Elliot Smith [Tue, 8 Mar 2016 11:32:12 +0000 (11:32 +0000)]
bitbake: toaster: rework task buildstats storage and display
The data available from buildstats is now more fine grained than
previously, so take advantage of that to enrich the data we save
against tasks:
* Store the CPU usage for user and system separately, and display
them separately.
* Disk IO is now measured in bytes, not ms. Also store the
read/write bytes separately.
* Store started and ended times, as well as elapsed_time. This
will enable future features such as showing which tasks were
running at a particular point in the build.
There was also a problem with how we were looking up the Task
object, which meant that the buildstats were being added to
new tasks which weren't correctly associated with the build. Fix
how we look up the Task (only looking for tasks which match the
build, and the task and recipe names in the build stats data) so
the build stats are associated with the correct task.
Signed-off-by: Elliot Smith <elliot.smith@intel.com> Signed-off-by: Michael Wood <michael.g.wood@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Sujith H <sujith.h@gmail.com> Signed-off-by: Elliot Smith <elliot.smith@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
bitbake: fetch2: Make SRC_URI[md5sum] and SRC_URI[sha256sum] expand their values
For some reason, the values for SRC_URI[md5sum] and SRC_URI[sha256sum]
were not being expanded. That lead to the following code not working
as expected:
Ed Bartosh [Wed, 9 Mar 2016 08:07:40 +0000 (10:07 +0200)]
bitbake: xmlrpc: fix bug in setting XMLRPCServer.single_use
XMLRPCServer.single_use attribute was always set to False.
This caused xmlrpc server to keep running after build is done as
BitBakeServerCommands.removeClient only shuts down server if its
single_use attribute is set to True.
The output of "npm view dependencies" isn't entirely JSON if there are
multiple results, but the code here was just discarding the output if
the entire thing didn't parse as JSON. Split the output into lines and
iterate over it, parsing JSON fragments as we find them; this way we end
up with the last package's dependencies since it'll be last in the
output.
Digging further, it seems that the dependencies field reported by "npm
view" also includes optional dependencies. That wouldn't be a problem
except some of these optional dependencies may be OS-specific; for
example the "chokidar" module has "fsevents" in its optional
dependencies, but fsevents only works on MacOS X (and is only needed
there). If we erroneously pull in fsevents, not only is it unnecessary
but it causes "npm shrinkwrap" to throw a tantrum. In the absence of a
better approach, look at the os field and discard the module (along with
any of its dependencies) if it isn't for Linux.
As part of this, we can reduce the calls to npm view to one per package
since we get the entire json output rather than querying twice for two
separate fields. Overall the time taken has probably increased since we
are being more thorough about dependencies, but it's not quite as bad as
it could have been.
Paul Eggleton [Wed, 9 Mar 2016 04:22:12 +0000 (17:22 +1300)]
bitbake: fetch2/npm: fix errors with some version specifications
"2 || 3" is a valid version specification for a dependency in an npm
package.json file, but of course that looks like something else when
sent to a shell. Quote the version value to avoid this.
The patch commit message should have mentioned that the changes were
to update populate_sdk_ext after the changes to uninative now the
download is placed into a specific directory in DL_DIR. We also
need to specify the uninative tarball checksum.
Paul Eggleton [Wed, 9 Mar 2016 04:48:55 +0000 (17:48 +1300)]
recipetool: create: shrinkwrap and lockdown npm modules
"npm shrinkwrap" creates a file that ensures that the exact same
versions get fetched the next time the recipe is built. lockdown is
similar but also includes sha1sums of the modules thus validating they
haven't changed between builds. These ensure that the build is
reproducible.
Paul Eggleton [Wed, 9 Mar 2016 04:48:54 +0000 (17:48 +1300)]
recipetool: create: support creation of additional files by plugins
Allow plugins to create additional files to go alongside the recipe. The
plugins don't know what the output filename is going to be, so they need
to put the files in a temporary location and add them to an "extrafiles"
dict within extravalues where the destination filename is the key and
the temporary path is the value.
devtool add was also extended to ensure these files get moved in and
preserved upon reset if they've been edited by the user.
