On 29/11/2018 12:16, John Paul Adrian Glaubitz wrote:

Hi!

On 11/29/18 10:21 AM, Normand wrote:

...

Since 20181119, there are more than 500 unresolvables in Leap:15.1:PowerPC (1) osc jobhist (2) show that at least yast2 packages need rebuilds

I'm maintaining multiple unofficial ports in Debian, among these is also Debian for 32-bit PowerPC. I would be interested in co-maintaining SUSE for 32-bit PowerPC as well.

One of the things that I wanted to change is to enable Rust for PowerPC as of recently we got it working in Debian again after helping upstream finding and fixing some issues in LLVM and Rust itself.

Adrian

Hello John Paul, If you want to help on Rust for PowerPC, then better to address that to the devel:languages:rust (1) that is related development project. I changed Subject title as your discussion is not about specific thread I am addressing before (about generic build problem in Leap:15.1:PowerPC project that is anyway ppc64le only. openSUSE Tumbleweed do not support anymore ppc 32-bit since a while. we still have ppc64 (BE) 64-bits and ppc64le 64-bits (2) (1) https://build.opensuse.org/project/show/devel:languages:rust (2) https://build.opensuse.org/project/show/openSUSE:Factory:PowerPC -- Michel Normand -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

On Nov 29 2018, John Paul Adrian Glaubitz wrote:

Well, Factory is building packages for ppc, ppc64 and ppc64le so it's not so relevant at the moment that there are no Tumbleweed releases for 32-bit PowerPC:

...

https://build.opensuse.org/project/show/openSUSE:Factory:PowerPC

If you want to work on ppc for Factory, the biggest obstacle right now is mariadb. Andreas. -- Andreas Schwab, SUSE Labs, schwab@suse.de GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE 1748 E4D4 88E3 0EEA B9D7 "And now for something completely different." -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

Hi Andreas! On 11/29/18 2:16 PM, John Paul Adrian Glaubitz wrote:

...

If you want to work on ppc for Factory, the biggest obstacle right now is mariadb.

Ah, I already guessed that. I saw the build was missing which makes quite a number of reverse dependencies unresolvable.

Ok, this is the same issue as on Debian [1, 2]: ../include/my_atomic.h:121:2: error: #error atomic ops for this platform are not implemented #error atomic ops for this platform are not implemented ^~~~~ I will have a look. I wonder why they're not the atomic stuff from C++11. Adrian

[1] https://build.opensuse.org/package/live_build_log/openSUSE:Factory:PowerPC/m... [2] https://buildd.debian.org/status/fetch.php?pkg=mariadb-10.3&arch=powerpc&ver=10.3.0-0%2Bexp2&stamp=1536336776&raw=0 -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

On Thu, 29 Nov 2018, 19:50:40 +0100, John Paul Adrian Glaubitz wrote:

On 11/29/18 7:38 PM, John Paul Adrian Glaubitz wrote:

...

On 11/29/18 2:16 PM, John Paul Adrian Glaubitz wrote:

...

...

If you want to work on ppc for Factory, the biggest obstacle right now is mariadb.

Ah, I already guessed that. I saw the build was missing which makes quite a number of reverse dependencies unresolvable.

Ok, this is the same issue as on Debian [1, 2]:

../include/my_atomic.h:121:2: error: #error atomic ops for this platform are not implemented #error atomic ops for this platform are not implemented ^~~~~

I will have a look. I wonder why they're not the atomic stuff from C++11.

Oh, they actually do that! Their test is just flawed in cmake:

CHECK_CXX_SOURCE_COMPILES(" int main() { long long int var= 1; long long int *ptr= &var; return (int)__atomic_load_n(ptr, __ATOMIC_SEQ_CST); }" HAVE_GCC_C11_ATOMICS)

On ppc64, this works:

glaubitz@redpanda:~/mariadb$ g++ cpp11test.cpp -o cpp11test glaubitz@redpanda:~/mariadb$ ./cpp11test

On powerpc, we need -latomic:

root@kapitsa:~# g++ cpp11test.cpp -o cpp11test -Wl,--as-needed /usr/bin/ld: /tmp/ccyivhlO.o: in function `main': cpp11test.cpp:(.text+0x48): undefined reference to `__atomic_load_8' collect2: error: ld returned 1 exit status root@kapitsa:~#

root@kapitsa:~# g++ cpp11test.cpp -o cpp11test -latomic root@kapitsa:~#

Now, I don't know from the top of my head how to fix the cmake file.

Any suggestions?

Does CHECK_CXX_SOURCE_COMPILES use anything like LDFLAGS which could be predefined in a host specific way?

