Richard Purdie [Wed, 1 Mar 2017 15:16:44 +0000 (15:16 +0000)]
oeqa/selftest: Drop http sstate sharing
Using httpServer from python for sharing sstate is known to be buggy, it can't
cope with the number/type of requests coming from bitbake and quietly fails
to share files.
This causes intermittent build failures which are hard to debug. We can
use a file:// url for the sstate mirror instead, removing the need for
the http server.
The sdk-update test is simply dropped since the SDK is never published
to this location and hence it would never have any update. Its equiavalent
to pointing at an empty web server. There is a better eSDK update test in
testsdk so rather than improve this one, lets drop it and concentrate on
the one there.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Wed, 1 Mar 2017 14:47:50 +0000 (14:47 +0000)]
populate_sdk_ext: Allow generation of meta-extsdk-toolchain even for minimal SDKs
If you build a minimal eSDK currently, you don't build meta-extesdk-toolchain
even if you will have built most of its dependencies. This means when you try
and install a toolchain into the eSDK, it fails, breaking our automated testing
of the eSDK.
Therefore add the dependency unconditionally even when a minimal eSDK is being
built and allow the automated testing to work.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Wed, 1 Mar 2017 14:45:46 +0000 (14:45 +0000)]
oeqa/sdkext/context: Work around broken dependency checks to get sdk tests running
This is admitted a bit of a hack but it does allow a number of significant sdk
tests to run successfully and hence improves testing of eSDK which is good.
I'm therefore proposing we do this until we come up with a better solution
since the current lack of testing is worrying and would have caught other issues
had it been present.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Wed, 1 Mar 2017 10:59:13 +0000 (10:59 +0000)]
oeqa/sdkext: Ensure we run a deterministic set of tests
The directory list of sdk tests to run can vary so this code effectively selects
a random set of SDK tests to run in the eSDK. We want to attemp all the SDK tests
so remove the element selection.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Wed, 1 Mar 2017 01:13:34 +0000 (01:13 +0000)]
devtool/sdk: Run build-sysroots after installing new things
After running sdk-install we need to ensure that the standalone sysroots are
updated as done when the eSDK is originally built. Add such a call so this
happens automatically and the envrionment scripts in the SDK work correctly
after updates.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Wed, 1 Mar 2017 01:12:10 +0000 (01:12 +0000)]
testsdk: Handle minimal eSDK and avoid download costs
When using a minimal eSDK, testing currently fails as the sdk isn't
populated. We therefore setup the eSDK under test to point at local
sstate and execute a command to ensure the toolchain is populated
since most of the tests depend on this being present.
At the same time, add in a link to DL_DIR through own-mirrors so
that tests which fetch source (e.g. the kernel module one) can
use the local stash. This cuts test execution of the kernel module
test from 2000s to 120s.
We did try using DL_DIR directly but that causes uninative issues
requiring other workarounds so own-mirrors is neater.
Together these fixes unbreak eSDK testing on the autobuilder.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Jussi Kukkonen [Wed, 22 Feb 2017 14:43:05 +0000 (16:43 +0200)]
weston: Upgrade 1.11.1 -> 2.0.0, separate libweston
* Drop two patches that are upstream. Rebase other patches.
* Separate libweston into its own package, modify the recipe
as needed because files have changed location.
* Remove "--disable-rpi-compositor": the backend does not exist
anymore.
Libweston is already at version 2 and is likely to have new major
versions. The versions should be parallel installable (but weston
itself will not be).
Signed-off-by: Jussi Kukkonen <jussi.kukkonen@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Mariano Lopez [Tue, 21 Feb 2017 18:20:35 +0000 (12:20 -0600)]
insane.bbclass: Add missing dependencies
package_qa task requires some tools installed in sysroot; with
the introduction of recipe specific sysroot this task won't
have such tools installed if it's forced to run.
Signed-off-by: Mariano Lopez <mariano.lopez@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Ming Liu [Sun, 26 Feb 2017 07:44:45 +0000 (08:44 +0100)]
generate-manifest-3.5.py: add logic to generate native manifest
python3-native supposes to RPROVIDE all native packages as added in
generate-manifest-3.5.py, but it does not so far, this leads a problem
that sometimes bitbake cant find a runtime provider for a python3-*-native
when a new runtime dependency on it being required, this usualy happens
after a new native python3-* recipe is created or the old native python3-*
recipes are upgraded.
To avoid manually extending RPROVIDE every time when a new runtime
dependency is introduced, an argument '-n/--native' is added to the
manifest generator, allowing it create a native python3 manifest, with a
RPROVIDE line only, the RPROVIDE should contain all the sub-packages.
