This work-around was added in commit
716f8b34c5 ("package/grpc: work around
gcc bug 85180") to work around a bug on Microblaze. However, since
commit
ae0557403a ("package/libabseil-cpp: add
BR2_PACKAGE_LIBABSEIL_CPP_ARCH_SUPPORTS"), grpc depends on
libaseil-cpp, which isn't available on Microblaze, making the
GCC_BUG_85180 no longer needed, and effectively dead code. This commit
drops it.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
Add new -DgRPC_BUILD_CODEGEN=OFF config option which is required if
no plugins are enabled.
Drop "add GPR_DISABLE_WRAPPED_MEMCPY" patch which is upstream.
Rebase remaining patches.
Signed-off-by: James Hilliard <james.hilliard1@gmail.com>
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
Fix the following build failure raised on uclibc and musl since the
reintroduction of the package in commit
16ff948444:
/home/buildroot/autobuild/instance-1/output-1/host/opt/ext-toolchain/bin/../lib/gcc/arm-buildroot-linux-uclibcgnueabi/10.3.0/../../../../arm-buildroot-linux-uclibcgnueabi/bin/ld: /home/buildroot/autobuild/instance-1/output-1/host/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/lib/libabsl_stacktrace.so.2111.0.0: undefined reference to `backtrace'
Fixes:
- http://autobuild.buildroot.org/results/63ab2bc86cad03d5258492b17d1707078761d9b3
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Currently, grpc depends on the full host-grpc, which in turn depends on host
versions of many other libraries. One of these, host-libabseil-cpp, also
requires a host gcc 4.9 or larger, a dependency which is not met on CentOS
7.
But in fact, the target grpc only needs the 'grpc_cpp_plugin' binary from
host-grpc. And that binary does not depend on host-libabseil-cpp or other
libraries, only on host-protobuf.
Given the above, simplify the grpc/host-grpc situation.
- Add a patch to the (host-)grpc CMakeLists.txt file to add an option to
only build grpc_cpp_plugin.
- Update grpc.mk and Config.in to remove the unnecessary dependencies, and
change the host-grpc configure options to make cmake happy.
The advantages of these changes are:
- making grpc available to older hosts with gcc < 4.8, like CentOS 7
- significantly reducing the build time of host-grpc and its dependencies
The patch was proposed upstream but not accepted with below rationale.
Perhaps input from others can help in persuading upstream in a future
attempt.
'What you're doing sounds like quite a narrow use case. But we simply
cannot provide a cmake option for every possible scenario in the world.
Introducing a new cmake option isn't for free and requires careful
design and maintenance.'
Nevertheless, given the benefits in terms of build time and dependency
reduction, it makes sense to apply this patch in spite of the disadvantage
of a local non-upstreamed patch.
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
[Arnout: propagate removed dependency to collectd]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
* update patch 0001 to match changed target code
* BSD-3c and MPL-2.0 licenses were added to LICENSE
Signed-off-by: Michael Nosthoff <buildroot@heine.tech>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Some of the third party code is BSD-licensed. In addition, the roots.pem
certificate store is MPL-licensed.
This was probably already the case in earlier versions as well, but it
was only noticed while updating to 1.42.0 because the LICENSE file was
adapted for it.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
* drop CMakeLists.txt patch applied upstream.
* Update patch for wrap_memcpy.cc to match changed target file.
* update patch numbering.
Signed-off-by: Michael Nosthoff <buildroot@heine.tech>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
grpc has plugins for multiple programming languages, which are needed on
development machines only. Examples are grpc_cpp_plugin, grpc_ruby_plugin,
etc.
Even though before commit fedf3318e3,
grpc_cpp_plugin was not installed for target, all other plugins still were.
This causes additional build time and rootfs space.
As Buildroot does not support building a development environment for target,
these tools can be disabled.
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
In commit fedf3318e3, an obsolete patch to
support cross-compilation was removed, in favor of the upstream solution.
However, this caused a small change in behavior: for the target grpc, the
tool 'grpc_cpp_plugin' is now also built, while before it was not.
This tool is only really needed on development machines. Since Buildroot
does not support compilers and such on target itself, the tool is not
needed.
There exists an option gRPC_BUILD_GRPC_CPP_PLUGIN which can be set to 'OFF',
but disabling it in a cross-compilation context yields build failures.
Add a patch to fix that. This patch is intended to be upstreamed to grpc.
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
When cross-compiling grpc, a native tool 'grpc_cpp_plugin' is needed.
Patch '0001-target-build-using-host-plugin.patch' in Buildroot provides a
way to pass the path to this tool via a configure option
'gRPC_NATIVE_CPP_PLUGIN'.
