Summary: | Convert virtual/linux-sources to a new-style virtual | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | Bruno <bonbons> |
Component: | Eclasses | Assignee: | Gentoo Kernel Miscellaneous <kernel-misc> |
Status: | RESOLVED FIXED | ||
Severity: | normal | CC: | anigel, Marc-Antoine, mips, qa, ulm |
Priority: | High | Keywords: | QAcanfix |
Version: | unspecified | ||
Hardware: | All | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Bug Depends on: | 365105 | ||
Bug Blocks: | 350792 | ||
Attachments: |
Proposed linux-sources-2.4.ebuild
Proposed linux-sources-2.6.ebuild Proposed linux-sources-2.6.ebuild |
Description
Bruno
2006-01-09 11:41:37 UTC
Bruno, could you please post an emerge -pvt of that's shown, when you try to install kde w/ lm_sensors ? Because PROVIDE's (thus VIRTUAL's) don't have a version ... (afair) Here is the output of emerge -pvt kdebase: These are the packages that I would merge, in reverse order: Calculating dependencies ...done! [ebuild N ] kde-base/kdebase-3.4.3-r1 +arts +cups -debug -hal +ieee1394 -java -kdeenablefinal -ldap +lm_sensors -logitech-mouse -openexr +opengl -pam +samba +ssl +xinerama 0 kB [ebuild N ] sys-apps/eject-2.0.13-r2 +nls 0 kB [ebuild N ] sys-apps/lm_sensors-2.9.2 -sensord 0 kB [ebuild N ] sys-kernel/gentoo-sources-2.6.14-r5 -build -doc -symlink (-ultra1) 0 kB [ebuild N ] sys-libs/libraw1394-1.1.0 0 kB [ebuild N ] kde-base/kdelibs-3.4.3 +alsa +arts +cups -debug -doc -jpeg2k -kdeenablefinal -kerberos -openexr +spell +ssl +tiff +xinerama -zeroconf 0 kB [ebuild N ] kde-base/arts-3.4.3 +alsa +arts -artswrappersuid -debug -esd -hardened -jack -kdeenablefinal +mp3 -nas +vorbis +xinerama 0 kB [ebuild N ] dev-libs/glib-2.8.4 -debug -doc -hardened 0 kB [ebuild N ] media-libs/audiofile-0.2.6-r1 0 kB [ebuild N ] media-libs/libmad-0.15.1b -debug 0 kB [ebuild N ] media-libs/libvorbis-1.1.0 -aotuv 0 kB [ebuild N ] media-libs/libogg-1.1.2 0 kB [ebuild N ] media-libs/libart_lgpl-2.3.17 -debug 0 kB [ebuild N ] kde-base/kde-env-3-r4 0 kB [ebuild N ] dev-libs/libxslt-1.1.15 -crypt -debug -python -static 0 kB [ebuild N ] dev-libs/libpcre-6.3 -doc 0 kB [ebuild N ] app-admin/fam-2.7.0-r2 0 kB [ebuild N ] net-nds/portmap-5b-r9 (-selinux) +tcpd 0 kB [ebuild N ] x11-libs/qt-3.3.4-r8 +cups -debug -doc -examples -firebird +gif -immqt -immqt-bc +ipv6 +mysql -nas -odbc +opengl -postgres -sqlite +xinerama 0 kB [ebuild N ] x11-libs/libXrandr-1.1.0.2 0 kB [ebuild N ] media-libs/libmng-1.0.8-r1 0 kB [ebuild N ] media-libs/lcms-1.13-r1 +jpeg -python +tiff +zlib 0 kB [ebuild N ] x11-libs/libXinerama-1.0.1 0 kB [ebuild N ] x11-libs/libXcursor-1.1.5.2 0 kB [ebuild N ] x11-libs/libXfixes-3.0.1.2 0 kB [ebuild N ] net-fs/samba-3.0.14a-r2 -acl +cups -doc -examples -kerberos -ldap +libclamav +mysql -oav -pam -postgres -python -quotas +readline (-selinux) -winbind -xml +xml2 0 kB [ebuild N ] dev-libs/libxml2-2.6.22 -debug -doc +ipv6 -python +readline -static 0 kB [ebuild N ] net-print/cups-1.1.23-r4 -cjk -gnutls +nls -pam +samba -slp +ssl 0 kB [ebuild N ] media-libs/tiff-3.7.3 0 kB [ebuild N ] app-text/xpdf-3.01-r3 +X +motif 0 kB [ebuild N ] media-libs/t1lib-5.0.2 +X -doc 0 kB [ebuild N ] app-text/ghostscript-esp-7.07.1-r10 +X -cjk +cups -emacs -gtk 0 kB [ebuild N ] media-fonts/gnu-gs-fonts-std-8.11 0 kB [ebuild N ] media-libs/libpng-1.2.8 -doc 0 kB [ebuild N ] media-libs/jpeg-6b-r5 0 kB [ebuild N ] media-libs/alsa-lib-1.0.10 -doc -jack 0 kB [ebuild N ] media-sound/alsa-headers-1.0.10 0 kB [ebuild N ] app-text/aspell-0.50.5-r4 +gpm 0 kB [ebuild N ] dev-libs/cyrus-sasl-2.1.20 -authdaemond +berkdb -gdbm -java -kerberos -ldap +mysql -pam -postgres +ssl -static 0 kB Total size of downloads: 0 kB Note: when disabling lm_sensors useflag both lm_sensors and gentoo-sources ebuilds are removed from the output (only difference) To get portage happy and not wanting to install gentoo-sources I have to put