The OSVERSION for FreeBSD 10.3 is 1003000, not 1030000. The last commit
blocked FreeBSD 10.3 too. OSVERSION is also supposed to be used in
conjunction with OPSYS.
Approved by: just-fix-it
In particular, only support amd64 for FreeBSD 12+ (-CURRENT),
Nail the coffin of EOL distros: IGNORE builds before 10.3-RELEASE,
and force -mstackrealign on i386 gcc to put an end to i386
instabilities. (for some reason, FreeBSD 12-CURRENT chooses GCC rather
than base clang, and then breaks the stack alignment for SSE2 again,
causing SIGBUS[1].)
[1] http://beefy11.nyi.freebsd.org/data/head-i386-default/p432608_s312909/logs/rawtherapee-devel-5.0.log
Reported by: pkg-fallout@ [1]
Changes: <http://rawtherapee.com/blog/rawtherapee-5-released>
Due to GCC issues on i386 with stack misalignment for SSE2 instructions,
use the base system clang there, and use GCC 5+ on amd64. Since i386
has no OpenMP support, even clang 3.4 in FreeBSD 10.3 is good enough.
Note that the cache directory suffix, which used to be 4.1, is now
empty, so the cache directory is now ~user/.cache/RawTherapee.
~user/.cache/RawTherapee4.1 should be renamed before the start, or
removed.
Miscellaneous cleanups all over the place.
Prune unused dependencies via -Wl,--as-needed.
Build with as few requirements as works for 10.3-RELEASE and
11.0-RELEASE, i386 and amd64. This should use the base CC on anything
but 10.3-RELEASE amd64, where we use GCC 4.9 for OpenMP support.
Specifically, 11.0-RELEASE compilations do *not* work with GCC 4.9 and
the result fails with SIGBUS with apparent bogus SSE code generation and
misalignment.
Fix sound theme support via libcanberra (change Linux ifdefs to !Apple
ifdefs).
Apply -DPROC_TARGET_NUMBER="1" (i. e. build an executable with generic
tuning) only when building packages and unless -march is in CFLAGS.
USE_GNOME upgraded from gtkmm20 to gtkm24, and add devel/openmp as
requisite on amd64 (rather than implicit clang or gcc libs), bumping
PORTREVISION.
Silence cmake developer warnings, and be sure to export linker flags to
cmake (so it finds -lomp in its configuration phase)
[-DCMAKE_POLICY_DEFAULT_CMP0056:STRING=NEW]
Cleanups: leverage USES=localbase, and drop/avoid duplicate definitions
and dead code from the Makefile and support for unsupported older
FreeBSD releases.
Based in part on v0.3 patch of...
PR: 215675
Submitted by: jbeich@
MFH: 2017Q1
- drop -ffast-math from the compiler flags [1]
- drop libc++ hacks for FreeBSD 9.x/10.1 and use the new gcc-c++11-lib [2]
- be sure to optimize for packaging when PACKAGE_BUILDING [3]
- simplify CMake OPTION_OMP passing [2]
- bump PORTREVISION
Reported by: Ingo Weyrich [1], jbeich@ [2], DrSlony
Note: untested in poudriere since texlive-base is currently defunct.
I could reproduce the problem on 10.1-amd64 on a bare metal build
and check that Christoph's patch works, so committing it. Please report
fall-out in the PR indicated below.
+ Bump PORTREVISION to trigger rebuild.
PR: 202250
Submitted by: Christoph Moench-Tegeder
Instead of defining a variable that is almost always based on CONFIGURE_ENV,
just use CONFIGURE_ENV directly.
This also matches the behavior of other ports that do not use autotools (so
most ports can just worry about CONFIGURE_ENV). Additionally, the fact that
we do not use ?= means we do not have problems if another file in Uses/
needs to set CONFIGURE_ENV (with CMAKE_ENV, the order of the arguments to
USES would matter).
Ports which set CMAKE_ENV have been adjusted accordingly. In most cases,
CMAKE_ENV was just replaced with CONFIGURE_ENV, the exceptions being:
* databases/sqliteman: CMAKE_ENV line removed; setting QMAKESPEC there has
no effect on the build system.
* devel/freeocl: CMAKE_ENV line removed; FREEOCL_CXX_COMPILER is already
retrieved from the CMAKE_CXX_COMPILER variable in the build
system.
* graphics/openimageio: CMAKE_ENV line removed; setting Qt variables there
has no effect on the build system.
Reviewed by: makc
Differential Revision: https://reviews.freebsd.org/D3403
This got broken by the way how the port detects that the system is
clang-based in order to force libc++, and should be fixed now.
PR: ports/186240
Reported by: Christoph Moench-Tegeder
Upstream changes are described at:
<http://rawtherapee.com/blog/rawtherapee-4.0.12-released>
[2] brings back OpenMP support and fully optimized sources, but it does
require GCC 4.8 to avoid internal compiler errors in older GCC releases,
and adds a build dependency on the libc++ port to bring in the required
compile-time namespace transformations to have Rawtherapee use libc++
(as the other C++-based requisite ports do).
