| Age | Commit message (Collapse) | Author |
|---|
|
Now that none of the packagegroups depend on virtual/kernel, we have the problem
that MACHINE=qemumips bitbake core-image-minimal doesn't put a kernel
into the deploy directory. This breaks many common usecases and
user expectations.
To avoid this, add a dependency on the kernel deploy to image do_build tasks.
This should avoid any circular dependency issues but equally ensure users
have their expectations met.
[YOCTO #5581]
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
gcc has cross and target components with a shared workdir. The unpack umask
settings need to match for all of these. We need to use strings in each
case to ensure the sstate code matches them correctly.
This patch tweaks various things to ensure the change adding the unpack umask
change doesn't break the compiler builds.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* when git checkouts files from fetched clone it respects system umask
and creates files with different permissions, if such files are copied
to packages, resulting target images have also different permissions
on them.
* we need reproducible builds across different builders with different
system umask, so set 022 umask
[YOCTO #5590]
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The comment was originally written for module.bbclass and is now
slightly misleading. This updates it to match the current code.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This reverts commit 6a6735cb98d529c8c5376aedb5c951339f54fe96.
The module class already ensures the scripts are rebuilt correctly. Running
this at sstate installation time is problematic since it can require the
cross compiler. Adding such a dependency would cause issues of its own.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The current code is a little bit overcomplicated, deficient and also
possibly broken.
Issues include:
a) Not maximally optisming rpaths (e.g. a lib in usr/lib might get an
rpath of $ORIGIN/../../usr/lib)
b) The return in the middle of the for loop look suspiciously like
it might break on some binaries
c) The depth function, loops of "../" prepending and so on can
be replaced with a call to os.path.relpath
This patch cleans up the above issues.
Running binaries should result in less "../" resolutions which can't
hurt performance either.
[YOCTO #3989]
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* Fix icecc.bbclass failing with:
ERROR: recipe-name NULL prefix
when it's used with empty TARGET_PREFIX.
* Allarch recipes cannot use compiler at all (even the local one)
so there is no point of using icecc for them.
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Write a list of installed packages to a .manifest file next to the
image, so we can find out what went into the image after it has been
constructed without necessarily having to have buildhistory enabled
(although that will provide more detail.) We can make use of this for
example in the testimage class associated code that checks for installed
packages for determining whether or not to run specific tests.
Note: this replaces the previous ipk-specific manifest code with
something that works for ipk, rpm and deb, and instead of a pruned
status file, packages are listed one per line, in the following format:
<packagename> <packagearch> <version>
Tests for all three backends have shown that the performance impact of
this change is negligible (about 1.5s max).
Implements [YOCTO #5410]
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
check_create_long_filename used a fixed filename for its test files. This
meant that os.remove(testfile) could fail with ENOENT if two instances were
running at the same time against the same sstate directory. Using a
randomly generated filename stops this from happening.
(Although it might seem unlikely, this race did appear to occur multiple
times with Jenkins - presumably because the matrix jobs were all kicked off
at the same time.)
Signed-off-by: Mike Crowe <mac@mcrowe.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Although ENAMETOOLONG is 36 on Linux x86 and x86_64 it does isn't on other
architectures so the value shouldn't be hard coded.
Signed-off-by: Mike Crowe <mac@mcrowe.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This avoids crashing on newer svn layouts where the entries files
don't contain three lines. If someone wants to fix this to
get the right version on newer subversion checkouts, patches
welcome but this at least stops things crashing.
[YOCTO #5363]
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The new do_bundle_initramfs task introduced in
609d5a9ab9e58bb1c2bcc2145399fbc8b701b85a defeats using the sstate
cache. The kernel is resurrected from the sstate cache but ends up being
built again since do_bundle_initramfs depends on do_compile.
The task is no longer nostamp to avoid causing unnecessary rebuilds. The
sstate checksum stamps should know when to rebuild.
The task now runs before do_deploy and part of the work has been moved to
do_deploy where it now writes to ${DEPLOYDIR} rather than
${DEPLOY_DIR_IMAGE} so that the files end up in sstate.
