RUSTC_TARGET_ARCH comes from the GNU triple, so every 32-bit PowerPC
subtarget gets powerpc-unknown-linux-musl. That triple is hard-float and
mpc85xx is not: its e500v2 has no floating-point unit, and the toolchain
reports _SOFT_FLOAT and calls __divdf3 for a division.
So rustc emits lfd, stfd and fdiv for a core with no floating-point
registers, and passes doubles in a way nothing else in the image does. It
does not fail outright — a speed test reported negative throughput, and
formatting one of those values tripped an assertion inside core.
Pass -Ctarget-feature=-hard-float where CONFIG_SOFT_FLOAT says so. The same
flag has to reach the std that rust/host builds, or std and the packages
linked against it disagree about how a double is passed.
Signed-off-by: Josef Schlehofer <pepe@bloodkings.eu>
(cherry picked from commit 388c3b6ad4)
LLVM places the release fence of a cmpxchg between the load-linked and the
store-conditional. On e500v2 that clears the reservation, so the
store-conditional always fails. The strong form retries and re-reserves; the
weak form has no retry and can never succeed.
fetch_update in core is a weak retry loop, so it spins forever and anything
using it hangs. A tokio reactor livelocks on its first I/O event, burning
CPU without issuing a syscall, which makes async Rust unusable on the
subtarget.
Carry the fix as a patch until it reaches a rustc release. It adds a
TargetLowering hook, defaulting to false, so only PowerPC changes.
Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>
(cherry picked from commit 9ea0d00d93)
Also refresh 020-riscv64-support.patch from the current state of
openresty/luajit2#236. The previous snapshot defined
asm_tail_prep(ASMState *as), but since openresty/luajit2 commit
68354f44 the callers pass a second TraceNo argument, so the riscv64
build failed with:
lj_asm.c:2559:5: error: too many arguments to function 'asm_tail_prep'
The refreshed patch carries the upstream adaptation and restores the
build. The other four patches apply unchanged.
Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>
(cherry picked from commit 578758e1f6)
luajit2 currently fails to link for powerpc: lj_vm.o emits PLT calls
that force BSS-PLT, which the inline-PLT relocations emitted by GCC 12
and newer do not support. Add the same GOT-based fix that
lang/lua/luajit has carried as 060-ppc-musl.patch since 2019.
With that in place the package builds, but ipairs() and pairs() turn
out to be broken on 32-bit big-endian soft-float targets. Both are
regressions in luajit2's PPC64 patch and do not affect upstream
LuaJIT. To be submitted to openresty/luajit2 as well.
Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>
(cherry picked from commit a6b248d6b8)
old just disabled version test, but as LuaJIt 2.1. part is stable.
So we can use that for test if luajit2 itself able to run.
Fixes: c7ca6d46f7 luajit2: ("add test-version.sh")
Suggested-by: Josef Schlehofer <pepe.schlehofer@gmail.com>
Signed-off-by: Seo Suchan <tjtncks@gmail.com>
(cherry picked from commit 5504291b81)
Add PYPI_SOURCE_NAME:=drf_nested_routers as the 0.95.0 sdist
uses underscores in the filename.
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit 0f833b1b9c)
Update package to 3.17.1.
Bug fix:
- HTMLFormRenderer crashed with empty datetime field values
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit b5a553b457)
Major changes since 3.10.0:
4.0:
- add CORS_ALLOW_PRIVATE_NETWORK setting for Local Network Access spec support
- add async middleware support, reducing overhead on async views
- remove deprecated CORS_REPLACE_HTTPS_REFERER and CorsPostCsrfMiddleware
- remove three header names from default allowed list (accept-encoding, dnt, origin)
4.3:
- fix access-control-allow-credentials header not being omitted on
non-allowed responses
4.6:
- drop support for Django 3.2 through 4.1
4.9:
- add Django 6.0 support
- tighten validation of sequence-type settings (e.g. CORS_ALLOW_METHODS)
Add PYPI_SOURCE_NAME:=django_cors_headers as the 4.x sdist uses
underscores in the filename.