The generated python-native-3.5-manifest.inc is also added which is
included by python3-native recipe.
Signed-off-by: Ming Liu <peter.x.liu@external.atlascopco.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Ming Liu [Sun, 26 Feb 2017 07:39:19 +0000 (08:39 +0100)]
generate-manifest-2.7.py: add logic to generate native manifest
python-native supposes to RPROVIDE all native packages as added in
generate-manifest-2.7.py, but it does not so far, this leads a problem
that sometimes bitbake cant find a runtime provider for a python-*-native
when a new runtime dependency on it being required, this usualy happens
after a new native python-* recipe is created or the old native python-*
recipes are upgraded.
To give a example, the following commit is trying to address such a issue:
commit 4583cd1bb15306e8f0ab7bcd80732e6f35aa4533:
[ python-native: Make python-native also RPROVIDE python-unittest-native ]
To avoid manually extending RPROVIDE every time when a new runtime
dependency is introduced, an argument '-n/--native' is added to the
manifest generator, allowing it create a native python manifest, with a
RPROVIDE line only, the RPROVIDE should contain all the sub-packages.
The generated python-native-2.7-manifest.inc is also added which is
included by python-native recipe.
Signed-off-by: Ming Liu <peter.x.liu@external.atlascopco.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Humberto Ibarra [Mon, 27 Feb 2017 16:49:09 +0000 (10:49 -0600)]
oelib/buildhistory.py: Add unittest for buildhistory_analysis
The buildhistory_analysis module (in which buildhistory-diff is
based) was lacking unittest for its functions. Created selftest
module for this and a few testcases to cover basic cases.
[YOCTO #10727]
Signed-off-by: Humberto Ibarra <humberto.ibarra.lopez@intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Mariano Lopez [Mon, 27 Feb 2017 07:45:01 +0000 (07:45 +0000)]
oeqa/core/loader.py: Avoid importing tests with built-ins name
If importing a test with the same name as a built-in module,
it will silently import the built-in and check for tests in
built-in module. This happened with syslog module in debian
based machines, so add a raise to avoid this behavior.
[YOCTO #10978]
Signed-off-by: Mariano Lopez <mariano.lopez@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Mariano Lopez [Tue, 21 Feb 2017 14:33:08 +0000 (14:33 +0000)]
selftest: Optimize get_bb_var use
get_bb_var calls bitbake every time it is used and every call
would take about 7 seconds. There are tests that calls get_bb_var
several times when they can use get_bb_vars. Also there are tests
that calls it to fetch the same variable over and over again.
This will optimize the use of get_bb_var and get_bb_vars for a
little speed up in the tests.
[YOCTO #11037]
Signed-off-by: Mariano Lopez <mariano.lopez@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Maxin B. John [Fri, 24 Feb 2017 16:07:08 +0000 (18:07 +0200)]
gstreamer: update to 1.10.4
bug-fix release:
*) Various fixes for crashes, assertions, deadlocks and memory leaks on
fuzzed input files and in other situations (CVE-2017-5847, CVE-2017-5848)
*) gst-libav was updated to ffmpeg 3.2.4, fixing a couple of CVEs
Signed-off-by: Maxin B. John <maxin.john@intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Pavel Modilaynen [Fri, 24 Feb 2017 10:22:20 +0000 (11:22 +0100)]
buildstats.bbclass: Avoid index exception in /proc/PID/io parsing
There is some probability (depends on system load) to get empty
or line containing "0" as the last line while reading /proc/PID/io.
Avoid build failure by checking if line contains separator
":" before split.
Signed-off-by: Pavel Modilaynen <pavelmn@axis.com> Signed-off-by: Daniel Lublin <daniel@lublin.se> Signed-off-by: Ross Burton <ross.burton@intel.com>
Remove a reference to WebKit/LICENSE, as the directory is not there anymore
(and the LICENSE file in it hasn't been moved somewhere else) - it was
a relic of webkit 1.x era.
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Rebase monitor-option.patch and no-msgfmt-check.patch
Add support for python 3 probes
(this necessitates the 0001-Do-not-let-configure-write-a-python-location-into-th.patch
and the Add 0001-Install-python-modules-to-correct-library-dir.patch to
install python modules to correct destination).
Add perl to RDEPENDS to fix a QA warning about one of the installed files.
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
fcntl-fix-the-time-def-to-use-time_t.patch to
0039-fcntl-fix-the-time-def-to-use-time_t.patch
7. The new version of ltp add the checking for ksh and csh, we doesn't support
ksh and csh in oe-core, so remove the checking files when the system executes
do_install function.
