Since llama.cpp update in Buildroot commit [1], the test_aichat can
fail for several reasons:
The loop checking for the llama-server availability can fail if curl
succeed, but the returned json data is not formatted as expected.
This can happen if the server is ready but the model is not completely
loaded. In that case, the server returns:
{"error":{"message":"Loading model","type":"unavailable_error","code":503}}
This commit ignore Python KeyError exceptions while doing the
server test, to avoid failing if this message is received.
Also, this new llama-server version introduced a prompt caching, which
uses too much memory. This commit completely disable this prompt
caching by adding "--cache-ram 0" in the llama-server options.
[1] https://gitlab.com/buildroot.org/buildroot/-/commit/05c36d5d875713521f99b7bad48be316dcde2510
Signed-off-by: Julien Olivain <ju.o@free.fr>
Release notes: https://github.com/ggml-org/llama.cpp/releases
Merge BR2_PACKAGE_LLAMA_CPP_SERVER into BR2_PACKAGE_LLAMA_CPP_TOOLS, as
both of these options must be enabled to build tools like llama-cli and
llama-server. See upstream commit [1].
Since the Buildroot option BR2_PACKAGE_LLAMA_CPP_SERVER is removed, this
commit also removes it from support/testing/tests/package/test_aichat.py
which was using it.
[1] https://github.com/ggml-org/llama.cpp/commit/a180ba78c710f52af7c7edee562fbe98c36d6ddd
Signed-off-by: Joseph Kogut <joseph.kogut@gmail.com>
[Julien:
- reindent options in .mk
- remove BR2_PACKAGE_LLAMA_CPP_SERVER in test_aichat.py
]
Signed-off-by: Julien Olivain <ju.o@free.fr>
Python setuptools has removed the obsolete pkg_resources python module in
v82.0.0. This module is used by the binman command in U-Boot until
v2025.10.
Since the python-setuptools package has been updated to v82.0.0 in
Buildroot, the python test tests.boot.test_atf.TestATFAllwinner (using
U-Boot v2023.10 and binman) fails to build with the following error [1]:
ModuleNotFoundError: No module named 'pkg_resources'
Update test_atf to use a more recent U-Boot v2026.01 (and add the
dependency on GNU TLS), to fix the build.
Link: https://gitlab.com/buildroot.org/buildroot/-/jobs/13500946337 [1]
Fixes: 51365ff063 ("package/python-setuptools: bump to version 82.0.0")
Signed-off-by: Vincent Stehlé <vincent.stehle@arm.com>
Cc: James Hilliard <james.hilliard1@gmail.com>
Cc: Julien Olivain <ju.o@free.fr>
Signed-off-by: Julien Olivain <ju.o@free.fr>
Specifying a list of br2-external trees is poorly documented, and the
only example uses a colon to separate the br2-external paths.
Adding the support for colon-separated list is the biggest mistake that
was made when introducing support for multiple br2-external [0]. Indeed,
both space and colon can be used to separate entries in the list, and it
is also possible to mix the two. However, internally, the list is stored
as a space-separated list, and all the code will split on spaces.
Besides, all other lists in Buildroot are a space-separated:
BR2_ROOTFS_DEVICE_TABLE
BR2_ROOTFS_STATIC_DEVICE_TABLE
BR2_TARGET_TZ_ZONELIST
BR2_ROOTFS_USERS_TABLES
BR2_ROOTFS_OVERLAY
BR2_ROOTFS_PRE_BUILD_SCRIPT
BR2_ROOTFS_POST_BUILD_SCRIPT
BR2_ROOTFS_POST_FAKEROOT_SCRIPT
BR2_ROOTFS_POST_IMAGE_SCRIPT
...
So, using colons is odd.
The fact that BR2_EXTERNAL is passed on the command line rather than
being a Kconfig item is not a reason enough to justify that it be
colon-separated.
Change the documentation to only mention using a space-separated list.
Of course, for backward compatibility, we keep the code as-is to accept
a colon-separated list, but we just do not advertise it.
Note that keeping the split on colons means that colons are not accepted
in pathnames of br2-external trees; in practice, this is not a new
restriction, or one that could lift as usign colons in Makefiles are
problematic anyway.