Paul Eggleton [Wed, 9 Mar 2016 04:48:52 +0000 (17:48 +1300)]
recipetool: create: split npm module dependencies into packages
Rather than rolling all of an npm module's dependencies into the same
package, split them into one module per package, setting the SUMMARY and
PKGV values from the package.json file for each package. Additionally,
mark each package with the appropriate license using the license
scanning we already do, falling back to the license stated in the
package.json file for the module if unknown. All of this is mostly in
aid of ensuring all modules and their licenses now show up in the
manifests for the image.
Additionally we set the main LICENSE value more concretely once we've
calculated the per-package licenses, since we have more information at
that point.
Paul Eggleton [Wed, 9 Mar 2016 04:48:51 +0000 (17:48 +1300)]
recipetool: create: add license file crunching
Matching license texts directly to md5sums only goes so far. Some
licenses make the copyright statement an intrinsic part of the license
statement (e.g. MIT) which of course varies between projects. Also,
people often seem to take standard license texts such as GPLv2 and
reformat them cosmetically - re-wrapping lines at a different width or
changing quoting styles are seemingly popular examples. In order to
match license files to their actual licenses more effectively, "crunch"
out these elements before comparing to an md5sum. (The existing plain
md5sum matching has been left in since it's a shortcut, and our list of
crunched md5sums isn't a complete replacement for it.)
As always, this code isn't providing any guarantees (legal or otherwise)
that it will always get the license correct - as indicated by the
accompanying comments the LICENSE values it writes out to the recipe are
indicative and you should verify them yourself by looking at the
documentation supplied from upstream for the software being built if you
have any concerns.
Paul Eggleton [Wed, 9 Mar 2016 04:48:49 +0000 (17:48 +1300)]
recipetool: create: improve mapping for autotools program macros
Make the following improvements to mapping items specified in
AC_CHECK_PROG, AC_PATH_PROG and AX_WITH_PROG to recipes/classes:
* Produce a map of native recipe -> binary for all binaries currently in
STAGING_BINDIR_NATIVE and use this when mapping items
* Add some more entries to the class map
* Ignore autotools binaries since they are covered by the inherit of
autotools
* Ignore coreutils-native since that would almost always be a bogus
dependency
python3: fix patching get_python_lib() in distutils/sysconfig.py
Previous, two things were wrong:
1) lib_basename was set from STAGING_LIBDIR only if prefix parameter was empty or missing
2) if prefix was not empty, lib_basename reverted to sys.lib, even if STAGING_LIBDIR
should've overriden it
Mark Hatle [Thu, 25 Feb 2016 15:49:57 +0000 (09:49 -0600)]
rpm: Enable specific crypto and digest settings via variables
Allow the user to set the specific digest and non-repudiable signature
algorithms. This should be done on a distribution wide basis.
See recipe for exact instructions, but values are now set using:
RPM_FILE_DIGEST_ALGO (default 1 - md5)
RPM_SELF_SIGN_ALGO (default DSA)
Also, change the PACKAGECONFIG to define the default crypto engine for
RPM5. Not just the available crypto engines. If a crypto engine is not
selected, the system will default to the internal beecrypt version.
Mark Hatle [Fri, 4 Mar 2016 19:03:52 +0000 (13:03 -0600)]
security_flags.inc: Special flags are needed for RPM
RPM interally has support to build and work with the stack protector.
This is disabled by default in the RPM package, and the proper settings
should be specified in the security_flags. Using the default setting of
stack-protector-strong causes linking problems due to issues with libtool
selecting the wrong GCC objections to link against.
Falling back to the RPM values of stack-protector will permit linking to
work properly, and some level of protection.
