Richard Purdie [Thu, 19 Jan 2017 16:56:30 +0000 (16:56 +0000)]
gtk-icon-cache: Add PACKAGE_WRITE_DEPS for postinst
gdk-pixbuf-native and gtk-icon-utils-native are needed by the postinstall
scripts so mark the dependency. The utils may be needed at icon build
time too so DEPENDS is unchanged.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Thu, 19 Jan 2017 16:53:53 +0000 (16:53 +0000)]
sstate: Drop the depchain isPostDep() checks
The dependencies of do_package_write_* tasks are either going to be packaging
tools needed to build the packages, or, native tools needed at postinst
time. Now we've formalised this dependency pattern, drop the hardcoded
list and work based on the rule. The package creation tools are usually
the same tools needed at rootfs/postinst time anyway so the difference is
moot.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Thu, 19 Jan 2017 16:49:00 +0000 (16:49 +0000)]
classes/package*: Add support for PACKAGE_WRITE_DEPS
Add a new variable to allow markup of postinstall (and preinst)
script dependnecies on native/cross tools.
If your postinstall can execute at rootfs creation time rather than on
target but depends on a native tool in order to execute, you need to
list that tool in PACKAGE_WRITE_DEPENDS.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The OVS fails to function and the kernel modules cannot be found by
any of the kernel tools such as depmod because they are installed
into the wrong directory in multilib 64bit/32bit bulids.
Signed-off-by: Jason Wessel <jason.wessel@windriver.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Paul Eggleton [Thu, 19 Jan 2017 18:29:17 +0000 (07:29 +1300)]
Revert "classes/sstate: add a mode to error if sstate package unavailable"
It turns out that this check cannot work. We don't have the information
to know whether an sstate package is really needed at this point in the
execution, so we check the availability for things that we won't
actually end up needing later on. Thus we can't fail if some of these
aren't found or we'll get needless failures.
This check was intended to give earlier more accurate errors when sstate
artifacts failed to download, but that's not practical so we'll rely
solely on the task execution check that was added within the runqueue.
This reverts most of commit 9e711b54487c3141d7264b8cf0d74f9465020190 (we
still need to allow BB_SETSCENE_ENFORCE through from the external
environment since the eSDK relies upon that.)
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Paul Eggleton [Thu, 19 Jan 2017 18:29:16 +0000 (07:29 +1300)]
classes/populate_sdk_ext: force a known value for TMPDIR
If TMPDIR is configured to be somewhere outside of TOPDIR (a not
uncommon configuration where you have multiple disks and space on /home
is at a premium) then our attempt to find out the location of paths
under TMPDIR by using a relative path led to horribly broken paths
ending up in the eSDK. To save pain, just force a known value for TMPDIR
(i.e. ${TOPDIR}/tmp) and then we can assume that everywhere else.
Fixes [YOCTO #10797].
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Thu, 12 Jan 2017 13:30:02 +0000 (13:30 +0000)]
staging: Add dependency on virtual/binutils for strip binary
The staging code strips binaries and we need virtual/binutils for that.
Add a specific dependency since the one from do_configure and others
may not be enough to ensure the binaries are in our own sysroot.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Thu, 12 Jan 2017 13:30:53 +0000 (13:30 +0000)]
oe-selftest: Error if the user has buildhistory enabled
buildhistory interfers with various tests so error if the user has it enabled,
hence avoiding time spent running the tests only for them to fail.