sys-kernel/gentoo-sources-2.6.x into /etc/portage/profile/package.proviede (putting there virtual/linux-sources-2.6.15 has no effect) Just curious, are the vserver-sources in your world file ? (`grep vserver-sources /var/lib/portage/world`) But I think that error, you're seeing, doesn't evolve from the PROVIDE but from the check the ebuild does for x86 --- x86? ( || ( >=virtual/linux-sources-2.5 sys-apps/lm_sensors-modules ) )") --- You could try to install the lm_sensors-modules instead and after that, emerge should stop to pull in gentoo-sources ... But the thing is, you'll need to emerge the gentoo-sources, since the ebuild checks whether some CONFIG_ options are set or not ... And it normally depends on a running kernel, which in case of a vserver *is* the hosts kernel .. It's right in this case that it depends and requires >=virtual/linux-sources-2.5. The only thing I could do about that, is to change virtual/linux-sources for the vserver-profile to sys-kernel/vserver-sources, but that in fact would pull in sys-kernel/vserver-sources on that emerge, you're trying .. :) Aaaaaaaaaar, you're trying to do that on the host, right ?! You might see in your emerge output the 'default-linux/x86/2005.1' and in that profile, virtual/linux-sources is set to gentoo-sources. So, it has to do something with the VIRTUAL/PROVIDE handling portage does. I think I'm gonna ask Brian (ferringb) about that tomorrow morning. Yes, I'm installing KDE on the Host. I agree with you that the handling of provide/depend of portage probably causes the issue. Looking at the version-numbering of kernel sources they all use vanilla numbering except vserver-sources and openzv-sources. I guess the problem would be identical with openvz. Would certainly be useful to know how exactly portage handles version-checks for packages providing virtuals. (If this is even allowed, I think there were issues related to this with modular Xorg) Issues with tools depending on kernel for inside guests is another question (and there only /etc/portage/profiles/package.provide seems to be a reasonable solution when used with care, unless the vserver profile provides the kernel by default <- potentially dangerous with verions or required presence of configured kernel source tree) Harring sez: >=virtual/linux-sources-2.5 is not valid for old style virtuals (which is what virtual/linux-sources is). Portage interprets it rather wrongly, and is part of the reason intending on moving to empty ebuilds for glep37 (new virtuals/metapkgs) this needs to be fixed in lm_sensors or with a new-style virtual/linux-sources (In reply to comment #7) > this needs to be fixed in lm_sensors or with a new-style virtual/linux-sources Odd, this is how we have depended on linux-2.6.x+ for a long time. Assigning to kernel@ instead since this is not only an lm_sensors problem. The following ebuilds also depend on a versionized virtual/linux-sources. I'll be happy to write a new-style virtual for this if needed. x11-drivers/ati-drivers-8.14.13-r3 x11-drivers/ati-drivers-8.20.8 x11-drivers/ati-drivers-8.21.7-r1 x11-drivers/ati-drivers-8.18.6 x11-drivers/ati-drivers-8.23.7 x11-drivers/ati-drivers-8.19.10 x11-drivers/ati-drivers-8.16.20-r1 x11-drivers/ati-drivers-8.14.13-r5 x11-drivers/ati-drivers-8.18.8 x11-drivers/ati-drivers-8.18.6-r1 x11-drivers/ati-drivers-8.18.8-r2 x11-drivers/ati-drivers-8.14.13-r2 x11-drivers/ati-drivers-8.16.20 x11-drivers/ati-drivers-8.14.13-r4 x11-drivers/ati-drivers-8.21.7 x11-drivers/ati-drivers-8.22.5 x11-drivers/ati-drivers-8.18.8-r1 media-gfx/plustek-parallel-0.42.9 sys-apps/lm_sensors-2.9.2 sys-apps/lm_sensors-2.10.0 net-misc/tulip-devel-1.1.8-r1 dev-lang/ifc-6.0 dev-lang/ifc-7.1.008 dev-lang/ifc-7.0.064-r1 dev-util/systemtap-20060325 sys-fs/dazuko-2.1.1 sys-fs/dazuko-2.1.0 sys-fs/dazuko-2.2.0 media-tv/rivatv-0.8.6 media-tv/rivatv-0.8.6-r1 I'd like to note that I have new-style virtuals