The task can also race against do_install since both call into the kernel
build system. This is fixed by making do_bundle_initramfs run after
do_install (which therefore also fixes the problem that
3baa63b4d588c3262254528b406ede265dd117bf was addressing.)
Signed-off-by: Mike Crowe <mac@mcrowe.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This reverts commit 3baa63b4d588c3262254528b406ede265dd117bf. It broke
builds that aren't using kernel-yocto.
Signed-off-by: Mike Crowe <mac@mcrowe.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
When preping a read-only rootfs and finding some post-install
scripts that can not be run, list the names of said scripts to
avoid having to look around the rootfs to find a list.
Signed-off-by: Jeffrey C Honig <jeffrey.honig@windriver.com>
Signed-off-by: Jeff Polk <jeff.polk@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
On a multilib system when one of the multibs has a different OS then
other multilibs a failure can occur during the install process because
RPM assumes all systems have the same OS.
When an n32 platform is selected as an alternative multilib, it shows
up as mips64_n32-.*-linux-gnun32 in /etc/rpm/platform. This causes
problems when the smart tool tries to add a channel for the multilib.
RPM archScore call always returns zero for arch "mips64_n32" -
after appending default vendor and os, it finds "mips64_n32-wrs-linux"
doesn't match any predefined platforms. Fix this by removing the
restriction of -gnun32 suffix in platform file.
Signed-off-by: Lei Liu <lei.liu2@windriver.com>
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Jeff Polk <jeff.polk@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
PREFERRED_PROVIDER_virtual_pkg has been incorrectly set with more
than one multilib prefixes. For example, if we have two alternative
multilibs lib64 and lib32, PREFERRED_PROVIDER_virtual_pkg will be
set to lib32-lib64-pkg or lib64-lib32-pkg, depending on which
multilib shows up first in the list.
Signed-off-by: Lei Liu <lei.liu2@windriver.com>
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The buildhistory code has fallback paths for older bitbakes for now. The
distrodata class is much less used and it can be assumed a recent bitbake
is used in that case rather than adding fallback code.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The license_create_manifest function contains bashism, this will lead
to unexpected results on ubuntu build host, as sh is linked to dash on
ubuntu. Even if COPY_LIC_MANIFEST and COPY_LIC_DIRS are enabled, the
license files will still be missing on target.
This patch fixes the above problem.
[YOCTO #5549]
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
${KERNEL_OUTPUT} is being renamed/restored in bundle_initramfs task, so we
must ensure bundle_initramfs run after kernel_link_vmlinux where the link
of vmlinux is created as the bootable image.
Signed-off-by: Ming Liu <ming.liu@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
no xorg-abi
...
WARNING: QA Issue: Package lib32-xf86-video-vesa contains Xorg driver (vesa_drv.so) but no xorg-abi- dependencies
WARNING: QA Issue: Package lib32-xf86-video-intel contains Xorg driver (intel_drv.so) but no xorg-abi- dependencies
WARNING: QA Issue: Package lib32-xf86-video-omap contains Xorg driver (omap_drv.so) but no xorg-abi- dependencies
WARNING: QA Issue: Package lib32-xf86-video-vmware contains Xorg driver (vmware_drv.so) but no xorg-abi- dependencies
WARNING: QA Issue: Package lib32-xf86-video-fbdev contains Xorg driver (fbdev_drv.so) but no xorg-abi- dependencies
WARNING: QA Issue: Package lib32-xf86-video-modesetting contains Xorg driver (modesetting_drv.so) but no xorg-abi- dependencies
...
In oe-core 87be2314fe0e67a616b4f713106b4ae46931f5cb, it fixed add_abi_depends
issue while enable multilib.
We did the same thing for qa check: package_qa_check_xorg_driver_abi.
[YOCTO #5545]
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
If kernel_do_compile() encounters a .lzma image for the initramfs, it
attempts to decompress it using the little-known "lzmash" utility. This
may not be present, which will cause the build to fail.