[0] in 20cd497387 core: add support for multiple br2-external trees
Reported-by: Fiona Klute (WIWA) <fiona.klute@gmx.de>
Reported-by: Brandon Maier <Brandon.Maier@collins.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Romain Naour <romain.naour@smile.fr>
Boring changes: either do what shellcheck suggested, or comment why we
don't want to fix the code.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Romain Naour <romain.naour@smile.fr>
The trap was initially introduced in c5fa9308ea (core/br2-external:
properly report unexpected errors), in 2017, to catch all unexpected
errors, back when a single file was generated, and errors emitted to
stderr.
Since commit d027cd75d0 (core: generate all br2-external files in
one go), in 2019 the single output file 'ofile' is no longer created,
as multiple output files were then introduced, while messages for
*expected errors* were redirected to a Makefile variable assignment
emitted on stdout, at which point the script just exits (in error);
expected failures only occur in do_validate().
Unexpected errors can only occur on failure to create, or write to,
output files, either '.br2-external.mk' in do_validate() or do_mk(),
or any of the kconfig fragments in do_kconfig(). Cause for failure to
create those can only be a no-space-left-on-device condition, as they
are created in a directory that was just created by the script earlier
in main(), and thus has the necessary mode; failure to create that
directory is now caught explicitly.
A trap on ERR is not called when the shell exits explicitly with a call
to 'exit', thus, only failures to create or write to output file would
be caught. In that case, we are better off not trying to write to those
files anyway: failure to create the file would already be reported by
the shell on stderr, while disk-full would not allow to store the output
anyway...
In any case, the script exits in error, which is going to be caught by
the caller, which will terminate.
So, drop the trap altogether.
As a side effect, that squelches a shellcheck error.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Romain Naour <romain.naour@smile.fr>
Subproject tarballs are not provided anymore:
https://discourse.llvm.org/t/llvm-22-1-0-released/89950
"Please note since the last release the subproject tarballs have been
removed"
https://discourse.llvm.org/t/rfc-do-something-with-the-subproject-tarballs-in-the-release-page/75024/14
Used upstream tarball llvm-project-22.1.0.src.tar.xz for all packages
and linked subproject hash files to ../llvm-project.hash.
Removed patches which fix build errors caused by subproject tarballs.
Removed handling of third-party-21.1.8.src.tar.xz which is included in
the monolithic tarball.
Added _SUBDIR variable or updated _INSTALL_CMDS when needed.
For compiler-rt:
Removed both patches, they are not needed anymore.
Added patch to fix aarch64 build.
Added dependency on gcc >= 15.x because libcxx now depends on gcc >=
15.x: https://github.com/llvm/llvm-project/pull/165684
warning "Libc++ only supports GCC 15 and later"
Building with gcc 14.x causes many build errors like
output/build/compiler-rt-22.1.0/compiler-rt/buildroot-build/lib/fuzzer/libcxx_fuzzer_x86_64/build/include/c++/v1/__type_traits/is_array.h:43:68:
error: expected primary-expression before ')' token
output/build/compiler-rt-22.1.0/compiler-rt/buildroot-build/lib/fuzzer/libcxx_fuzzer_x86_64/build/include/c++/v1/__type_traits/is_array.h:43:44:
error: there are no arguments to '__is_unbounded_array' that depend on
a template parameter, so a declaration of '__is_unbounded_array' must
be available [-fpermissive]
output/build/compiler-rt-22.1.0/compiler-rt/buildroot-build/lib/fuzzer/libcxx_fuzzer_x86_64/build/include/c++/v1/__type_traits/decay.h:22:32:
error: expected type-specifier before '__decay'
A corresponding bug report sent upstream
https://github.com/llvm/llvm-project/issues/174203
was answered:
https://github.com/llvm/llvm-project/issues/174203#issuecomment-3711113919
"Our policy is rather clear: Only the latest GCC is supported."
and an update to supported compiler versions was committed:
https://github.com/llvm/llvm-project/commit/d1146b1ddd03aea3d67ce8f413607e8e8be67f4b
Updated TestClangCompilerRT to use a gcc 15-based toolchain.
Signed-off-by: Bernd Kuhls <bernd@kuhls.net>
Signed-off-by: Julien Olivain <ju.o@free.fr>
The artifacts URL has been moved to https and domain to .org to
improve security and avoid redirection issues. This ensures that
downloads of kernels and rootfs images during runtime tests are
encrypted and verified.