already for all this, however im not comitting them because of some outstanding issues. Profiles however have already been updated ready. Please get in touch with me directly if you want an update on this. Sounds like we're in safe hands *** Bug 133015 has been marked as a duplicate of this bug. *** Should this be closed? vserver-sources-2.1.0-r1 is no longer in portage. No, it should not be closed as the issue is not just affecting that single vserver-sources version but any kernel package that does not follow vanilla version numbers. Starting with >=sys-apps/lm_sensors-2.10.1 it does not pull in gentoo-sources anymore because lm_sensors depends on || ( >=virtual/linux-sources-2.5 >=sys-kernel/vserver-sources-2.0 ) The virtual problem as described in comments #6, #7 and #8 has not been solved yet. As listed by Henrik Brix Andersen in comment #8 there are multiple packages triggering the same problem with versionned old-style virtuals. The only sensible change would be changing the summary to express something like the following: multiple packages depend on versionned old-style virtual virtual/linux-sources which fails to match vserver-sources It should interest PMS and Package Manager or QA developers... The correct summary would be: "virtual/linux-sources needs to be converted to a new style virtual". Why hasn't this happened yet? Could this be converted to a new-style virtual please? See tracker bug 350792. virtual/linux-sources packages: sys-kernel/cell-sources sys-kernel/ck-sources sys-kernel/gentoo-sources sys-kernel/git-sources sys-kernel/hardened-sources sys-kernel/mips-sources sys-kernel/mm-sources sys-kernel/openvz-sources sys-kernel/pf-sources sys-kernel/sparc-sources sys-kernel/tuxonice-sources sys-kernel/usermode-sources sys-kernel/vanilla-sources sys-kernel/vserver-sources sys-kernel/xbox-sources sys-kernel/xen-sources sys-kernel/zen-sources Note: mips-sources (redundantly) have this in PROVIDE of the ebuild, in addition to providing it via kernel-2.eclass. Adding mips to CC. Created attachment 271129 [details]
Proposed linux-sources-2.4.ebuild
Created attachment 271131 [details]
Proposed linux-sources-2.6.ebuild
Ebuilds for the new-style virtual attached, please review.
Open issues: - Should the dependency on sys-kernel/vserver-sources in linux-sources-2.6 be versioned? CCing vserver-devs. - Some 150 packages depend on virtual/linux-sources. Out of these, 5 depend on >=virtual/linux-sources-2.6.XX, which is not allowed with an old-style virtual, and wouldn't be satisfied by new-style virtual/linux-sources-2.6. AFAICS, most of them explicitly use kernel_is from linux-info.eclass in addition, or do a CONFIG_CHECK for specific features. So the versioned dependency may be dispensable. CCing maintainers of the packages in question: app-accessibility/espeakup (williamh, accessibility) net-firewall/xtables-addons (pva) net-wireless/broadcom-sta (matsuu) sys-cluster/gfs-kernel (cluster) sys-fs/vhba (vanquirius) (In reply to comment #18) > - Should the dependency on sys-kernel/vserver-sources in linux-sources-2.6 > be versioned? CCing vserver-devs. the Linux-VServer patch for 2.6 kernels is >=2, but the very old (7 years) vserver-sources for kernel 2.4 were not versioned with the VServer patch version, but the kernel version instead, so they would match >=2 too. so i'd say no version dependency is needed, noone should have 2.4 vserver sources installed anymore. (In reply to comment #18) > app-accessibility/espeakup (williamh, accessibility) > net-firewall/xtables-addons (pva) > net-wireless/broadcom-sta (matsuu) > sys-cluster/gfs-kernel (cluster) > sys-fs/vhba (vanquirius) Reverse dependencies fixed. Created attachment 271505 [details]
Proposed linux-sources-2.6.ebuild
Slightly updated ebuild: hardened-sources will be preferred with USE=hardened.
I'll commit this in a few days if there are no objections.
Committed. |