Fortunately, it appears that the more mainstream "lzma" tool is
compatible so we can use that one instead.
Signed-off-by: Phil Blundell <philb@gnu.org>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
The icecc class often calls 'which' for determining paths. This leads
to many messages on stderr in case 'which' doesn't find the
executable. Using bb.utils.which is more appropriate here and doesn't
pollute stderr.
Signed-off-by: Tobias Henkel <tobias.henkel@bmw-carit.de>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
The existing code in checkpkg from distrodata.bbclass had similar
functionality with fetch when searching for latest package version.
For packages that use svn protocol that part was rewrote in order
to use fetcher API.
It now calls latest_revision method from Svn class in fetch2 so
that it gets latest version.
[ YOCTO #1813 ]
Signed-off-by: Irina Patru <irina.patru@intel.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Previously, it supported the situation that /boot area with separate boot
partition:
...
menuentry "Update bzImage-3.10.10-WR6.0.0.0_standard-3.10" {
set root=(hd0,1)
linux /bzImage-3.10.10-WR6.0.0.0_standard root=/dev/sdb1 rw ip=dhcp
}
...
But didn't consider the situation that /boot within root partition:
...
menuentry "Update bzImage-3.10.10-WR6.0.0.0_standard-3.10" {
set root=(hd0,1)
linux /boot/bzImage-3.10.10-WR6.0.0.0_standard root=/dev/sdb1 rw ip=dhcp
}
...
This fix supported them both.
[YOCTO #5514]
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
The commit [kernel.bbclass: test for hardlinks before installing] doesn't
work on all build machines without a space between the ! and the test for
hardlink equivalance. The build continues, while the test fails and no
copy is made into the sysroot. Eventually tasks like build_hdimg will fail
with the missing kernel with the root cause being less than obvious.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
commit df564c4 [kernel: Use hardlinks for do_populate_sysroot for speed]
does indeed speed up the build, but it also means that the KERNEL_OUPUT
and KERNEL_IMAGETYPE may be hardlinks.
If they are hardlinks, install complains that they are actually the same
file and stops the build. We can easily test and avoid the copy if the
file is already in place.
[YOCTO #5527]
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The expected usage of COPY_LIC_MANIFEST and COPY_LIC_DIRS appears to be
to set them to "1" to enable; however the test here is just testing
whether they have a value at all, so setting them to "0" would also
enable them which is somewhat disingenuous. Actually check if they are
set to "1" instead in order to fix this.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This is very old, and we now have pkgdata which is much more complete.
Nobody appears to be using this class, and even OE-Classic had no
current references to it.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This appears to be a very old way of archiving recipes and associated
files. We have better ways of doing this now, and nobody appears to be
using this class - even OE-Classic had no current references to it.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
It's DISTRO_FEATURES; DISTRO_FEATURE is invalid.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
It's LICENSE_FLAGS, LICENSE_FLAG is invalid.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
It's gtk-immodules-cache and GTKIMMODULES_PACKAGES.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
If SDKMACHINE is set then check that a configuration file matching it
actually exists, otherwise the user won't know that they've set it
incorrectly.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* If the md5 parameter is specified with no value, report that the
checksum is not specified instead of reporting that it has changed
* If the md5 checksum has changed, point directly to the license file in
a way that is easy to copy and paste and give the line numbers in an
easy to read form, as well as asking the user to verify that the new
contents matches the current LICENSE value.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
We don't actually need this dummy class; "inherit" can be used with an
expression that evaluates to nothing with current BitBake.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This saves about 2GB on a core-image-sato build so is worth doing and is
consistent with our efforts to try and decrease our build footprint.
Build time in my local test seemed unaffected but on more IO bound
machines it should help.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
bb.fetch.decodeurl() will throw if it doesn't like the look of the URL that
it's given. (Bitbake's idea of what constitutes a valid URL is somewhat
idiosyncratic so it is fairly easy to trip over this by mistake when writing
a recipe.)