For example, if there is an sstate cache and something comes from that cache,
you can get a "version went backwards" warning from buildhistory which would
then change the outcome of a test. Its safer/easier to disable it.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Paul Eggleton [Mon, 16 Jan 2017 23:14:36 +0000 (12:14 +1300)]
devtool: fix source extraction to work with recipe specific sysroots
When extracting source for a recipe within devtool (for extract, modify
or upgrade) We need to redirect WORKDIR, STAMPS_DIR etc. under a
temporary directory so that:
(a) we pick up all files that get unpacked to the WORKDIR, and
(b) we don't disturb the existing build
However, with recipe-specific sysroots the sysroots for the recipe will
be prepared under WORKDIR, and if we used the system temporary directory
i.e. usually /tmp) as used by mkdtemp by default, then our attempts to
hardlink files into the recipe-specific sysroots will fail on systems
where /tmp is a different filesystem, and we'd have to fall back to
copying the files which is a waste of time. Put the temp directory under
the WORKDIR to prevent that from being a problem.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
[RP: Add needed mkdirhier call] Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Tue, 17 Jan 2017 17:42:40 +0000 (17:42 +0000)]
sstate: Fix EXTRA_STAGING_FIXMES handling
The code wasn't working properly, tweak so that it works as expected and
the grep expression includes the right patterns. Not sure this code has ever
worked prior to this.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Patrick Ohly [Fri, 13 Jan 2017 14:52:33 +0000 (15:52 +0100)]
rm_work.bbclass: clean up sooner
Having do_rm_work depend on do_build had one major disadvantage:
do_build depends on the do_build of other recipes, to ensure that
runtime dependencies also get built. The effect is that when work on a
recipe is complete and it could get cleaned up, do_rm_work still
doesn't run because it waits for those other recipes, thus leading to
more temporary disk space usage than really needed.
The right solution is to inject do_rm_work before do_build and after
all tasks of the recipe. Achieving that depends on the new bitbake
bb.event.RecipeTaskPreProcess and bb.build.preceedtask().
It can't just run in an anonymous function, because other anonymous
functions that run later may add more tasks. There's still such a
potential conflict when some future RecipeTaskPreProcess event handler
also wants to change task dependencies, but that's not a problem
now. Should it ever occur, the two handlers will have to know about
each other and cooperate to resolve the conflict.
Benchmarking (see "rm_work + pybootchart enhancements" on the OE-core
mailing list) showed that builds with the modified rm_work.bbclass
were both faster (albeit not by much) and required considerably less
disk space (14230MiB instead of 18740MiB for core-image-sato).
Interestingly enough, builds with rm_work.bbclass were also faster
than those without.
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Patrick Ohly [Fri, 13 Jan 2017 14:52:32 +0000 (15:52 +0100)]
rm_work_and_downloads.bbclass: more aggressively minimize disk usage
rm_work.bbclass never deletes downloaded files, even if they are not
going to be needed again during the
build. rm_work_and_downloads.bbclass is more aggressive in minimizing
the used disk space during a build, but has other disadvantages:
- sources required by different recipes need to be fetched once per
recipe, not once per build
- incremental builds do not work reliably because sources get
removed without ensuring that sources gets fetched again
That makes rm_work_and_downloads.bbclass useful for one-time builds in
a constrained environment (like a CI system), but not for general use.
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
By default, do_rm_work either skips recipes entirely (when listed in
RM_WORK_EXCLUDE) or removes everything except for temp.
In meta-swupd, virtual image recipes collaborate on producing update
data for the base recipe. Tasks running in the base recipe need some
information from the virtual images.
Those files could be passed via a new shared work directory, but that
scatters data in even more places. It's simpler to use the normal
WORKDIR and teach rm_work.bbclass to not remove the special output
with the new RM_WORK_EXCLUDE_ITEMS.
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Ross Burton [Tue, 17 Jan 2017 23:55:46 +0000 (23:55 +0000)]
oeqa/selftest/devtool: rewrite modify testcase
The modify testcase had to be updated as it started failing when mdadm was
upgraded due to hardcoding version numbers in the test. I then noticed how
inefficient the test was and mostly rewrote it.
Start by changing the minor modification to change "Linux Software RAID" (the
subtitle of the man page) to "antique pin sardine" (a nonsense phrase that is
unlikely to appear upstream), and neaten the logic.
Start by not removing sstate at the beginning of the test. To ensure builds
happen we can use -f and -C, and iterating the sstate cache is time consuming.
Don't bitbake mdadm repeatedly until it stabilizes, we can start with bitbake -C
unpack to ensure that a full build is done from scratch.
os.path.join has the interesting quirk that join(/foo, /bar) results in /bar, so
use oe.path.join instead of working around that manually.