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit ccd69a148f)
Update package to 6.0.4.
Security fixes:
- CVE-2026-33033: DoS fix in MultiPartParser -- base64-encoded multipart
uploads with excessive whitespace could cause repeated memory copying
- CVE-2026-3902: ASGI header spoofing fixed -- headers containing underscores
are now ignored by ASGIRequest to prevent hyphen/underscore conflation
attacks
- CVE-2026-4277: Privilege abuse in GenericInlineModelAdmin -- add permissions
on inline model instances were not validated against forged POST data
- CVE-2026-4292: Privilege abuse in ModelAdmin.list_editable -- changelist
forms incorrectly allowed new instances to be created via forged POST data
- CVE-2026-33034: DoS via ASGI memory upload limit bypass -- missing or
understated Content-Length could bypass DATA_UPLOAD_MAX_MEMORY_SIZE
Bug fixes:
- alogin/alogout regression where request.user was not set/cleared if already
materialized by sync middleware
- RelatedFieldWidgetWrapper regression incorrectly wrapping all widgets in a
fieldset in admin forms
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit 06eb22a606)
Changes since 0.4.4:
- v0.5.0: Improved handling of CTEs and window functions
- v0.5.0: Better token classification for SQL dialects
- v0.5.x: Various bug fixes for edge cases in formatting
and statement splitting
Also add test.sh to verify formatting, splitting, and parsing.
Link: https://github.com/andialbrecht/sqlparse/blob/master/CHANGELOG
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit 2eef8732a3)
Verify that asgiref.__version__ strictly equals the expected version
argument ($2), instead of just checking that it is non-empty.
Also remove single quotes from EOF to enable variable expansion.
Signed-off-by: Wei-Ting Yang <williamatcg@gmail.com>
(cherry picked from commit bdab2f1f59)
Changelog: https://github.com/django/asgiref/blob/main/CHANGELOG.txt
Multiple minor/patch releases since 3.7.2 with bug fixes and
Python 3.13+ compatibility improvements. Resets PKG_RELEASE to 1.
Add test.sh to verify async_to_sync and sync_to_async adapters.
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit 5a7fbc8303)
Changelog since 0.9.6.1:
- add product version number to the PRODID tag in iCalendar output
- add support for GEO tags in vCards
- various bugfixes and compatibility improvements
Also need to update setup.cfg at build time to insert the version
to allow the package to build.
Otherwise we get
```
2026-04-21T08:17:17.9341927Z File "/builder/staging_dir/hostpkg/lib/python3.14/site-packages/setuptools/config/setupcfg.py", line 296, in __setitem__
2026-04-21T08:17:17.9342520Z parsed = self.parsers.get(option_name, lambda x: x)(value)
2026-04-21T08:17:17.9343117Z File "/builder/staging_dir/hostpkg/lib/python3.14/site-packages/setuptools/config/setupcfg.py", line 602, in _parse_version
2026-04-21T08:17:17.9343777Z return expand.version(self._parse_attr(value, self.package_dir, self.root_dir))
2026-04-21T08:17:17.9344201Z ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-04-21T08:17:17.9344938Z File "/builder/staging_dir/hostpkg/lib/python3.14/site-packages/setuptools/config/setupcfg.py", line 421, in _parse_attr
2026-04-21T08:17:17.9345544Z return expand.read_attr(attr_desc, package_dir, root_dir)
2026-04-21T08:17:17.9345876Z ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-04-21T08:17:17.9346409Z File "/builder/staging_dir/hostpkg/lib/python3.14/site-packages/setuptools/config/expand.py", line 191, in read_attr
2026-04-21T08:17:17.9346947Z return getattr(module, attr_name)
2026-04-21T08:17:17.9347261Z AttributeError: module 'vobject' has no attribute 'VERSION'
2026-04-21T08:17:17.9598878Z
```
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit 970bbeb3b5)
Don't override the upstream build ID logic to hopefully improve
reproducibility.