In version 1.20.0, the upstream grpc project added better support for
cross-compiling via commit 0d7a0ded [1], searching for the native
grpc_cpp_plugin via PATH (rather than specifying it as configure option as
our patch was doing).
This change renders the mentioned Buildroot patch obsolete, so remove it.
[1] https://github.com/grpc/grpc/commit/0d7a0ded1cc93bb7f4d69a156b0a69829557cbf2
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Reviewed-by: Michael Nosthoff <buildroot@heine.tech>
Tested-by: Michael Nosthoff <buildroot@heine.tech>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Due to libabseil dependencies the host gcc is at least 4.9.
So the fix for host gcc 4.8 is no longer needed.
Signed-off-by: Michael Nosthoff <buildroot@heine.tech>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
re2 unconditionally uses -pthread and add it to re2.pc
Moreover, it fails to buid without threads on:
In file included from /home/peko/autobuild/instance-0/output-1/per-package/re2/host/opt/ext-toolchain/arm-buildroot-linux-uclibcgnueabihf/include/c++/8.3.0/arm-buildroot-linux-uclibcgnueabihf/bits/os_defines.h:39,
from /home/peko/autobuild/instance-0/output-1/per-package/re2/host/opt/ext-toolchain/arm-buildroot-linux-uclibcgnueabihf/include/c++/8.3.0/arm-buildroot-linux-uclibcgnueabihf/bits/c++config.h:508,
from /home/peko/autobuild/instance-0/output-1/per-package/re2/host/opt/ext-toolchain/arm-buildroot-linux-uclibcgnueabihf/include/c++/8.3.0/bits/stl_algobase.h:59,
from /home/peko/autobuild/instance-0/output-1/per-package/re2/host/opt/ext-toolchain/arm-buildroot-linux-uclibcgnueabihf/include/c++/8.3.0/memory:62,
from /home/peko/autobuild/instance-0/output-1/build/re2-2020-08-01/re2/filtered_re2.h:24,
from /home/peko/autobuild/instance-0/output-1/build/re2-2020-08-01/re2/filtered_re2.cc:5:
/home/peko/autobuild/instance-0/output-1/per-package/re2/host/arm-buildroot-linux-uclibcgnueabihf/sysroot/usr/include/features.h:218:5: warning: #warning requested reentrant code, but thread support was disabled [-Wcpp]
# warning requested reentrant code, but thread support was disabled
^~~~~~~
In file included from /home/peko/autobuild/instance-0/output-1/build/re2-2020-08-01/re2/prog.h:22,
from /home/peko/autobuild/instance-0/output-1/build/re2-2020-08-01/re2/bitstate.cc:28:
/home/peko/autobuild/instance-0/output-1/build/re2-2020-08-01/re2/re2.h:768:16: error: 'once_flag' in namespace 'std' does not name a type
mutable std::once_flag rprog_once_;
^~~~~~~~~
/home/peko/autobuild/instance-0/output-1/build/re2-2020-08-01/re2/re2.h:768:11: note: 'std::once_flag' is defined in header '<mutex>'; did you forget to '#include <mutex>'?
/home/peko/autobuild/instance-0/output-1/build/re2-2020-08-01/re2/re2.h:218:1:
+#include <mutex>
Fixes:
- http://autobuild.buildroot.org/results/7d7c6dcac3cb8ea6deb753178e85eb1c5c74c8e3
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Reviewed-by: Michael Nosthoff <buildroot@heine.tech>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Related changes:
- add dependency on Google RE2 package
- update patches to new offsets
Tested on Ubuntu 20.04
Signed-off-by: Michael Nosthoff <buildroot@heine.tech>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
The libabseil-cpp package fails to build on a number of CPU
architectures in our autobuilders.
On most CPU architectures, the first issue looked like this:
libabseil-cpp-20200225/absl/base/internal/direct_mmap.h: In function 'void* absl::lts_2020_02_25::base_internal::DirectMmap(void*, size_t, int, int, int, off64_t)':
libabseil-cpp-20200225/absl/base/internal/direct_mmap.h:121:39: error: static assertion failed: Platform is not 64-bit
121 | static_assert(sizeof(unsigned long) == 8, "Platform is not 64-bit");
| ~~~~~~~~~~~~~~~~~~~~~~^~~~
libabseil-cpp-20200225/absl/base/internal/direct_mmap.h:123:15: error: 'SYS_mmap' was not declared in this scope; did you mean 'SYS_mmap2'?