Don't repeatedly call get_bb_var(), each call results in a call to bitbake.
These changes reduce the runtime of the test from over 600 seconds to around 160
seconds on my machine.
Signed-off-by: Ross Burton <ross.burton@intel.com>
Mariano Lopez [Fri, 13 Jan 2017 14:33:55 +0000 (14:33 +0000)]
oeqa/utils/qemurunner.py: Be sure to stop qemu-system
When runqemu fails, qemu-system process would keep running
and won't be killed, setpgrp() was used when runqemu was
a shell script but it seems it doesn't work always with python.
This would kill qemu-system explicity and to avoid leaving
it behind.
Signed-off-by: Mariano Lopez <mariano.lopez@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Mariano Lopez [Fri, 13 Jan 2017 14:33:54 +0000 (14:33 +0000)]
oeqa/utils/qemurunner.py: Add missing sys module
This adds the missing sys module used by the child process
to exit. It seems the exception was cached in testimage and
selftest. It seems nobody noticed this because the module
is only used for sys.exit().
Signed-off-by: Mariano Lopez <mariano.lopez@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Markus Lehtonen [Fri, 13 Jan 2017 13:12:45 +0000 (15:12 +0200)]
oeqa.utils.metadata: allow storing any bitbake config variables
Make it possible to store any bitbake config variables in the metadata.
Config values will be stored under a new config element in the xml report:
<config>
<variable name="MACHINE">qemux86</variable>
</config>
The value of MACHINE is moved there instead of having a dedicated
<machine> element.
[YOCTO #10590]
Signed-off-by: Markus Lehtonen <markus.lehtonen@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Markus Lehtonen [Fri, 13 Jan 2017 13:12:43 +0000 (15:12 +0200)]
oeqa.utils.metadata: have layer name as an attribute in xml
Have the layer name as an attribute instead of of the name of the
element itself. That is, have <layer name="layer_name"/> instead of
<layer_name/>. A bit better XML design.
[YOCTO #10590]
Signed-off-by: Markus Lehtonen <markus.lehtonen@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Markus Lehtonen [Fri, 13 Jan 2017 13:12:41 +0000 (15:12 +0200)]
oeqa.utils.metadata: rename 'revision' to 'commit'
Revision is a bit vague and could point to a tag, for example. Git
commit objects are unambiguous and persistent so be explicit that the
element should contain git commit hash.
[YOCTO #10590]
Signed-off-by: Markus Lehtonen <markus.lehtonen@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Markus Lehtonen [Fri, 13 Jan 2017 13:12:40 +0000 (15:12 +0200)]
oeqa.utils.metadata: fix retrieval of git branch and revision
Always return a valid branch name, or, '(nobranch)' if the current HEAD
is detached. Also, always return the hash of the commit object that HEAD
is pointing to. Previous code returned an incorrect branch name (or
crashed) e.g. in the case of detached HEAD.
[YOCTO #10590]
Signed-off-by: Markus Lehtonen <markus.lehtonen@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Markus Lehtonen [Fri, 13 Jan 2017 13:12:37 +0000 (15:12 +0200)]
oeqa.utils.metadata: re-organise host distro information
Put all host distro data under one <host_distro> element. In addition
take the data directly from /etc/os-release instead of the "lsb API".
The /etc/os-release file is virtually ubiquitous, now, and using its
field names and values provides a more standardized and extensible
format.
[YOCTO #10590]
Signed-off-by: Markus Lehtonen <markus.lehtonen@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
André Draszik [Tue, 17 Jan 2017 15:54:39 +0000 (15:54 +0000)]
time: BBCLASSEXTEND to native and nativesdk
Some recipe might use time, e.g.
http://git.yoctoproject.org/cgit/cgit.cgi/meta-swupd/tree/classes/swupd-image.bbclass?id=531a8e636ca891c34f2596eb7f6365a11d4b29a7#n481
Time is a bash built-in, or a separate utility (typically
in /usr/bin/), but not everybody uses bash, or has
GNU time installed by default.
