There's been a recent discussion about how we can make the Yocto SDK
experience better [1]. One of the ideas was to eliminate the SDK
as a separate artefact altogether and simply provide everything
that the SDK and eSDKs do directly in a yocto build. This does not
mean that people have to 'learn Yocto', but rather that the integrators
should provide a well-functioning sstate cache infrastructure (same as
with minimal eSDK, really), and a few wrapper scripts for setting up the build
and the SDK environment that run layer setup and bitbake behind the scenes.
[1] https://lists.openembedded.org/g/openembedded-architecture/topic/thoughts_on_the_esdk/90990557
So without further ado, here's how you get a 'SDK' without building one:
1. Set up all the needed layers and a yocto build directory.
2. Run:
$ bitbake meta-ide-support
$ bitbake -c populate_sysroot gtk+3
(or any other target or native item that the application developer would need)
$ bitbake populate-sysroots
3. Set up the SDK environment:
. tmp/deploy/images/qemux86-64/environment-setup-core2-64-poky-linux
(adjust accordingly)
Et voila! The Unix environment is now set up to use the cross-toolchain from
Yocto, exactly as in the SDK. And devtool/bitbake are available to extend it,
exactly as in the eSDK.
Theare are numerous benefits here: no need to produce, test, distribute and maintain
separate SDK artifacts. No two separate environments for the yocto build and the SDK.
Less code paths where things can go wrong. Less awkward, gigantic tarballs. Less
SDK update headaches: 'updating the SDK' simply means updating the yocto layers with
git fetch or layer management tooling. Built-in SDK extensibility: just run bitbake
again to add more things to the sysroot, or add layers if even more things are required.
How is this tested?
Exactly same as the regular SDK:
$ bitbake -c testsdk meta-ide-support
This runs the same toolchain tests from meta/lib/oeqa/sdk/cases as the regular
sdk testing does.
(From OE-Core rev: 5c845d7f4ea6ae7ba18ed43180dad28775cace31)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
When using meson-native directly from bitbake this has no effect and
everything works as before; the use case is being able to build
meson-driven projects from the 'direct SDK' environment. Build systems
like cmake/autotools are entirely driven by environment variables,
but meson needs configuration files, and so this replicates how SDKs
set things up.
In particular, with this change the libepoxy build test from the SDK
test suite is able to pass.
(From OE-Core rev: 87d4f6d176f27c22dbb99abc271b9a6eaca314f8)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This is a simpler version of the same class in testsdk.py, as it does not
need to unpack and set up the SDK, and can proceed to the tests straight away.
(From OE-Core rev: be21c62e5bd96164aab9e01168b7a43c6de44c17)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Specifically:
1. Add a few more toolchain items to DEPENDS so they're available in the recipe sysroot.
2. Use deploy mechanism to place the SDK environment setup file and testdata for testing it
into the image deploy directory.
3. Add ability to run SDK tests via the testsdk class and task. This also requires providing
a testdata json file.
4. Ensure sysroot population always runs, as those items are mean to be picked up
by 'bitbake build-sysroots' into the unified sysroot.
(From OE-Core rev: 93f419451cc18b5d31b6fe134ca52a2750fa567d)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Specifically:
1. Place the environment setup file into $B and not into $TMPDIR,
so that the recipe using the class can itself better decide what to do with the file.
2. Use global, unified sysroots (provided through build-sysroots recipe)
and not recipe-specific ones, as this allows flexible on-the-fly management of what
libraries are available to build applications, without having to modify any
recipes, similar to eSDK 'extensible' part.
This also requires adjustment of the sstate sametune_samegsigs test, as meta-ide-support
becomes dependent on $MACHINE (unified sysroots have it in their paths)
and needs to be excluded from the test.
3. Add a few missing settings that have been added to SDK environment files.
4. Add a snippet to the environment setup file that also runs the relocation scripts.
In regular SDKs this is executed by the SDK installer, in direct SDK we can do it when
setting up the environment.