Bruce Ashfield [Tue, 8 Mar 2016 04:22:22 +0000 (23:22 -0500)]
linux-yocto/4.4: explicitly enable ftrace in tracing fragment
The recent split/factoring between production and development
kernel configurations, EXPERT, EMBEDDED and DEBUG_KERNEL are no longer
selected for all kernel types. This means that ftrace is no longer
selected by default in in standard BSPs, causing breakage in things
that relied on it.
Signed-off-by: California Sullivan <california.l.sullivan@intel.com> Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The firmware can perform a scheduled scan with not matchsets passed,
but it can't send notification that results were found. Since the
userspace then cannot know when we got new results and the firmware
wouldn't trigger a wake in case we are sleeping, it's better not to
allow scans without matchsets.
This fixes https://bugzilla.kernel.org/show_bug.cgi?id=110831
Bruce Ashfield [Fri, 4 Mar 2016 18:21:16 +0000 (13:21 -0500)]
linux-yocto/kernel-meta: ktype refactoring: move DEBUG_KERNEL, EXPERT and EMBEDDED
Integrating the folliwing patch series from Cal:
This patch series refactors the ktypes so that base and standard ktypes
do not enable EMBEDDED, EXPERT, or DEBUG_KERNEL. The reason this
decision was made is because production platforms likely do not want
DEBUG_KERNEL enabled, and EMBEDDED selects EXPERT which selects
DEBUG_KERNEL.
A new ktype called "developer" was also created. This ktype
enables the options now missing from standard and base, making it easy
to maintain the functionality of a BSP through simply swapping the ktype
from standard to developer. The preempt-rt ktype is now based off of
developer in order to maintain its functionality.
The new standard ktype does not include EMBEDDED, EXPERT, or
DEBUG_KERNEL. Without DEBUG_KERNEL it loses a number of debug features
that are selected by default. Without EXPERT it gains RFKILL_INPUT and
DEBUG_MEMORY_INIT, while losing VMSPLIT_3G. These are only available to
configure with CONFIG_EXPERT=y and default to EXPERT or !EXPERT. Not
selecting EMBEDDED has no apparent impacts.
Some coordination is required for this change, as existing BSPs WILL be
affected, and will either need to accept the changes in the standard
ktype or move to the developer ktype.
California Sullivan (12):
features/debug: add debug-kernel feature
ktypes: add developer ktype
ktypes/base: Disable EMBEDDED and DEBUG_KERNEL
CONFIG_PROCESSOR_SELECT: do not enable
intel-common-drivers.scc: move profiling and latencytop to a new file
romley.scc remove profiling and latencytop features
bsp/intel-common: add intel-core* developer BSPs
preempt-rt.scc: include developer ktype instead of standard
intel-common: add intel-developer-drivers.scc to preempt-rt BSPs
CONFIG_I2C_I801: set option to yes in intel-core* BSPs
bsp: add developer common-pc BSPs
bsp: remove profiling and latencytop from non-developer common-pc BSPs
Signed-off-by: California Sullivan <california.l.sullivan@intel.com> Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com> Signed-off-by: Ross Burton <ross.burton@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ross Burton [Tue, 8 Mar 2016 11:27:35 +0000 (11:27 +0000)]
xmlto: tell xmlto where cp is
xmlto looks for a cp binary, and on e.g. Fedora 23 will find it at /usr/bin/cp
but most other distros have it at /bin/cp. This causing problems with sharing
sstate between distributions, but as /bin is a symlink on F23 we can safely
force the path to /bin/cp.
Elliot Smith [Tue, 8 Mar 2016 11:27:24 +0000 (11:27 +0000)]
toaster.bbclass: improve how we gather buildstats for Toaster
Clean up the code which gathers buildstats for Toaster,
and modify the field names so that the correct parts of the
buildstats files are used to derive the CPU usage values.
Also derive elapsed time for the build here, rather than in
Toaster, as we have ready access to the data in the correct
format.