Signed-off-by: André Draszik <adraszik@tycoint.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
André Draszik [Tue, 17 Jan 2017 15:28:28 +0000 (15:28 +0000)]
python3-pygobject: enable cairo only when a display is available
It doesn't seem to make sense to unconditionally enable cairo
support, as this pulls in lots of other dependencies, where
none of these can actually be used without some sort of
graphical interface.
Not having a generic distro feature to detect this, we just
use any of DirectFB / Wayland / X11
Signed-off-by: André Draszik <adraszik@tycoint.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Nathan Lynch [Mon, 16 Jan 2017 18:51:45 +0000 (12:51 -0600)]
lttng-tools: upgrade to 2.9.3
Notable recipe changes:
* switch to using release tarball
* brokensep no longer required
* drop upstreamed error.h patch
* change do_install_ptest to selectively copy build artifacts to
install tree.
* use backported patch to address file-rdeps warning
Signed-off-by: Nathan Lynch <nathan_lynch@mentor.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Nathan Lynch [Mon, 16 Jan 2017 18:51:43 +0000 (12:51 -0600)]
lttng-ust: upgrade to 2.9.0
buildhistory-diff says:
/usr/lib/liblttng-ust-python-agent.so.0 was removed
/usr/lib/liblttng-ust-python-agent.so.0.0.0 was removed
This is because prior versions of lttng-ust unconditionally built
the Python agent libraries; this was fixed upstream in 21ddb8e ("Only
build python lib when agent is enabled").
Switch to using release tarballs.
Signed-off-by: Nathan Lynch <nathan_lynch@mentor.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Nathan Lynch [Mon, 16 Jan 2017 18:51:42 +0000 (12:51 -0600)]
lttng-ust: don't exclude manual pages from build
Adding "manpages" to PACKAGECONFIG results in a lttng-ust-doc package
with no manpages. Fix the patch we carry to only exclude the examples
directory, which seems to have been its original intent anyway.
Signed-off-by: Nathan Lynch <nathan_lynch@mentor.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Ross Burton [Mon, 16 Jan 2017 13:11:10 +0000 (13:11 +0000)]
perf: use PACKAGECONFIG instead of hand-rolled logic
PERF_FEATURES_ENABLE and perf_feature_enabled() was basically a poor man's clone
of PACKAGECONFIG, without the automatic handling of dependencies and configure
options.
As part of the port to PACKAGECONFIG the options have been changed to remove the
perf- prefix, but are otherwise unchanged.
Also remove BUILDPERF_libc_uclibc assignment as nothing in the metadata uses a
BUILDPERF variable.
Signed-off-by: Ross Burton <ross.burton@intel.com>
Maxin B. John [Fri, 13 Jan 2017 12:57:09 +0000 (14:57 +0200)]
mdadm: upgrade to version 4.0
3.4 -> 4.0
Removed the following upstreamed or backported patches:
a) 0001-Fix-some-type-comparison-problems.patch
b) 0001-Fix-typo-in-comparision.patch
c) 0001-mdadm.h-bswap-is-already-defined-in-uclibc.patch
d) 0001-raid6check-Fix-if-else-indentation.patch
e) 0001-util.c-include-poll.h-instead-of-sys-poll.h.patch
f) mdadm-3.2.2_fix_for_x32.patch
Signed-off-by: Maxin B. John <maxin.john@intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Patrick Ohly [Fri, 13 Jan 2017 14:52:31 +0000 (15:52 +0100)]
gcc-source.inc: cleanly disable do_rm_work
Using "deltask" assumes that do_rm_work has been added already, which
won't be the case anymore in the upcoming improved rm_work.bbclass,
because then an anonymous python method will add do_rm_work.
Setting RM_WORK_EXCLUDE works with the current and upcoming
rm_work.bbclass and is the API that is meant to be used for excluding
recipes from cleaning, so use that.