The change has been tested by running a runtime test, confirming the
correct download and renaming of artifacts from the new URL:
Downloading to .../tmpyotq8uor
Renaming from .../tmpyotq8uor to .../kernel-versatile-5.10.202
Signed-off-by: Dowan Gullient <dowan.gullient@smile.fr>
Signed-off-by: Julien Olivain <ju.o@free.fr>
Commit 3d2141bcee("support/testing/run-tests: specify multiprocessing
method") added a call to multiprocessing.set_start_method('fork') as a
workaround for python 3.14, which changed the default start method to
forkserver - Which is incompatible with the nose2 setup.
multiprocessing.set_start_method() is only supposed to be called a maximum
of 1 time per process and throws a RuntimeError if called more than that
(even with the same arguments):
>>> import multiprocessing
>>> multiprocessing.set_start_method('fork')
>>> multiprocessing.set_start_method('fork')
Traceback (most recent call last):
File "<python-input-2>", line 1, in <module>
multiprocessing.set_start_method('fork')
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^
File "/usr/lib/python3.13/multiprocessing/context.py", line 247, in set_start_method
raise RuntimeError('context has already been set')
Debian included a similar patch in python3-nose2 0.51.1-2 (currently in
testing/unstable) which adds its own call to set_start_method():
https://salsa.debian.org/python-team/packages/nose2/-/blob/debian/0.15.1-2/debian/patches/0004-plugins-mp-set-context-to-fork-for-Python-3.14-mp-AP.patch?ref_type=tags
Which comes from:
https://github.com/nose-devs/nose2/pull/644
As discussed in the upstream PR, this is not a correct fix is wrong and
breaks various use cases. An issue has been opened to get this fixed in the
Debian packaging at:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1129350
But until that is done, rework the patch to:
- Only override set_start_method() if needed to limit impact
- Monkey patch set_start_method() so additional calls are ignored
To unbreak run-test on affected Debian systems and add some documentation to
make it clear why this is done.
[Peter: use allow_none / force optional arguments as pointed out by Julien]
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Removed patch that fixed building with libseccomp and old kernel headers
as systemd has fixed that upstream now.
Notable changes:
- Add -Dlibmount=enabled to host and target build. Disabling this
option disables building most of systemd's tools apart from
libsystemd.
- remove gcrypt and gnutls from dependencies and build options.
openssl is the only supported crypto library now.
- Disable nspawn build for host variant and add a Config.in option
for the target variant.
- systemd now requires kernel >= 5.4 while the recommended kernel
version moved to >=5.7. This make systemd depend on
BR2_TOOLCHAIN_HEADERS_AT_LEAST_5_4, removing various dependencies
on older kernel header versions from sub options.
- The sha256sum of LICENSES/README.md changed due to various
license clarification, removal of some vendored files as well as
changing some file paths. Licensing did not actually change.
For changelog, see:
https://github.com/systemd/systemd/blob/v258.3/NEWS
Tested with `./support/testing/run-tests tests.init.test_systemd`
All tests passed
One of the tests now needs a bigger rootfs size.
Signed-off-by: Marcus Hoffmann <buildroot@bubu1.eu>
[Peter: fix check-package warnings]
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The CPIO filesystem generated by the test_firewalld test is too
large, and doesn't fit as an initramfs in the 256MB of RAM available
in the versatilepb machine. This causes a "Initramfs unpacking failed:
write error" when booting, and many files being missing from the root
filesystem, ultimately causing the test to fail.
The test_firewalld test initially started to fail following a systemd
update [1][3]:
[BRTEST# systemctl is-active firewalld
failed
But really started to crash at boot following a python 3.14 update
[2][4]:
Run /init as init process
/init: exec: line 15: /sbin/init: not found
Also, update TestFirewalldSysVInit to use ext2 instead of cpio.
[1] 926e0504d0
[2] a0a6abc8b1
Fixes:
[3] https://gitlab.com/buildroot.org/buildroot/-/jobs/12944797059
[4] https://gitlab.com/buildroot.org/buildroot/-/jobs/11856840940
Signed-off-by: Romain Naour <romain.naour@smile.fr>
Signed-off-by: Julien Olivain <ju.o@free.fr>
Buildroot commit [1] introduced a new Buildroot-Initialize.cmake file,
to mimic the upstream Linux-Initialize.cmake.
However, this upstream file was introduced in commit [2], which was
introduced in CMake 3.27.