Signed-off-by: Elliot Smith <elliot.smith@intel.com> Signed-off-by: Michael Wood <michael.g.wood@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Hongxu Jia [Mon, 29 Feb 2016 19:41:42 +0000 (19:41 +0000)]
ghostscript: 9.16 -> 9.18
- Backport patches to fix build failure caused by
'--without-jbig2dec' and '--without-jbig2dec'.
...
|make[1]: *** No rule to make target `obj/sjbig2_.dev',
needed by `obj/sjbig2.dev'. Stop.
...
http://bugs.ghostscript.com/show_bug.cgi?id=696497
- Previously, it did not build the whole local libpng
source in ghostscript, only picked up specific files
and compile them. But on ghostscript 9.18, when the
arm's FPU has been set to NEON (-mfpu=neon * with GCC),
the selected file "libpng/pngrutil.c" needs to link
'png_init_filter_functions_neon' which should be
compiled by a non-selected file "libpng/arm/arm_init.c".
...
|./obj/pngrutil.o: In function `png_init_filter_functions':
|armv7a-neon-poky-linux-gnueabi/ghostscript/9.18-r0/build/../
ghostscript-9.18/libpng/pngrutil.c:3921: undefined reference to
`png_init_filter_functions_neon'
...
So do not compile local libpng source in ghostscript,
use shared libpng to instead.
Signed-off-by: Li Zhou <li.zhou@windriver.com> Signed-off-by: Ross Burton <ross.burton@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Randy Witt [Tue, 1 Mar 2016 18:08:21 +0000 (10:08 -0800)]
populate_sdk_ext: Make populate_sdk_ext nostamp
Since the metadata for multiple layers will be put into the sdk, anytime
those layers change the sdk needs to be rebuilt. Ideally, this would use
checksums to only rebuild when necessary.
However, since this feature needs to be in the release, opt for a less
intrusive change by setting the task to nostamp. Unnecessary rebuilds
may occur, but it's better than a user wondering why their changes
didn't get added.
Signed-off-by: Randy Witt <randy.e.witt@linux.intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
license.bbclass: fix host contamination warnings for license files
We get below host contamination warnings of license files for
each recipe, when we try to create a separate ${PN}-lic package (which
contains license files), by setting LICENSE_CREATE_PACKAGE equal to "1"
in local.conf.
-- snip --
WARNING: QA Issue: libcgroup: /libcgroup-lic/usr/share/licenses/libcgroup/generic_LGPLv2.1 is owned by uid 5001, which is the same as the user running bitbake. This may be due to host contamination [host-user-contaminated]
WARNING: QA Issue: attr: /attr-lic/usr/share/licenses/attr/libattr.c is owned by uid 5001, which is the same as the user running bitbake. This may be due to host contamination [host-user-contaminated]
WARNING: QA Issue: bash: /bash-lic/usr/share/licenses/bash/COPYING is owned by uid 5001, which is the same as the user running bitbake. This may be due to host contamination [host-user-contaminated]
-- CUT --
Since the license files from source and OE-core, are populated in a normal
shell environment rather in pseudo environment (fakeroot); the ownership of
these files will be same as host user running bitbake. During the do_package
task (which runs in pseudo environment (fakeroot)), os.link preserves the
ownership of these license files as host user instead of root user.
This causes license files to have UID same as host user id and resulting in
above warnings during do_package_qa task.
Changing ownership of license files to root user (which has UID and GID as 0)
under pseudo environment will solve above warnings, and on exiting pseudo
environment the license files will continue to be owned by host user. Perform
this manipulation within try/except statements, as tasks which are not exected
under pseudo (such as do_populate_lic) result in OSError when trying to
change ownership of license files.