Signed-off-by: Dengke Du <dengke.du@windriver.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Nathan Rossi [Wed, 22 Feb 2017 05:58:57 +0000 (15:58 +1000)]
glib-2.0/glib.inc: If 'charset.alias' exists drop it
Delete the 'charset.alias' file during the do_install task if it exists.
This file is not shipped in any packages and on the targets that it is
generated (musl and mingw32) it is undesired.
Signed-off-by: Nathan Rossi <nathan@nathanrossi.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Nathan Rossi [Wed, 22 Feb 2017 05:58:57 +0000 (15:58 +1000)]
glib-2.0/glib.inc: Only apply MLPREFIX renaming to class-target
When renaming the 'gio-querymodules' binary to prefix with the current
MLPREFIX, only do this for target packages where the MLPREFIX makes
sense. This skips renaming for nativesdk targets which have MLPREFIX set
to 'nativesdk-'.
Signed-off-by: Nathan Rossi <nathan@nathanrossi.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Markus Lehtonen [Wed, 22 Feb 2017 15:26:50 +0000 (17:26 +0200)]
license.bbclass: run chown only under pseudo
The copy_license_files() function in license.bbclass is called in two
different contexts. First, it is run as part of the do_populate_lic
task. In addition, it is called from do_package task when
LICENSE_CREATE_PACKAGE is enabled. The function has code that changes
the owner of license files to root which is meant only to happend in the
latter case - i.e. under do_package which is run under pseudo.
Previously, the code was blindly always running chown and just ignored
errors that happened when running from do_populate_lic. This patch
changes it to be more intelligent, only doing chown when running under
pseudo.
[YOCTO #10897]
Signed-off-by: Markus Lehtonen <markus.lehtonen@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Drop backported 0002-boost-test-execution_monitor.hpp-fix-mips-soft-float.patch
Rebase consider-hardfp.patch to 0001-When-using-soft-float-on-ARM-we-should-not-expect-th.patch
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Zhixiong Chi [Wed, 22 Feb 2017 07:14:42 +0000 (15:14 +0800)]
unzip: CVE-2014-9913 CVE-2016-9844
Backport the patches for CVE-2014-9913 CVE-2016-9844
CVE-2016-9844:
Buffer overflow in the zi_short function in zipinfo.c in Info-Zip
UnZip 6.0 allows remote attackers to cause a denial of service
(crash) via a large compression method value in the central
directory file header.
CVE-2014-9913:
Buffer overflow in the list_files function in list.c in Info-Zip
UnZip 6.0 allows remote attackers to cause a denial of service
(crash) via vectors related to the compression method.
Patches come from:
https://github.com/NixOS/nixpkgs/blob/master/pkgs/tools/archivers/unzip/ or
https://release.debian.org/proposed-updates/stable_diffs/unzip_6.0-16+deb8u3.debdiff
Mike Looijmans [Wed, 22 Feb 2017 14:32:09 +0000 (15:32 +0100)]
initscripts/sysfs.sh: mount configfs if present
configfs is another kernel virtual file system that should be mounted
if configured, so if it's configured into the kernel, mount it. It is
used to configure e.g. USB gadget mode and devicetree overlays.
Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl> Signed-off-by: Ross Burton <ross.burton@intel.com>
Andre McCurdy [Wed, 22 Feb 2017 10:21:42 +0000 (02:21 -0800)]
bitbake.conf: fix ineffective include conf/target/${TARGET_SYS}.conf
TARGET_SYS is defined in terms of TARGET_ARCH, so it's not valid
until after TUNE_ARCH has been set by the machine config. The
original order of includes resulted in an attempt to include
non-existent files such as:
conf/target/INVALID-oe-linux.conf
Signed-off-by: Andre McCurdy <armccurdy@gmail.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Mariano Lopez [Tue, 21 Feb 2017 14:33:07 +0000 (14:33 +0000)]
selftest: Avoid sstate corruption by calling cleansstate
Currently selftest doesn't use sstates because some tests
clean sstate cache; using sstates would give a performance
boost instead of building everything from scratch.
With this sstates are not corrupted using different methods
depending on tests:
devtool: These tests needed to delete the cache so SSTATE_DIR
as SSTATE_MIRRORS and set a temporal SSTATE_DIR.
sstatetests: This module already used a temporal SSTATE_DIR, so
just set up the SSTATE_MIRRORS.
Rest: Removed cleansstate, some of them required to force a
certain task, others were just removed or changed for another
task.
[YOCTO #10929]
Signed-off-by: Mariano Lopez <mariano.lopez@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Markus Lehtonen [Thu, 23 Feb 2017 10:24:18 +0000 (12:24 +0200)]
rpm: support customizing gpg command line
Add a new %_gpg_sign_cmd_extra_args macro that allows customizing the
gpg options used when signing rpm packages. This is needed to be able to
sign packages with gpg 2.1 which requires "--pinentry-mode loopback" to
allow non-interactive signing.