Thanks to: Baptiste Daroussin (bapt@) for help with [2]
PR: ports/186240 [1]
Submitted by: Christoph Moench-Tegeder [1]
- Add desktop-file-utils to the USES list, bumping PORTREVISION.
- Force USE_GCC only on OSVERSION < 1000024, to avoid running into
a libc++/libstdc++ mismatch in glibmm-2.4 on FreeBSD 10
(which caused linker troubles in the Glib:: namespace due to
libc++-based different mangling - glibmm is compiled to use libc++
on 10-STABLE, so we cannot USE_GCC there).
- Force disable OpenMP option on non-GCC systems, because clang does
not yet support it.
- In order to be able to use clang, force ipsharpen.cc to be compiled
with -O1 only because -O2 or higher optimization levels lead to
excessive compile times, so long that I never waited for the compile
to complete (over 10 minutes on a Phenom II at 2.5 GHz).
- Quench unused-command-line-arguments warnings on clang.
This should fix the compile on 10-STABLE and HEAD.
Whether the software actually works I cannot currently test due to
lack of a full desktop install of 10-STABLE.
- Switched to automake 1.11.6, see CVE-2012-3386.
- #14669: Fixed extraction of CC from gmp.h.
- Fixed case of intermediate zero real or imaginary part in mpc_fma,
found by hydra with GMP_CHECK_RANDOMIZE=1346362345.
This is on top of the following changes from version 1.0
- Licence change towards LGPLv3+ for the code and GFDLv1.3+ (with no
invariant sections) for the documentation.
- 100% of all lines are covered by tests
- Renamed functions
. mpc_mul_2exp to mpc_mul_2ui
. mpc_div_2exp to mpc_div_2ui
- 0^0, which returned (NaN,NaN) previously, now returns (1,+0).
- Removed compatibility with K&R compilers, which was untestable due
to lack of such compilers.
- New functions
. mpc_log10
. mpc_mul_2si, mpc_div_2si
- Speed-ups
. mpc_fma
- Bug fixes
. mpc_div and mpc_norm now return a value indicating the effective
rounding direction, as the other functions.
. mpc_mul, mpc_sqr and mpc_norm now return correct results even if
there are over- or underflows during the computation.
. mpc_asin, mpc_proj, mpc_sqr: Wrong result when input variable has
infinite part and equals output variable is corrected.
. mpc_fr_sub: Wrong return value for imaginary part is corrected.
Convert to the new LIB_DEPENDS standard and remove hard-coded
.so versions from a couple of dependent ports.
Bump PORTREVISIONS of all dependent ports.
PR: 183141
Approved by: portmgr (bdrewery)
Right now this is a noop in the former case and a noop in the latter
case unless lang/gcc44 has been installed explicitly.
This puts a bit more emphasis on standardizing on a canonical version
"current" GCC and makes it easier to update that canonical version
by changing the default in Mk/bsd.gcc.mk and updating the lang/gcc port.
That is, USE_GCC=yes means "use a decent/modern version of GCC" without
having to worry about details.
Approved by: portmgr (bdrewery)
- change cmake to cmake:outsource to avoid a warning [1]
- paranoia: check rawtherapee --help if it prints the version number, to
make sure that it links to the right shared libraries (base libgomp is
insufficient), just in case RPATH propagation or setting breaks later
on. [by yours truly]
PR: ports/178206 [1]
Submitted by: Waitman Gobble <uzimac@da3m0n8t3r.com> [1]
This should fix the issues mentioned in the PRs given below,
in particular, it should do away with the need for libmap.conf hacks
for libgcc_s, libgomp, libobjc, libssp, libstdc++.
(the Makefile as I took it already set the linker's -rpath, and
patch-CMakeLists.txt ensures CMake keeps it during "make install").
ChangeLog 4.0.10: http://rawtherapee.com/blog/rawtherapee-4.0.10-released
ChangeLog 4.0.11: http://rawtherapee.com/blog/rawtherapee-4.0.11-released
PR: ports/171079
PR: ports/178206
are no more self hosting so we are stuck with 0.25 version while pkgconf provide
the same set of features as 0.27 and a compatible frontend. A symlink to
pkg-config has been added for convenience and compatibility
This also introduces a new macro to use pkgconf in your ports:
USE_PKGCONFIG
it can take the following arguments:
- yes (meaning build only dep)
- build (meaning build only dep)
- run (meaning run only dep)
- both (meaning run and build dep)
From now USE_GNOME= pkgconfig is deprecated in favour of USE_PKGCONFIG
The old gnome macro has been modified to use pkgconf but still the sameway: run
and build dep to avoid large breakage.
While here fix some ports relying on pkg-config but not specifying it, fix some
ports broken because testing wrong .pc files, and fix ports using pkg-config
--version to determine pkg-config version instead of
pkg-config --modversion pkg-config like recommanded by pkg-config
With Hat: portmgr
Exp-runs by: bapt (pointhat-west), beat (pointyhat)