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Robert Yang [Wed, 23 Nov 2016 08:00:18 +0000 (00:00 -0800)]
runqemu-export-rootfs: fix inconsistent var names
Fixed:
$ runqemu nfs qemux86-64
[snip]
On your target please remember to add the following options for NFS
nfsroot=IP_ADDRESS:/path/to/nfsroot,nfsvers=3,port=,mountprog=,nfsprog=,udp,mountport=
[snip]
Note that the values are null, this is because their var names are
inconsistent.
[YOCTO #10519]
Signed-off-by: Robert Yang <liezhi.yang@windriver.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Robert Yang [Tue, 22 Nov 2016 09:26:32 +0000 (01:26 -0800)]
scripts/runqemu: fix checking for <file>.cpio.gz
When "runqemu /path/to/<file>.cpio.gz", it used the last suffix "gz" as
the fstype which was wrong. Check filename against self.fstypes firstly
can fix the problem.
Signed-off-by: Robert Yang <liezhi.yang@windriver.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
David Vincent [Mon, 16 Jan 2017 14:30:16 +0000 (15:30 +0100)]
extrausers: Use static ids when available
When creating users at the image level using extrausers class, the
current behavior is to ignore the status of USERADDEXTENSION. This could
lead to undefined behavior when static ids are expected but the system
falls back to dynamic ones.
Signed-off-by: David Vincent <freesilicon@gmail.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Maciej Borzecki [Mon, 16 Jan 2017 10:41:51 +0000 (11:41 +0100)]
wic: partitionedfs: account for non-table partitions when checking if logical parititon is needed
Commit 8c1c43b7901a9fcd8b279eb4250b08157ad345b7 `wic: Create a logical partition
only when it is really mandatory` did not account for partitions that are not
present in partition table.
Signed-off-by: Maciej Borzecki <maciej.borzecki@rndity.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Ola Redell [Fri, 13 Jan 2017 17:22:23 +0000 (18:22 +0100)]
kernel-module-split: Append KERNEL_VERSION string to kernel module name
The KERNEL_VERSION string is added to kernel module package names in order
to make the kernel modules for different kernel versions distinct packages
instead of different versions of the same package. With this change, when
a new kernel is installed together with its kernel modules (e.g. by upgrade
of the packages kernel and kernel-modules) using some package manager such
as apt-get or rpm, the kernel modules for the older kernel will not be
removed. This enables a fall back to the older kernel if the new one fails.
Also, for backwards compatibility and to enable kernel version agnostic
dependencies to kernel modules, create a virtual package with the old
(shorter) kernel module package name using RPROVIDES.
Signed-off-by: Ola Redell <ola.redell@retotech.se> Signed-off-by: Ross Burton <ross.burton@intel.com>
The name of the output image for a fitImage that contains a ramdisk
should match the same as for the fitImage that does not contain a
ramdisk. As such it should not be assumed that KERNEL_IMAGETYPE is
"fitImage". This change explicitly sets the name of the output
ramdisk/initramfs to start with fitImage as does the non-ramdisk output.
Signed-off-by: Nathan Rossi <nathan@nathanrossi.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Nathan Rossi [Fri, 21 Oct 2016 12:07:27 +0000 (22:07 +1000)]
kernel-fitimage.bbclass: Allow unset load/entry addresses for ramdisks
Allow the load and entry addresses to remain unset if the UBOOT_RD_*
variables are also unset for ramdisk entries in the image tree. This
allows for U-Boot to decide dynamically where to load the ramdisk.
Signed-off-by: Nathan Rossi <nathan@nathanrossi.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Juro Bystricky [Thu, 12 Jan 2017 22:07:58 +0000 (14:07 -0800)]
sanity.bbclass: Improved error message
When a non-existing MACHINE is specified, sanity check issues
the following message:
Please set a valid MACHINE in your local.conf or environment
However, MACHINE can also be set in multiconfig .conf file(s).
Hence we may have several different MACHINE settings within one
(multiconfig) build, so the present error message is fairly
ambiguous.
This patch remedies this by explicitly naming the offending MACHINE and
by amending the list of places where this erroneous MACHINE definition
could have originated.