Signed-off-by: George Sapkin <george@sapk.in>
(cherry picked from commit 6149f90711)
On macOS (Darwin) hosts, building host Python C extensions (such as Cython) using the '-shared' flag and linking against '-lpython3.x' causes the host Python interpreter to load a duplicate copy of the Python runtime. This leads to type checking mismatches and segmentation faults (SIGSEGV) when importing the compiled extension.
For example, running:
./staging_dir/hostpkg/bin/python3 -c "import Cython.Utils"
crashes with:
Segmentation fault: 11
To build shared modules correctly on macOS, they must be compiled as bundles using the '-bundle -undefined dynamic_lookup' flags instead of '-shared', and they should not link against the Python library (no '-lpython3.x' in LDFLAGS).
Fix this by dynamically adjusting LDSHARED and LDFLAGS in python3-host.mk when the host OS is Darwin.
Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>
(cherry picked from commit ea541f48b5)
rust/host fails to compile on macOS running on Intel x64
because the host target triple is autogenerated to be
'arm64-unknown-linux-'. Rust doesn't have such a target triple, thus the
build fails because there are no pre-built artefacts for bootstrapping.
Fix this by setting RUSTC_HOST_ARCH to 'x86_64-apple-darwin' in case
our host is HOST_ARCH=x86_64 and HOST_OS=Darwin.
This fix is based on the existing fix for Apple silicon [1].
Fixes:
File "/Volumes/pepe/py/openwrt/build_dir/target-x86_64_musl/host/rustc-1.96.0-src/x.py", line 53, in <module>
bootstrap.main()
~~~~~~~~~~~~~~^^
File "/Volumes/pepe/py/openwrt/build_dir/target-x86_64_musl/host/rustc-1.96.0-src/src/bootstrap/bootstrap.py", line 1418, in main
bootstrap(args)
~~~~~~~~~^^^^^^
File "/Volumes/pepe/py/openwrt/build_dir/target-x86_64_musl/host/rustc-1.96.0-src/src/bootstrap/bootstrap.py", line 1366, in bootstrap
build.download_toolchain()
~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/Volumes/pepe/py/openwrt/build_dir/target-x86_64_musl/host/rustc-1.96.0-src/src/bootstrap/bootstrap.py", line 697, in download_toolchain
download_component(download_info)
~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^
File "/Volumes/pepe/py/openwrt/build_dir/target-x86_64_musl/host/rustc-1.96.0-src/src/bootstrap/bootstrap.py", line 529, in download_component
get(
~~~^
download_info.base_download_url,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...<3 lines>...
verbose=download_info.verbose,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/Volumes/pepe/py/openwrt/build_dir/target-x86_64_musl/host/rustc-1.96.0-src/src/bootstrap/bootstrap.py", line 58, in get
raise RuntimeError(
...<6 lines>...
)
RuntimeError: src/stage0 doesn't contain a checksum for dist/2026-04-16/rust-std-1.95.0-x86_64-unknown-linux-darwin24.6.0.tar.xz. Pre-built artifacts might not be available for this target at this time, see https://doc.rust-lang.org/nightly/rustc/platform-support.html for more information.
[1] https://github.com/openwrt/packages/commit/105fa3920e12f557bdf1fcbc566fc286fb53e319
Signed-off-by: Georgi Valkov <gvalkov@gmail.com>
(cherry picked from commit aa32dd256e)
Signed-off-by: Georgi Valkov <gvalkov@gmail.com>
Install libperl.so with executable permissions (0755) instead of copying
it as-is.
This allows the OpenWrt build system to recognize the library as a binary and
automatically run the "strip" utility on it. Stripping removes unnecessary
debugging metadata and helper symbols that are only needed for development but
not for running Perl on the router.