Adrian

Cheers. l8er manfred

On 11/29/18 7:50 PM, John Paul Adrian Glaubitz wrote:

Now, I don't know from the top of my head how to fix the cmake file.

Found something from Debian [1]: # Check whether atomic operations require -latomic or not # See https://github.com/facebook/hhvm/issues/5217 INCLUDE(CheckCXXSourceCompiles) set(OLD_CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS}) set(CMAKE_REQUIRED_FLAGS "-std=c++1y") CHECK_CXX_SOURCE_COMPILES(" #include <atomic> int main() { struct Test { int val; }; std::atomic<Test> s; s.is_lock_free(); } " NOT_REQUIRE_ATOMIC_LINKER_FLAG) if(NOT "${NOT_REQUIRE_ATOMIC_LINKER_FLAG}") message(STATUS "-latomic is required to link hhvm") target_link_libraries(hhvm atomic) endif() set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS}) Adrian

[1] https://sources.debian.org/src/hhvm/3.24.7+dfsg-2/hphp/hhvm/CMakeLists.txt/?... -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

Hi! On 11/30/18 1:08 PM, John Paul Adrian Glaubitz wrote:

The attached proof-of-concept patch fixes mariadb for me on Debian unstable on powerpc. I was not able to make it build on openSUSE yet.

An improved patch which works against MariaDB 10.3 but not 10.2.x which requires more work as 10.2 doesn't fully use C++11 atomics yet. Adrian

Hi! On 11/30/18 9:50 PM, John Paul Adrian Glaubitz wrote:

I have a patch now which fixes the build on ppc but the testsuite failes two tests [1]. Patch attached but also available in my home project.

Ok, package is fixed for ppc and submitted now:

https://build.opensuse.org/request/show/653203

Adrian -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

On Nov 29 2018, Jan Engelhardt wrote:

Isn't the ppc target more or less just an ILP32 variant of a ppc64 anyway, similar to what x32 is to x86_64?

Not quite. The ppc target uses only the 32-bit subset of the powerpc ISA. Andreas. -- Andreas Schwab, SUSE Labs, schwab@suse.de GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE 1748 E4D4 88E3 0EEA B9D7 "And now for something completely different." -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

On Thursday 2018-11-29 15:36, John Paul Adrian Glaubitz wrote:

On 11/29/18 3:22 PM, Jan Engelhardt wrote:

...

...

Well, Factory is building packages for ppc, ppc64 and ppc64le so it's not so relevant at the moment that there are no Tumbleweed releases for 32-bit PowerPC:

Isn't the ppc target more or less just an ILP32 variant of a ppc64 anyway, similar to what x32 is to x86_64?

It's a different instruction set. x32 is basically x86_64 with 32-bit pointers with the full x86_64 instruction set available. 32-bit PowerPC is the PowerPC 74xx series (based on the POWER3 ISA IIRC) while 64-bit PowerPC is the 970 series (based on the POWER4 ISA).

For 32-bit SPARC, people usually use SPARCv9 (the first 64-bit UltraSPARC) with 32-bit pointers. So it's very much comparable with x32 and x86_64.

That last part has always been a bit strange, because sparc64-linux-gcc -m32 does not actually make use of the basic 64-bit instructions like ldx/stx, noticable for example when playing with uint64_t-based time. -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

On 03/12/2018 09:45, Ludwig Nussel wrote:

Normand schrieb:

...

Since 20181119, there are more than 500 unresolvables in Leap:15.1:PowerPC (1) osc jobhist (2) show that at least yast2 packages need rebuilds

I thought there was automatic rebuild of unresolvable in OBS, is it disabled for this project ?

The project is set to rebuild="local"¹. So rebuilds are only triggered on source changes or manually. There's also a script that can be used in rebuild=local mode to trigger rebuilds of unresolvables². It's not enabled for 15.1 PPC as it is only halfway safe with rebuild=local which it doesn't actually check for.

Nevertheless rebuild=local is not a clean way to build the distro. So unless you are very short on build power or have other constraints that outweigh the disadvantages of rebuild=local I would recommend to use at least rebuild=direct.

Michal, as per Ludwig suggestion, could you change meta to rebuild=direct, if you do not other see other drawback ? --- Michel Normand

[1] https://build.opensuse.org/project/meta/openSUSE:Leap:15.1:PowerPC [2] https://github.com/openSUSE/openSUSE-release-tools/blob/master/rebuildpacs.p...

    Volunteers to rewrite that script in a less hackish way welcome :-)

-- Michel Normand -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org