If these exceptions are allowed to propagate all the way up to better_exec()
then we will get a large amount of python stack trace spew when they are
finally caught. Avoid that by catching them locally and throwing
bb.build.FuncFailed() with a suitable explanation instead.
Signed-off-by: Phil Blundell <philb@gnu.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
An empty distro value leads to OVERRIDES and FILESOVERRIDES containing
"::" entries which causes odd issues such as files being included when
they shouldn't be. We could put in anonymous python to guard against
empty entries but its messy and setting a default value for DISTRO to
something harmless is much easier.
This patch adds a weak default and ensures the sanity test doesn't
complain about it.
DISTRO_VERSION and SDK_VERSION are also updated to match.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
In the current implementation there can be a race condition while
creating the toolchain archive causing the build to break.
This is fixed by locking the toolchain archiving step using flock.
Signed-off-by: Tobias Henkel <tobias.henkel@bmw-carit.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
There are some recipes which parse the PARALLEL_MAKE variable by their
own and set them to an empty string afterwards. This disables icecc
for this recipe.
Adding a whitelist for forcing icecc makes it possible to use icecc
also with these recipes.
Signed-off-by: Tobias Henkel <tobias.henkel@bmw-carit.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
In the current implementation a KERNEL_CC variable containing shell
evaluation breaks the build process. Shell expansion is not happening
before general expansion in get_cross_kernel_cc which results in a
syntax error and an aborted parse process.
Before expanding the KERNEL_CC variable get_cross_kernel_cc now checks
for backticks or '$(' in the KERNEL_CC variable and performs a shell
evaluation using a call to echo if it finds one.
Signed-off-by: Tobias Henkel <tobias.henkel@bmw-carit.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Some packages are calling the compiler in the install step. In this
case either the build breaks or icecc is not used for building. The
proper environment has to be set to enable icecc based building.
Signed-off-by: Tobias Henkel <tobias.henkel@bmw-carit.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Bitbakes prepend mechanism for the tasks disregards the type of the
function. Thus bitbaking recipes using python functions for configure,
compile or install steps fail due to the missing python version of
set_icecc_env.
Assuming that icecc doesn't need to be used in such situations adding
a dummy python version of set_icecc_env fixes this.
Signed-off-by: Tobias Henkel <tobias.henkel@bmw-carit.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Changing any of the ICECC blacklist variables should not change the
sstate checksum as this doesn't influence the build result.
Signed-off-by: Tobias Henkel <tobias.henkel@bmw-carit.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Currently the icecc class prints a note for every package which
disables parallel make at parse time. This is unneccessary as many
packages don't support parallel building. Changing the log level from
info to debug hides these messages in normal builds without removing
the information when needed.
Signed-off-by: Tobias Henkel <tobias.henkel@bmw-carit.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The current implementation doesn't give a hint about the cause in case
something went wrong in set_icecc_env. This makes it harder to find
out why a package is not being built using icecc. Therefore warnings
are inserted in the various error cases.
Signed-off-by: Tobias Henkel <tobias.henkel@bmw-carit.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
File command in some version could print extra space between
"LSB" and "executable" - it causes mklibs can't find any executables
using grep "LSB executable". Fix the grep pattern to catch
multiple spaces.
Signed-off-by: Lei Liu <lei.liu2@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Changing APPEND wasn't causing syslinux to re-run, so add a manual dependency.
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The kernel tree is large and doesn't need to be copied. Override
the default sysroot handling function to use a hardlink copying
function in python.
This commit also drops the copying of the /lib directory which
just contains the kernel modules. We never use those in the sysroot
so there is little point in carrying those around.
For linux-yocto this takes the do_populate_sysroot time 24s -> 14s.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
Richard Purdie
Martin Jansa
Paul Eggleton
Mike Crowe
Jeffrey C Honig
Lei Liu
Lei Liu
Chen Qi
Saul Wold
Ming Liu
Hongxu Jia
Phil Blundell
Tobias Henkel
Irina Patru
Bruce Ashfield
Ross Burton