="test" making portage take a hard road to Test-Tester, the residual cases that needed stabilization now simply assume that Test::Tester will be provided "Somehow", and the tests ( which are the only places Test::Tester is used ) are optimistically disabled if Test::Tester is not installed. In short, this means that: - Once you have perl-5.22 installed, Test::Tester *will* be present and tests *will* run. - If you already have Test::Tester but are on perl < 5.22, nothing interesting will happen, tests will still run. - If you are on perl < 5.22 and have installed perl-core/Test-Simple >= 1.1.10 via virtual/perl-Test-Simple, nothing interesting will happen, and tests will still run. - If you are on perl < 5.22 and have *not* installed perl-core/Test-Simple and have *not* installed Test-Tester, then the test suites will be skipped where relevant, and simply einfo that you need to solve this problem yourself if you want tests to work. dev-perl/Test-Tester is now package.masked' and portage should entirely cease prioritizing for it. Please report any additional problems. ( Though hopefully, in a new bug )
commit 1b4f3155b778b1794055ee10102bc6ff7765bef4 Author: Michał Górny <mgorny@gentoo.org> AuthorDate: Mon Sep 5 15:39:44 2016 Commit: Michał Górny <mgorny@gentoo.org> CommitDate: Mon Sep 5 15:53:55 2016 dev-perl/Test-Tester: Remove last-rited pkg, #584238