Signed-off-by: Daniel Istrate <daniel.alexandrux.istrate@intel.com> Signed-off-by: Ross Burton <ross.burton@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Alejandro del Castillo <alejandro.delcastillo@ni.com> Signed-off-by: Ross Burton <ross.burton@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Andre McCurdy <armccurdy@gmail.com> Signed-off-by: Ross Burton <ross.burton@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Li Zhou [Tue, 1 Mar 2016 08:35:19 +0000 (16:35 +0800)]
net-tools: Add SCTP option support
Porting three net-tools SCTP related patches from
<https://archive.fedoraproject.org/pub/archive/fedora/linux/releases/
12/Fedora/source/SRPMS/net-tools-1.60-95.fc12.src.rpm> to add support
for SCTP option.
Signed-off-by: Li Zhou <li.zhou@windriver.com> Signed-off-by: Ross Burton <ross.burton@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Patrick Ohly [Mon, 7 Mar 2016 14:51:14 +0000 (15:51 +0100)]
image.bbclass: support chaining compression (aka conversion) commands
It makes sense to use the compression mechanism also for conversion,
for example of a whole-disk image into .vdi (VirtualBox). That part
already works, like this:
But then it also makes sense to allow compressing the resulting image,
which only works after enhancing the image.bbclass.
For example, suppose a custom image command produces "dsk" images. Then
it becomes possible to set
IMAGE_FSTYPES = " dsk.xz dsk.vdi.xz"
and do_image_dsk will automatically produce the intermediate images,
convert to dsk.xz resp. dsk.vdi -> dsk.vdi.xz and delete all
intermediate images. Symlinks are also set correctly.
The patch for making the .rootfs configurable was incomplete: in the
python create_symlinks() method the new variable must be expanded
explicitly.
Not doing so broke the symlink creation and that led to hard build
failures in image types depending on the boot-directdisk.bbclass (like
qcow2) because the build_boot_dd() method relied on the symlink.
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ed Bartosh [Thu, 3 Mar 2016 06:47:42 +0000 (22:47 -0800)]
bitbake: toasterui: make toasterui to work in build mode
Currently toasterui works only in observer mode. This is
artificial limitation which was made to support current toaster
design. As we decided to stop using bitbake server we'll
need to run toasterui also in build mode.
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ed Bartosh [Thu, 3 Mar 2016 06:47:41 +0000 (22:47 -0800)]
bitbake: toasterui: check if setEventMask succeeded
Currently toasterui ignores return value of setEventMask
command, which created confusing difference between set of
events set by this command and the real set used in the code.
Checked if setEventMask succeeded. Print error message and
exit if it's not.
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ed Bartosh [Thu, 3 Mar 2016 06:47:40 +0000 (22:47 -0800)]
bitbake: command: make setEventMask readonly
Executing setEventMask command when bitbake server is in readonly
mode causes runCommand to fail with the following error:
'Not able to execute not readonly commands in readonly mode'
Set readonly attribute for setEventMask command to make it working
for Toaster UI. This should not do any harm as this command doesn't
influence cooker state.
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ed Bartosh [Thu, 3 Mar 2016 05:26:56 +0000 (21:26 -0800)]
bitbake: toaster: remove sshbecontroller module
The code of this module is broken for a long time.
The functionality of it can be easily achieved by running
'manage.py runbuilds.py' on remote machine.
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ed Bartosh [Thu, 3 Mar 2016 05:26:54 +0000 (21:26 -0800)]
bitbake: toaster: raise NotImplementedError
Raised NotImplementedError instead of Exception to be
able to catch it.
This is a preparation for removing sshbecontroller module.
It has to be done as code in bldcontrol/tests.py imports custom
NotImplementedException from sshbecontroller.
Signed-off-by: Ed Bartosh <ed.bartosh@linux.intel.com> Signed-off-by: brian avery <avery.brian@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
bitbake: toaster: change 'revision' to 'Git revision'
I've received some feedback on the 'Revision' label we use in the import
layer page. It is not quite communicating that what's required is a Git
revision. Changing it to 'Git revision' to make it a bit more specific.