When compiling a cmake package on a host with a cmake older than
version 3.27 (for example, the Buildroot Docker reference image has
cmake 3.25), the configuration fails with the error:
CMake Error at /buildroot/output/host/share/buildroot/Platform/Buildroot-Initialize.cmake:1 (include):
include could not find requested file:
Platform/Linux-Initialize
Call Stack (most recent call first):
/usr/share/cmake-3.25/Modules/CMakeSystemSpecificInitialize.cmake:21 (include)
CMakeLists.txt:20 (project)
CMake version < 3.27 is setting its LINUX predefine elsewhere (see
commit log of [1]), so this commit fixes the issue by simply making
the include optional. This will guarante thee cmake predefines will
be present in all the cases.
Fixes:
[1]
[1] https://gitlab.com/buildroot.org/buildroot/-/commit/ef9f0a07ed40970a7de4c608a5708ef6ceb75131
[2] https://gitlab.kitware.com/cmake/cmake/-/commit/cc737ae8295156812e697a9d2f3f7b79bf7b72c8
Signed-off-by: Julien Olivain <ju.o@free.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
This commit adds the -N/--needs-update option, disabled by default,
to list only packages with newer upstream versions. All other packages
will be excluded from the HTML or JSON output.
Signed-off-by: Kadambini Nema <kadambini.nema@gmail.com>
Signed-off-by: Arnout Vandecappelle <arnout@rnout.be>
Currently, the relocate-sdk.sh script scans the whole extracted SDK tree
to find instances of paths it needs to replace, which can take a
significant amount of time when the SDK is large, particularly relative
to the number of files that actually need to change.
However, the resulting list only depends on the SDK tarball itself, so
we can calculate it at build time and ship it with the tarball so
relocate-sdk.sh can use it directly.
Testing this on my machine with somewhat IOPS-limited rotating media,
the time goes down from:
$ time ./relocate-sdk.sh
Relocating the buildroot SDK from [...] to [...] ...
./relocate-sdk.sh 5.19s user 26.21s system 9% cpu 5:34.40 total
To:
$ time ./relocate-sdk.sh
Relocating the buildroot SDK from [...] to [...] ...
./relocate-sdk.sh 0.49s user 0.29s system 103% cpu 0.749 total
Signed-off-by: Florian Larysch <fl@n621.de>
Signed-off-by: Arnout Vandecappelle <arnout@rnout.be>
When running "make pkg-stats" on a host with Python 3.14 (e.g.
Fedora 43 for example), the execution fails with the error:
Checking URL status
Traceback (most recent call last):
File "/buildroot/support/scripts/pkg-stats", line 1387, in <module>
__main__()
~~~~~~~~^^
File "/buildroot/support/scripts/pkg-stats", line 1368, in __main__
loop = asyncio.get_event_loop()
File "/usr/lib64/python3.14/asyncio/events.py", line 715, in get_event_loop
raise RuntimeError('There is no current event loop in thread %r.'
% threading.current_thread().name)
RuntimeError: There is no current event loop in thread 'MainThread'.
This is due to a breaking change introduced in Python 3.14
asyncio.get_event_loop(). See [1]. Before Python 3.14, this call was
creating and setting an event loop if there was none. This situation
is now a runtime error.
In order to fix this issue with newer Python version, while keeping
backward compatibility, this commit replaces the code:
loop = asyncio.get_event_loop()
by an explicit event loop creation:
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
This commit was tested on a Fedora 43 host with Python-3.14.2, and
with the Buildroot Docker image plus the python3-aiohttp package
which is a Debian 12 with Python-3.11.2.
[1] https://docs.python.org/3.14/library/asyncio-eventloop.html#asyncio.get_event_loop
Signed-off-by: Julien Olivain <ju.o@free.fr>
Tested-by: Vincent Fazio <vfazio@xes-inc.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Marcus Hoffmann <buildroot@bubu1.eu>
[Peter: Fix flake8 warning, use http.server instead of relying on
connectivity]
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Testing with a really old toolchain is helpful to catch issues related
to kernel headers version issues, gcc version issues, etc. We chose a
glibc toolchain though as old musl or uClibc-ng versions tend to lack
a number of features that are needed by modern software.