(From OE-Core rev: db5dfd78ae441201778b1175f4fb9a3eba994899)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Correct the override name used for rust toolchain removal from the mingw sdk.
(From OE-Core rev: 8aeb0dfe6508daea51db4f0992eed194e925fdd7)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
It fails to compile mesa{,-native} when DEBUG_BUILD is enabled:
../mesa-22.1.3/src/compiler/nir/nir_inline_helpers.h: In function ‘nir_opt_move_block’:
../mesa-22.1.3/src/compiler/nir/nir_opt_move.c:55:1: error: inlining failed in call to
always_inline ‘src_is_ssa’: indirect function call with a yet undetermined callee
src_is_ssa(nir_src *src, void *state)
^~~~~~~~~~
Backport patch from mesa merge request to fix it.
(From OE-Core rev: 468715c42632a8467dc1f0abe0f3cd540de5f7c3)
Signed-off-by: Kai Kang <kai.kang@windriver.com>
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
When the workspace bbappends are still using the old override syntax
with EXTERNALSRC_pn-*, externalsrc_re will not match, and pn will never
be assigned, leading to a nondescript UnboundLocalError being raised on
the user's terminal. Try to detect that situation and give the user a
hint how to solve it.
(From OE-Core rev: d42ea8e849cf2df3708406418b961168268b316a)
Signed-off-by: Roland Hieber <rhi@pengutronix.de>
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add a new variable SDK_TOOLCHAIN_LANGS to allow different language support
to be selected within SDKs. Initially supported options are rust and go.
(From OE-Core rev: 0026b5687147410e4c7140ca5133089ebb51d9d4)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add a QA test to the SDK to test that a basic cargo build works.
[RP: Tweaked to work for multilibs and updated to match toolchain changes]
(From OE-Core rev: d0cfe587bc897e79ef01805cc9a42fbca28c883c)
Signed-off-by: Otavio Salvador <otavio@ossystems.com.br>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some recipes reference these. Rather than continually trying to chase down the references
and taskhash issues, clear the variables for an easier life and simpler code. These
wouldn't convey anything useful in a native build.
(From OE-Core rev: f8edeead263708889d31a7ff578ef8274cb678b4)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Remove further code duplication by allowing the main function to be
configurable.
(From OE-Core rev: 2d4b35a0db1220e7a7f1b6776223aa4d40a3f246)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Currently most of the rust recipes use this code but it is all piecemeal.
Turn the code into a class where things can start to be rationalised. Ultimately
some of the data and python code should be moved to a python library but one step
at a time.
No functionality changes.
(From OE-Core rev: 3795285cbf362e13b8151bfdbe1bce999ac28641)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some of the subtleties in the different codepaths for target rust json generation
were not easy to spot. Start to simplfy the code to make this clearer.
This patch should not have any functionality change although ABIEXTENSION
has to be excluded from the function signature, the triplet would normally
cover anything set there.
(From OE-Core rev: 0f83d959465e0d99f98ade6803281585931d1b02)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Match the code in rust-cross-canadian so that further simplifications
can be considered in future.
(From OE-Core rev: 5cb62d0a33777cb9afb1eea5f736a2580ce50dc7)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
rust-cross had special handling for armv7 targets but we also need this
for cross-canadian. Merge the code into the main function so everything is
consistent.
Also then fix the arm definition to be arm-eabi since ABI is correctly
being looked up.
(From OE-Core rev: ff3c3dbbd2bf1bb7bb70b55cca203e9eedcf14a8)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Based upon a patch from Otavio Salvador <otavio@ossystems.com.br>,
ensure the target json files are written in the correct order with
the most specific last incase it overwrites earlier files if the prefixes
match.
(From OE-Core rev: 1912c4e9e0ecf9655f3b3a41588b54d7956f5899)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
When a 'BUILD' target is requested we shouldn't be looking at TARGET_SYS but
at BUILD_SYS. Due to the way rust mangles triplets, we need the HOST_SYS triplet
to work with existing code - fixing that issue is a separate patch.