The change applies not only to the import layer page, but to all pages
showing revision information in the project configuration section.
Michael Wood [Mon, 7 Mar 2016 12:01:48 +0000 (12:01 +0000)]
bitbake: toaster: views api Package info return both kinds of RDEPENDS
As we do with the popover snippet which shows dependencies inline in the
table also show dependencies which have both TYPE_TRDEPENDS and
TYPE_RDEPENDS.
Signed-off-by: Michael Wood <michael.g.wood@intel.com> Signed-off-by: Elliot Smith <elliot.smith@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Dave Lerner [Fri, 4 Mar 2016 16:41:27 +0000 (16:41 +0000)]
bitbake: toaster: fixup dependency excludes for customimage
[YOCTO #9156]
For a customized image when adding a dependent package X that depends on
dependency package Y, in addition to adding X to appends_set and Y to
includes_set, make sure that Y is no longer in the excludes_set. Y may
have been added to the excludes_set by a prior package removal.
Signed-off-by: Dave Lerner <dave.lerner@windriver.com> Signed-off-by: Michael Wood <michael.g.wood@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Paul Eggleton [Mon, 7 Mar 2016 04:27:37 +0000 (17:27 +1300)]
bitbake: fetch2/npm: ignore unknown headers in tarballs
Tarballs that are fetched down via npm repositories seem to often have
unknown headers. This doesn't affect our ability to extract the contents
though so we don't really care to see those warnings.
Paul Eggleton [Mon, 7 Mar 2016 04:27:36 +0000 (17:27 +1300)]
bitbake: fetch2/npm: handle alternative dependency syntax
npm allows you to specify a dependency as a Github username+path, or
omit the version entirely. You can hit these if you don't use a
shrinkwrap file, with the result that the code later fails due to the
output of "npm view" being empty; so handle this lazily by just ignoring
this part of the dependency if it's not really a version.
Patrick Ohly [Mon, 7 Mar 2016 17:07:52 +0000 (18:07 +0100)]
image creation: allow overriding .rootfs suffix
By default, the image file name contains ".rootfs" to distinguish the
image file from other files created during image building. However,
for certain image types (for example, .hddimg) the ".rootfs" suffix is
redundant because the type suffix alone already uniquely identifies
the main image file (core-image-minimal-intel-corei7-64.hddimg instead
of core-image-minimal-intel-corei7-64.rootfs.hddimg).
With this change, distros that prefer the shorter image name can
override the .rootfs suffix unconditionally with
IMAGE_NAME_SUFFIX ?= '' in their distro configuration
or with some condition check like this:
python () {
if <whole-disk image format active>:
d.setVar('IMAGE_NAME_SUFFIX', '')
}
The exact logic when to remove the extra suffix depends on the distro
and how it enables its own image type.
Richard Purdie [Fri, 4 Mar 2016 16:30:57 +0000 (16:30 +0000)]
yocto-uninative: Add common include for uninative
uninative has some specific setup requirements. Rather than have everyone
doing this themselves, do this centrally and allow people to opt into it
based on some Yocto Project builds of the uninative tarballs.
Richard Purdie [Mon, 7 Mar 2016 12:04:57 +0000 (12:04 +0000)]
mtools: Drop GCONV_PATH manipulation
Now that nativesdk-glibc handles GCONV_PATH itself we don't need to do
this here. This unbreaks mtools for the native case without uninative
since the existing patch wasn't nativesdk specific.
Richard Purdie [Mon, 7 Mar 2016 12:02:51 +0000 (12:02 +0000)]
uninative: Add checksum support
We need to be able to update uninative if the version changes. To do this,
stash a checksum of the installed uninative tarball into a file. If this
changes, we update uninative.
For cleaner download messages, we place the tarballs into directories
based on the checksum.
