Under certain circumstances udev appears to be responsible (or part of)
a race condition, where input devices are treated and behave differently
than expected.
An attempt of documenting this issue:
https://bugreports.qt.io/browse/QTBUG-86630
They shouldn't be necessary, as nothing will or at least should link
against the resulting shared objects, as they aren't regular ones but
plug-ins.
However cmake-based projects complain about them missing - hence
provide/install them.
qmlplugindump appears to be called at the end of `make`-invocation for
qmake generated projects.
If it fails with exit code != 0 the OpenWrt build fails.
qmlplugindump is apparently needed as a host tool, but can not be
(easily) packaged as a host tool.
The project file states:
"qmlplugindump cannot be a build tool, because it loads target plugins."
which seems pretty mysterious to me.
So it's needed as a requirement on the host system, and/but natively
installed (newer) versions - at with my Debian's version - started to
fail with:
"QQmlComponent: Component is not ready"
I have no idea what qmlplugindump actually does or why it's needed.
It seems to do no harm to not invoke qmlplugindump - so
hack-not-even-fix it by just overriding each qmlplugindump invocation.
Without that patch the build rule including the input component depends
on other options being enabled, such as e.g. GUI support.
Since our host build is headless and we only need the input component
which actually does not have any such dependencies, we patch to force-
build it.
input component being present is a requirement for e.g. qt5tools
building certain host tools - such as 'kmap2qmap'.
Main objective is to make sure every Qt5 related file is clearly about
Qt version *5*. For files which do not contain the version number, make
sure the path contains it (e.g. suffixing the directory paths with
"/qt5"). This is done for host and target builds and accounts for the
following paths:
/bin -> /bin/qt5
/share -> /share/qt5
Every file inside /lib (except /lib/qt5/* which is covered by the
archdata-directory) is already versioned by its respective file names.
Keep target specific host builds inside $(STAGING_DIR)/host.
mesa-clc set to 'system' requires the system tool 'mesa-clc' which gets
built as part of the mesa/host build.
The mesa/host build however depends on LLVM and others - hence we're
trying to avoid pulling this in where possible if target configuration
allows.
meson option 'mesa-clc' allows for 3 options: enabled, system, auto.
While 'auto' sounds like it will only automagically go for either
'enabled' or 'system', it will also disable CLC support entirely, if
only drivers are enabled which don't need it (resulting in
`with_clc=false`) - which is exactly what we want.
This relies on only drivers being selectable, which don't automatically
enable CLC - if LLVM support is disabled.
Fix wrong PKG_HASH this has probably changed after xz vulnerability
discovery. Update to new PKG_HASH version.
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
* Use JavaScript JIT only on supported platforms.
* Fix ARM Thumb2 detection to allow using JavaScript JIT on ARMv7.
* Add inline assembly implementation returning the current stack pointer
on PPC64 (or PPC).
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Clover only implements OpenCL API for nouveau, r600 and radeon-si GPUs.
Remove OpenCL library packages for targets which anyway implement 0% of
the API and are hence useless.
See also https://mesamatrix.net
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Only LP64/LLP64 builds are officially supported. Please do not attempt
to build for other platforms; even if the program succeeds in a MAP01
smoke test, there are e.g. known visual artifacts that lead to a bad
user experience.
Still allow build on 32-bit platforms in case BROKEN is selected, so
users can try if it would work on their system as forum claims a fix
had actually been applied long ago.
Link: https://forum.zdoom.org/viewtopic.php?f=7&t=75673
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Define all CONFIG_PACKAGE_* symbols which may change build configuration
and only build broadcom drivers on ARM targets with FPU.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Previous attempts to prevent linking against libunwind were incomplete.
Simply don't call CheckLibUnwind macro.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Make sure build dependency to mesa/host is set for x86 and x86/64 builds,
add missing dependency on Python ply and make sure to always prefer LLVM
subproject to avoid picking up LLVM from the build host (which may not
match SPIR-V converter tool version).
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Install host-built binaries to STAGING_DIR_HOSTPKG as HOST_BUILD_DIR gets
cleared and should not be referenced after host build has completed.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Mirko Vogt