Also drop the arch_abi argument, it doens't make any sense to a getVar() call
and was a copy and paste error.
Based on a patch from Otavio Salvador <otavio@ossystems.com.br> but separated out
and tweaked.
Fixes: bd36593ba3 ("rust-common: Drop LLVM_TARGET and simplify")
(From OE-Core rev: d554161a045d12411f288394e253c54aa4c1257c)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
io_uring is enabled or disabled depending on whether liburing is available,
so add a PACKAGECONFIG to make this explicit, disabled by default.
(From OE-Core rev: daee79639c39ac6278855b35e0ddf71e52dd13f8)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
In cross-compiles CGO_ENABLED=1 needs to be set explicitly, as otherwise
Go refuses to use it even if CC is already set.
This fixes the selftest on setups where the host and the SDK target
don't have matching architectures.
[ YOCTO #14859 ]
(From OE-Core rev: 19be072619d39267df44f23c4c8b64f3808f6148)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
By default 'go mod' creates read-only files, but that just complicates
things. Add -modcacherw to make the cache read/write, so it can be
cleaned up without needing to chmod.
(From OE-Core rev: 7ff30e0d9fe8527cbc2f8ca84e0300fdc84663b6)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* it's not used by any other recipe AFAIK
* meta-python2 has own copy for python-setuptools there in:
meta-python2/recipes-devtools/python/python-setuptools/0001-conditionally-do-not-fetch-code-by-easy_install.patch
(From OE-Core rev: 3a5898861337eb4934360ef6ae448896ec061e26)
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Restore WHENCE_CHKSUM variable which is used to hold the WHENCE file
checksum. It is necessary to allow easily overriding it from local.conf
if the devupstream version is selected:
PREFERRED_VERSION_linux-firmware = "1:20220708+git%"
SRCREV:class-devupstream = "${AUTOREV}"
WHENCE_CHKSUM:class-devupstream:pn-linux-firmware = "abf1077491eeb261ecdcb680a34fc059"
Without the WHENCE_CHECKSUM one would need to manually patch the
LIC_FILES_CHKSUM variable to change the checksum of WHENC (e.g. using
the anonymous python function or remove expression).
(From OE-Core rev: 554be2af1e0a03a2d23032d48afbbe0913a45409)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ignore the rmdir cmd if using usrmerge distro feature since the
intention is to delete /lib or /lib64 but not libdir under /usr and
base_libdir = libdir when usrmerge is enabled in distro
(From OE-Core rev: 3b7f6b0e0f1f16f89cd924aac001b4f661c145ca)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
some standalone targets e.g. riscv64-elf disable shared linking for
baremetal ELF ABI in ld, therefore lets make it a static library
(From OE-Core rev: 3c6219dfcbcbde314648ba8cc54a90b32ea1c952)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This was added to musl specific backport patchset so it escaped testing
(From OE-Core rev: dd5f7e87b5bbde02eb50c7da6a6d3f928e0fd2f2)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
By naming this test class git.py, any attempt to import GitPython (as
needed by oelib.buildhistory) failed.
As this class exercises the intercepts, rename it to intercept.py.
(From OE-Core rev: d557cbbf86767bc2ebf2beb3d70af3b3ca5e0529)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Historically there's been a split between /lib for early boot and
/usr/lib for everything else, but with modern systems this split is
meaningless and incomplete. If a minimal system for early boot is
needed, it should be a full minimal system in a initramfs.
[RP: Fixed up selftest to match]
(From OE-Core rev: 990073dfc167354b4af41db83ac46c18b1aa99d5)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Set V=1 so that the compile log is useful when things go wrong.
(From OE-Core rev: 9815417bda5cbe88b5bc2524cd96c34e8b729050)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some opkg commands print an error during cleanup when the tmp_dir
does not exist and an attempt is made to delete it. The error messages
are harmless and the opkg commands eventually succeed.