Richard Purdie [Sat, 5 Mar 2016 08:22:33 +0000 (08:22 +0000)]
uninative: Use CXX11 ABI for interoperation between gcc4 and gcc5
We may see binaries built with gcc5 run or linked into gcc4 environment
so use the older libstdc++ standard for now until we don't support gcc4
on the host system. https://wiki.debian.org/GCC5 has more details about this.
Ross Burton [Fri, 4 Mar 2016 16:48:37 +0000 (16:48 +0000)]
uninative: correctly enable uninative
The previous attempt at soft-failing when uninative was enabled didn't actually
work, because the workers didn't evaluate the function that actually enabled
uninative.
In a BuildStarted handler we can check if we need to download or extract the
uninative tarball.
In a ConfigParsed handler on the workers we can check if the uninative loader is
present, and if so enable it.
Richard Purdie [Mon, 7 Mar 2016 10:46:11 +0000 (10:46 +0000)]
glibc: Add relocation of GCONV_PATH
We're seeing various failures where nativesdk glibc can't find the gconv
modules. We've tried various workaround but this needs fixing 'properly'.
One significant problem is knowing when a binary would use this glibc
and hence when to set the path and when not to.
Add the default path to its own special section which the sdk
relocation script already knows how to handle and remap.
This significantly improves the usability of uninative.
Richard Purdie [Mon, 7 Mar 2016 00:13:48 +0000 (00:13 +0000)]
dkpg: Use tar everywhere (not gtar)
All our systems usually have tar, if we want the native sstate to work
universally, we need to prefer this. This avoids a system with gtar
causing dpkg-native to use it and it not being present on some systems.
Richard Purdie [Fri, 4 Mar 2016 17:10:48 +0000 (17:10 +0000)]
sstate: Allow late expansion of NATIVELSBSTRING
uninative needs to adjust NATIVELSBSTRING fairly late in the
configuration parsing process but the sstate code encodes it into
variables. Since this string doesn't vary on a per recipe basis, we
defer its expansion until usage time.
Prelinking on x86-64 wasn't working out the box as it uses /lib and
not /lib64 for libs. Prelink was refusing to link as the dynamic loader
didn't match its idea of the right path. Passing in the --dyanmic-linker
option avoids this.
We can share code from image-mklibs so abstract that into a new class,
linuxloader.bbclass.
This does break prelinking of multilib images, I've opened a bug so we
can loop back and fix that problem, the code would need to iterate the
dynamic loaders (and setup ld.so.conf files for it).
Richard Purdie [Fri, 4 Mar 2016 16:27:11 +0000 (16:27 +0000)]
insane/prelink: Handle nonstandard library paths
Prelink contains some hardcoded assumptions about the path layout of
the target system. Unfortunately if the system doesn't match, prelink
doesn't work. This breaks:
a) prelink of those images
b) the unsafe-references-in-binaries QA test (which uses prelink-rtld)
One way to work around this is to construct an ld.so.conf file which
lists the library paths in question. We do this in sanity QA check and
in the rootfs prelink code, being careful not to trample any existing
target ld.so.conf.
There is an additional problem that $LIB references in RPATHs won't be
handled correctly, I've not see any system use these in reality though
so this change at least improves things.
Richard Purdie [Thu, 3 Mar 2016 17:18:25 +0000 (17:18 +0000)]
createrepo: Fix stat floating timestamps
When reading the mtime from disk, the system can get a floating point
value. Convert this to an int for comparision purposes, else some
packages always get reindexed as the value in the index is an int.
Ross Burton [Sat, 5 Mar 2016 00:12:03 +0000 (00:12 +0000)]
xmlto: ensure /bin/bash is used as bash
The xmlto script uses bashisms and checks at configure time to find a bash
binary. If the build host has /bin/sh as bash then this gets detected, which
causes problems in native builds if the sstate is then shared to a machine with
/bin/sh as dash.
projects / thirdparty / openembedded / openembedded-core-contrib.git / log
