(for clang-scan-deps which is needed for C++20 modules but not included
in base), we don't want to use it to compile.
this still adds as an unwanted BDEP on ports-gcc archs, but at least it
doesn't then get in the way of build. (we do want to use the module rather
than add a BDEP on a hardcoded lang/llvm/19).
found by jca@
Major update from 3.31.8 to 4.1.2.
CMake 4.0 introduces a breaking change: compatibility with versions older than
3.5 has been removed. Projects calling cmake_minimum_required() or
cmake_policy() with versions < 3.5 now error out.
To maintain compatibility with existing ports during the transition, the cmake
module now provides MODCMAKE_POLICY_VERSION_OVERRIDE (idea from sthen@)
support:
- MODCMAKE_POLICY_VERSION_OVERRIDE ?= No
- MODCMAKE_POLICY_VERSION_OVERRIDE_VER ?= 3.5
When MODCMAKE_POLICY_VERSION_OVERRIDE is set to "yes", CMake is instructed to
accept policy versions down to the specified minimum (default 3.5) via
-DCMAKE_POLICY_VERSION_MINIMUM. This allows older ports to build while
acknowledging they may use deprecated features.
https://cmake.org/cmake/help/latest/release/4.0.htmlhttps://cmake.org/cmake/help/latest/release/4.1.html
OK sthen@
It is suceffuly compiled by tb@ on sparc64, but it has some warnings on
alignment. I've checked a few code paths which leads to this warnings,
and all of them looks false positive.
tested and OK: tb@
While it's not 100% clear that this is needed, emulators/citra broke due
to the addition of noexcept to numerous callbacks, which changes a few
function signatures at least on the API level. It was a huge update, so
better be safe than sorry.
discussed with sthen
- Enable the SSE support on X86; there is run-time detection
- Disable NEON on arm
- The ALSOFT_CONFIG option has been deprecated
- The QSA backend was removed with the 1.21.0 release
- Sort the list of options
backend changes that were introduced in the 1.21.1 update.
Simplify channel handling in the sndio backend
620836f173ae6fc4505d0634984e0f2c46166367
Use non-block mode for sndio capture
1fd4c865fc084f134363db5155361d5483679235
From Brad, tested by me
ok maintainer (David CARLIER)
This was tracked down to two commits upstream which broke the sndio
backend due to >2 channel support being added.
Until this can be sorted out, this diff reverts two upstream commits:
5cffe7e50a2885d9b392c15f2cd3941cac49692c
b4a52321c4ad8e8bc0eb29d2cdae2c043fc405e1
From Brad, tested by me and kmos@
ok kmos@ sthen@
Backport a fix from upstream to avoid the crash. The code has changed
since and all backends have been converted to the "new style" backend
API. Fix tested by David Carlier (maintainer) and Thomas Frohwein.
clang fails to build those with 'could not read symbols: Bad value' (cf
http://build-failures.rhaalovely.net/sparc64/2017-03-20/multimedia/gstreamer1/core.log
& ttp://build-failures.rhaalovely.net/sparc64/2017-03-20/audio/openal.log)
Fix only those as they take out a large part of the portstree for 6.1
package builds, but reasonably MODCLANG_ARCHS=* shouldnt be used unless
the compiler has been *really* tested on every platform - or clang
should be marked *BROKEN* on sparc64 & macppc, or those platforms should
be taken out of LLVM_ARCHS.
audio/mpd, textproc/link-grammar, devel/keystone and devel/cbmc are in
the same boat and fail the same.
The problem is probably fixable directly in clang for sparc64, and on
powerpc clang doesnt package anyway for 'relocation truncated to fit:
R_PPC_PLTREL24' errors:
http://build-failures.rhaalovely.net/powerpc/2017-02-16/devel/llvm.log
tb