This toolchain is placed near the top of toolchain-configs.csv, so
that it is used as part of the "base" set of toolchain that test-pkg
uses, even without the -a option.
test-pkg takes the 6 first toolchains of this CSV file for its base
test, and actually the comment in toolchain-configs.csv was wrong
since commit 53a8c5150e, which removed a
toolchain from the base set, but not realizing that test-pkg would
anyway continue to test the first 6 toolchains.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The prebuilt MIPS64 toolchains are very old, causing build issues (for
example recently with the systemd v258 update). Replace them both a
single toolchain configuration that uses one of the mips64el Bootlin
toolchains.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
- br-i386-pentium4-full.config as an x86 32-bit toolchain test is
already reasonably covered by bootlin-x86-i686-musl.config
- br-microblazeel-full-internal.config as a Microblaze toolchain test is
already reasonably covered by bootlin-microblazeel-uclibc.config
- br-powerpc-internal-full.config and br-powerpc-603e-basic-cpp.config
as PowerPC 32-bit toolchain tests are already reasonably covered by
bootlin-powerpc-e500mc-uclibc.config
- br-powerpc64-power7-glibc.config as a PowerPC 64-bit toolchain test
is already reasonably covered by
bootlin-powerpc64le-power8-glibc.config
- br-riscv64-full-internal.config as a RISC-V 64-bit toolchain test is
already reasonably covered by bootlin-riscv64-glibc.config and
bootlin-riscv64-musl.config
- br-s390x-z13-internal-glibc.config as a s390 toolchain test is
already reasonably covered by bootlin-s390x-z13-glibc.config
- br-xtensa-full-internal.config as an Xtensa toolchain test is
already reasonably covered by bootlin-xtensa-uclibc.config
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Building internal toolchains takes a long time, and since the
differences between the 3 internal ARM toolchains is just the libc,
and we're already testing uclibc/musl with external toolchains, it
doesn't make much sense to build 3 different ARM internal toolchains.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
We're already testing the ARC architecture with one Bootlin toolchain,
it doesn't make sense to also test with two Buildroot internal
toolchains the ARC architecture, which is not a primary architecture
for Buildroot.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The package imports itself in setup.py to get the package-name and
version number. Dping this during the buildroot build would require
buisling host-python with sqlite support, which we are currently not set
up for. It also seems wasteful for just extracting the version number
and package name, so instead we replace the import by using a hardcoded
package-name and the version number stored in buildroot.
Signed-off-by: Marcus Hoffmann <buildroot@bubu1.eu>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Python 3.14 (not yet in Buildroot) introduced colors, enabled by
default, when the output is a terminal. This behavior can make the
pexpect pattern matching more difficult in some cases. See:
https://docs.python.org/3.14/using/cmdline.html#controlling-color
This commit globally disables the Python interpreter colors in the base
runtime Python test, by setting the NO_COLOR=1 environment variable.
Signed-off-by: Julien Olivain <ju.o@free.fr>
Enhance the Xen python tests to exercise block devices: this boils down
to switching from ramdisks to disk partitions for the domains rootfs.
(Refer to the comments in the python script for block devices details.)
- Add support for PCI and Xen block to the Linux kernel configurations.
- Add a few commands to list the xvda block device for good measure.
- Generate two partitions with the rootfs in the disk images; we use the
same rootfs contents twice, once for each domain.
- Add a paravirtualized block device to the Xen dom1 configurations and
adjust both domains kernel command lines, to specify the rootfs
locations.
- Build host-qemu for Arm v7, to workaround an issue with 32b Arm and
old Qemu versions, which is what we have on CI currently.
- While at it, bump Linux kernel to 6.18.4 and U-Boot to 2026.01.
Signed-off-by: Vincent Stehlé <vincent.stehle@arm.com>
Cc: Julien Olivain <ju.o@free.fr>
Signed-off-by: Julien Olivain <ju.o@free.fr>
Commit 3dd73c3 "package/openjdk: allow compiling without X11 support",
dropped the need for X11, so drop it from TestOpenJdk.
Signed-off-by: Thomas Devoogdt <thomas@devoogdt.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Rasdaemon was added in previous patch.
Add a test_rasdaemon that allows to test it.
Test is done on x86. It first checks the daemon's version, then mounts
debugfs and finally ensures that the init script allows to
start/restart/stop the daemon.
Signed-off-by: Bastien Curutchet <bastien.curutchet@bootlin.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
The run log of this ltp-testsuite test shows:
INFO: runltp script is deprecated, try kirk
https://github.com/linux-test-project/kirk
This commit updates this test to replace this deprecated runltp
shell script with the newer kirk Python script.
The logic of this runtime test remains the same: it runs a small number
of 'read' system call tests, and checks there is no failures and at
least one test succeed.
Cc: Petr Vorel <petr.vorel@gmail.com>
Signed-off-by: Julien Olivain <ju.o@free.fr>
Acked-by: Petr Vorel <petr.vorel@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Julien Olivain