When these commands are run and stderr is piped to stdout, the error
messages may clobber the stdout and cause unexpected results while
parsing the output of the command. Therefore, when parsing the output
of a command, do not pipe stderr to stdout. Instead, capture stderr
and stdout separately, and upon success, send stderr to bb.note().
(From OE-Core rev: f2167ae80258253eb47a5b148546b265320284cc)
Signed-off-by: Shruthi Ravichandran <shruthi.ravichandran@ni.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This disables tests for runtime system specific functions like fcntl
there is no other change seen
+configure:77372: checking for fcntl
+configure:77372: result: yes
In libsttdc++ configure log outputs.
[YOCTO #14803]
(From OE-Core rev: 586b6239252ecf2711930b60f7f0f617e38ac737)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Pulls in several CVE fixes.
Added a patch to avoid timer_create cross compile issue (and submitted upstream).
Also submit the race fix upstream.
We disable timer_create in the native case since some systems have it
and some don't so this makes us consistent.
(From OE-Core rev: d0c1de084c7ce030d47a428e4bbfbc4ce2996057)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
`rc` runs all the KILL scripts in a runlevel before the START scripts.
The umountnfs script is currently configured as a START script, and
runs after the networking KILL script. During shutdown, this causes a
~3 minute timeout after networking is shutdown when the system tries
to connect to and unmount any mounted network shares.
Fix this by changing the script configuration to "stop" so that it can
run before networking is stopped and unmount any network shares
safely.
(From OE-Core rev: c419bd4537756e9f6c2fe6da3a9b798526e27eca)
Signed-off-by: Shruthi Ravichandran <shruthi.ravichandran@ni.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Python support was dropped upstream and removed in 8a9a5885995c ("bind:
update 9.16.26 -> 9.18.1"), clean up the remaining pieces of python3 in
the recipe.
(From OE-Core rev: ee4e4eb16a3729dcafad075c42aec1695b8ea15f)
Signed-off-by: Alex Kiernan <alex.kiernan@gmail.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
In bc, we found that there are files newer than Changelog (e.g. scan.l)
which means after install runs, the timestamp is clamped to SDE which means
"bitbake bc -C compile" would then try and regenerate generated files and
complain flex was missing.
Rather than assuming Changelog/NEWS are magic for dates, drop that scan
method and scan all unpacked files. This shouldn't be that slow as they'd
be in the disk cache already after an unpack.
For bc, this changes SDE by about two minutes and avoids the problems
rebuilding.
Add a version comment to the task definition as changes in python library
code won't trigger a rebuild and we need one here.
(From OE-Core rev: 32dda0ad91a9e7946351c897578b4c97ae142341)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The linux kernel headers are necessary for glibc-dev, so
we need to use RDEPENDS instead of DEV_PKG_DEPENDENCY which
specifies RRECOMMENDS.
Currently, in case of NO_RECOMMENDATIONS set to "1", linux
kernel headers are not pulled in by glibc-dev, causing error
like below when compiling.
fatal error: linux/errno.h: No such file or directory
The problem could be reproduced by setting NO_RECOMMENDATIONS
to "1" and then running:
bitbake core-image-minimal -c populate_sdk
bitbake core-image-minimal -c testsdk
(From OE-Core rev: fdb16e1a78c2abcc8ac89678b1b250ca4fa9c0d9)
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If you build perf from Linux 5.4 it still contains buildpaths (pointing
to the sysroot's Python) as the existing fixes don't apply to the lines
in that old release.
Add further expressions to remove the final buildpaths.
(From OE-Core rev: 6783c1e295e5afafd73a02fd986ca8956a89dda7)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Only touching /tmp/.automount-$name is not good enough, it must contain
the mount name, otherwise umount could not get the path from it.
(From OE-Core rev: ccea69032329f3ba43c727d9eb71b1d063b89824)
Signed-off-by: Ming Liu <liu.ming50@gmail.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