This change reduces the installed size of libperl.so on the device:
- Before (not stripped): 3.7 MB (3,929,652 bytes)
```
root@turris:~# file /usr/lib/perl5/5.40/CORE/libperl.so
/usr/lib/perl5/5.40/CORE/libperl.so: ELF 32-bit MSB shared object, PowerPC or cisco 4500, version 1 (SYSV), dynamically linked, with debug_info, not stripped
```
- After (stripped): 3.5 MB (3,674,081 bytes)
```
root@turris:~# file /usr/lib/perl5/5.40/CORE/libperl.so
/usr/lib/perl5/5.40/CORE/libperl.so: ELF 32-bit MSB shared object, PowerPC or cisco 4500, version 1 (SYSV), dynamically linked, no section header
```
Saving ~255 KB of flash storage space on target devices.
Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>
(cherry picked from commit b6d2070fc8)
Removing the ncursesw path from include avoids seeing the wide-char
version of ncurses, so the host build picks up the narrow static lib
that the OpenWrt SDK stages. To keep that pin from leaking back in
through other curses headers the build host may have under
/usr/include, also force every header check in Python's configure.ac
AC_CHECK_HEADERS(curses.h ncurses.h ncursesw/ncurses.h
ncursesw/curses.h ncursesw/panel.h panel.h) line to "no" via cached
ac_cv_header_* variables.
The narrow ncursesw_* trio was enough on the original reporter's
host; Jan Kardell's review of the first cut hit a build where
`/usr/include/curses.h` and `/usr/include/ncurses.h` from the
distro's libncurses-dev were still being picked up. Adding the
non-wide entries (and panel.h for symmetry, in case a host also
ships /usr/include/panel.h alone) makes the override insensitive to
whichever curses dev packages the host distro happens to provide.
Reported-by: Jan Kardell <jan.kardell@telliq.com>
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit d883c02a41)
Add a simple CI compilation test to ensure that the Go toolchain is
functional.
Signed-off-by: George Sapkin <george@sapk.in>
(cherry picked from commit d9fdda8cf2)
On some build systems, the miniperl binary is created without execute
permission (errno 126 when running it as /bin/sh). This breaks building
the mro extension and cascades to all dependent packages.
Fix by ensuring chmod +x on miniperl after the main build step.
This matches how many other build systems handle this same issue.
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit d4d4e3e419)
perlbase-archive, perlbase-pod, and perlbase-test install Perl script
wrappers (ptar, pod2man, prove, etc.) that do not output the OpenWrt
package version string (5.40.0), causing generic version check failures
in CI.
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit 024b8ee60e)
Flup was heavily used in downstream distribution (Turris OS)
for their Web UI - reForis. Since there are no other
dependent packages in this repository, Flup is no longer needed.
The package appears to be abandoned and is no longer maintained
The latest version dates back to 2009.
It was previously required for Seafile.
Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>
(cherry picked from commit 5120ed91b8)
Bump to version 0.1.3 and drop patch, which is in upstream.
Signed-off-by: Erik Larsson <who+github@cnackers.org>
(cherry picked from commit fee669b6d4)
The release tarball already contains a pre-generated _version.py with
the version hardcoded, so versioneer is not needed at build time.
Patch pyproject.toml to only require setuptools and remove
PKG_BUILD_DEPENDS:=python-versioneer/host.
Remove python-versioneer package as it is no longer needed.
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit 8c3bc3fcf9)
Update package to 3.4.0.
Changes:
- Added binary wheels for RISC-V 64
- Fixed multiple rare crash paths during interpreter shutdown; now uses
the atexit module (subtle API change: getcurrent unavailable once
atexit fires)
- Fixed multiple race conditions in free-threaded (no-GIL) builds when
greenlets are accessed from multiple threads, some causing assertion
failures or interpreter crashes
- Several minor correctness fixes from automated code audit
Signed-off-by: Alexandru Ardelean <alex@shruggie.ro>
(cherry picked from commit 15e2da047b)
George Sapkin
Josef Schlehofer
Josef Schlehofer
Seo Suchan
Javier Marcet
hingbong
Hirokazu MORIKAWA
Andy Chiang
Alexandru Ardelean
Wei-Ting Yang
Georgi Valkov
Michael Heimpold
Jens Wagner
Erik Larsson
Rosen Penev