123 | syscall(SYS_mmap, start, length, prot, flags, fd, offset));
| ^~~~~~~~
| SYS_mmap2
Indeed, on 32-bit architectures, libabseil-cpp has some special code
to use the mmap2() system call, and it white-lists the supported
architectures. It is therefore trivial to add support for more
architectures.
However, once this is fixed, another issue arises:
absl/debugging/internal/examine_stack.cc uses the ucontext data
structures, which are not provided by uClibc-ng on all CPU
architectures, and even the code of libabseil-cpp does not exist for
all CPU architectures.
So, this commit solves that by simply making libabseil-cpp available
on architectures/C libraries where it is supported: it needs ucontext
support in the toolchain + a CPU architecture where
absl/debugging/internal/examine_stack.cc has the appropriate logic.
This new dependency is propagated to the reverse dependencies of
libabseil-cpp.
With this commit, libabseil-cpp passes a test-pkg -a test (so all
external toolchains used by the autobuilders):
andes-nds32 [ 1/45]: SKIPPED
arm-aarch64 [ 2/45]: OK
br-aarch64-glibc [ 3/45]: OK
br-arcle-hs38 [ 4/45]: SKIPPED
br-arm-basic [ 5/45]: SKIPPED
br-arm-cortex-a9-glibc [ 6/45]: OK
br-arm-cortex-a9-musl [ 7/45]: OK
br-arm-cortex-m4-full [ 8/45]: SKIPPED
br-arm-full [ 9/45]: OK
br-arm-full-nothread [10/45]: SKIPPED
br-arm-full-static [11/45]: SKIPPED
br-i386-pentium4-full [12/45]: OK
br-i386-pentium-mmx-musl [13/45]: OK
br-m68k-5208-full [14/45]: SKIPPED
br-m68k-68040-full [15/45]: SKIPPED
br-microblazeel-full [16/45]: SKIPPED
br-mips32r6-el-hf-glibc [17/45]: OK
br-mips64-n64-full [18/45]: OK
br-mips64r6-el-hf-glibc [19/45]: OK
br-mipsel-o32-full [20/45]: OK
br-nios2-glibc [21/45]: SKIPPED
br-openrisc-uclibc [22/45]: SKIPPED
br-powerpc-603e-basic-cpp [23/45]: SKIPPED
br-powerpc64le-power8-glibc [24/45]: OK
br-powerpc64-power7-glibc [25/45]: OK
br-powerpc-e500mc-full [26/45]: SKIPPED
br-riscv32 [27/45]: OK
br-riscv64 [28/45]: OK
br-riscv64-musl [29/45]: OK
br-sh4-full [30/45]: SKIPPED
br-sparc64-glibc [31/45]: SKIPPED
br-sparc-uclibc [32/45]: SKIPPED
br-x86-64-core2-full [33/45]: OK
br-x86-64-musl [34/45]: OK
br-xtensa-full [35/45]: SKIPPED
linaro-aarch64-be [36/45]: OK
linaro-aarch64 [37/45]: OK
linaro-arm [38/45]: OK
sourcery-arm-armv4t [39/45]: OK
sourcery-arm [40/45]: OK
sourcery-arm-thumb2 [41/45]: OK
sourcery-mips64 [42/45]: OK
sourcery-mips [43/45]: OK
sourcery-nios2 [44/45]: SKIPPED
sourcery-x86-64 [45/45]: OK
45 builds, 18 skipped, 0 build failed, 0 legal-info failed
Fixes:
http://autobuild.buildroot.net/results/ead663b4b67b0b57ed003a46db3182d95cc01bc0/
(and many similar build failures)
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Other changes:
- Add a dependency on libabseil-cpp
- Update the patches to apply properly.
Tested with the following distributions:
- Debian 9
- CentOS 7
- Fedora 32
Signed-off-by: Adam Duskett <Aduskett@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Other changes:
- Remove upstream patch 0004-Fix-gettid-naming-conflict.patch
- Remove upstream patch 0005-Rename-gettid-functions.patch
- Add a check for BR2_TOOLCHAIN_GCC_AT_LEAST_5 in grpc.mk. If the
selected toolchain is not at least version 5 or higher and the
optimization level is -Os, set the GRPC_CFLAGS and GRPC_CXXFLAGS
optimizations to -O2. This check prevents the following error:
error: failure memory model cannot be stronger than success memory model for '__atomic_compare_exchange'
Tested with test-pkg, all tests passed:
br-arm-full [1/6]: OK
br-arm-cortex-a9-glibc [2/6]: OK
br-arm-cortex-m4-full [3/6]: SKIPPED
br-x86-64-musl [4/6]: OK
br-arm-full-static [5/6]: SKIPPED
Signed-off-by: Adam Duskett <Aduskett@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Michael Nosthoff