MACHINE=xyz is invalid. Please set a valid MACHINE in your local.conf, environment or other configuration file.
[YOCTO#10810]
Signed-off-by: Juro Bystricky <juro.bystricky@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Presently there is no check to verify the existence of configuration
files as listed in BBMULTICONFIG.
For example, BBMULTICONFIG = "foobar" in local.conf does not trigger
an error or even a warning when there is no conf/multiconfig/foobar.conf.
The missing file is silently ignored.
This patch changes the inclusion of all multiconfig configuration files as
a non-optional requirement. If the file is missing, we get an error such as:
ERROR: ParseError at /data/master-multi/poky/meta/conf/bitbake.conf:704: Could not include required file conf/multiconfig/foobar.conf
Although the "default" configuration is not listed in BBMULTICONFIG,
this change also requires the file multiconfig/default.conf to exist.
The "default" (non-multiconfig) configuration is normally configured via local.conf,
so although this file is required, it can/should be empty. This patch creates
an empty file default.conf in meta/conf/multiconfig.
[YOCTO#10917]
Signed-off-by: Juro Bystricky <juro.bystricky@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This patch makes sure the directory which will contain
the systemd configuration (loader.conf) is created before
the configuration file is written, fixing errors when it
tried to write it to a non-existent directory
Signed-off-by: Alejandro Hernandez <alejandro.hernandez@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Tom Zanussi [Wed, 7 Sep 2016 14:45:15 +0000 (09:45 -0500)]
scripts: python3 fixes and new tool ksum
'ksum.py' generates a combined summary of vmlinux and module sizes for
a built kernel, as a quick tool for comparing the overall effects of
systemic tinification changes. Execute from the base directory of the
kernel build you want to summarize. Setting the 'verbose' flag will
display the sizes for each file included in the summary.
Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com> Signed-off-by: Alejandro Hernandez <alejandro.hernandez@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
André Draszik [Thu, 12 Jan 2017 10:34:39 +0000 (10:34 +0000)]
lib/oe/rootfs: reliably handle alternative symlinks
When removing unneeded packages from a (read-only) rootfs
during rootfs creation, alternative symlinks from those
packages may or may not be removed.
The reason is as follows:
update-alternatives(-native) is used during package
installation as part of the image creation. It uses
a database which contains entries for all the
alternative symlinks possible, and the -native version
uses the target's database by means of $OPKG_OFFLINE_ROOT,
i.e. the rootfs we're in the process of creating.
Once the rootfs has been created, OE removes certain
packages because we have a read-only rootfs - in
particular ROOTFS_RO_UNNEEDED which includes
VIRTUAL-RUNTIME_update-alternatives, i.e. the
update-alternatives. Recently, a change was made in
OE, where uninstallation of update-alternatives from the
rootfs causes removal of its database, too, to save space
(700KiB (uncompressed) in a busybox system) b24a63d71b517af701dfedbc7f7b541d25af708f
http://git.openembedded.org/openembedded-core/commit/meta/recipes-devtools/opkg-utils/opkg-utils_git.bb?id=b24a63d71b517af701dfedbc7f7b541d25af708f
Following from that, if update-alternatives is removed
from the target file system, update-alternatives-native
has no database anymore, meaning it can't manage any of
the alternative symlinks anymore.
Because the order of packages to uninstall is
non-deterministic, and update-alternatives could well
be removed before any packages that use the mechanism
provided, sometimes the extra symlinks are removed,
sometimes not.
By sorting the list of packages to be removed such that
update-alternatives is removed last, we can ensure that
that tings work reliably. (Certainly opkg seems to
uninstall packages in the order given on the command
line.)
[YOCTO #10916]
Signed-off-by: André Draszik <adraszik@tycoint.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Ross Burton [Thu, 12 Jan 2017 12:33:59 +0000 (12:33 +0000)]
selftest/base: don't fetch DISTRO variable in constructor
Fetching the DISTRO variable in the base constructor means that we have to start
bitbake for every test case instance, which adds minutes to the startup time.
Signed-off-by: Ross Burton <ross.burton@intel.com>