[YOCTO #11054]
Signed-off-by: Markus Lehtonen <markus.lehtonen@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Chen Qi [Wed, 22 Feb 2017 03:13:44 +0000 (11:13 +0800)]
sudo: upgrade to 1.8.19p2
The license checksum for doc/LICENSE is changed. It's a small change.
'2015' is changed to '2017'. Nothing else is changed. So the licenses
remain the same.
Signed-off-by: Chen Qi <Qi.Chen@windriver.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Aníbal Limón [Tue, 21 Feb 2017 18:34:52 +0000 (12:34 -0600)]
dpkg: Upgrade to 1.18.10
Add a patch to don't use --clamp-time when call tar because
isn't supported in tar hosts versions. See
0007-dpkg-deb-build.c-Remove-usage-of-clamp-mtime-in-tar.patch
patch for details.
Multiple people have expressed issues with flex-2.6.2; personally I had
problems compiling libsepol from meta-selinux (for libselinux). I tried
upgrading to flex-2.6.3, but that caused binutils-cross_2.27 to fail.
The simplest for now is to downgrade to flex-2.6.0.
Signed-off-by: Ross Burton <ross.burton@intel.com>
It is generated by kernel's "make clean _mrproper_scripts", the
kernel-devsrc includeds full sources, this symlink is not needed, and
the path is invalid on target, so remove it.
Signed-off-by: Robert Yang <liezhi.yang@windriver.com> Signed-off-by: Ross Burton <ross.burton@intel.com>
Aníbal Limón [Thu, 16 Feb 2017 16:45:32 +0000 (10:45 -0600)]
classes/license.bbclass: Don't copy unneeded licenses by package
Usually a recipe only provides one package but when provides more
than one package the LICENSE variable per package (i.e. linux-firmware)
needs to take into account to avoid unnecesary copy of licenses into
packages.
The patch validates if LICENSE exists in package LICENSES in order to
don't copy unneeded licenses.
As result of this patch some packages will not contain licenses there
are not into LICENSE variable.
For example:
acl contains GPLv2+ instead of GPLv2+ and LGPLv2.1+
libacl contains LGPLv2+ instead of GPLv2+ and LGPLv2.1+
This behaviour is declared on the acl recipe as:
SUMMARY = "Utilities for managing POSIX Access Control Lists"
HOMEPAGE = "http://savannah.nongnu.org/projects/acl/"
SECTION = "libs"
Aníbal Limón [Thu, 16 Feb 2017 16:45:31 +0000 (10:45 -0600)]
license.bbclass: Don't copy again LICENSE already handled as no-generic
The NO_GENERIC_LICENSE mapping was added [1] to enable copy LICENSES
from upstream source code into recipe licenses, previously that only
common-licenses was processed.
This result on copy twice the NO_GENERIC_LICENSE specified because there
is a mapping between license in LIC_CHKSUM and NO_GENERIC_LICENSE.
In order to avoid double copy one as generic_ and other as LICENSE. keep
track of licenses already copied.
For linux-firmware the result will be only generic_ licenses into
common-licenses.
It isn't clear that the README_-_DO_NOT_DELETE_FILES_IN_THIS_DIRECTORY.txt
file in the deploy directory warrants the complexity it brings elsewhere.
Let's just remove it entirely.
In particular, if two do_image_complete tasks run in parallel they risk
both trying to put their image into ${DEPLOY_DIR_IMAGE} at the same time.
Both will contain a README_-_DO_NOT_DELETE_FILES_IN_THIS_DIRECTORY.txt
file. In theory this should be safe because "cp -alf" will just cause one
to overwrite the other. Unfortunately, coreutils cp also has a race[1]
which means that if one copy creates the file at just the wrong point the
other will fail with:
cp: cannot create hard link ‘..../tmp-glibc/deploy/images/pantera/README_-_DO_NOT_DELETE_FILES_IN_THIS_D.txt’ to
+‘..../tmp-glibc/work/rage_against-oe-linux-gnueabi/my-own-image/1.0-r0/deploy-my-own-image-complete/README_-_DO_NOT_DELETE_FILES_IN_THIS_DIRECTORY.txt’: File exists
Patrick Ohly [Thu, 23 Feb 2017 17:38:55 +0000 (18:38 +0100)]
ovmf: increase path length limit
The VfrCompile tool has a hard-coded maximum length for path names
which turned out to be too small by around 20 characters in the
Yocto autobuilder setup. Increasing the maximum by a factor of 4
is relatively easy and makes the problem less likely.
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com>
