Build failure on sparc64 with obj.as<bottom>

[jamessan]

I recently updated Debian's packaging to the 1.4.2 release and ran into this build failure on sparc64. All other architectures built fine.

object_with_zone.cpp: In member function 'virtual void object_with_zone_user_defined_non_virtual_Test::TestBody()':
object_with_zone.cpp:745:32: error: '*((void*)& v +16)' may be used uninitialized in this function [-Werror=maybe-uninitialized]
     bottom br = obj.as<bottom>();
                                ^
object_with_zone.cpp:745:32: error: '*((void*)& v +12)' may be used uninitialized in this function [-Werror=maybe-uninitialized]
object_with_zone.cpp:745:32: error: '*((void*)& v +8)' may be used uninitialized in this function [-Werror=maybe-uninitialized]
object_with_zone.cpp:745:32: error: '*((void*)& v +4)' may be used uninitialized in this function [-Werror=maybe-uninitialized]
object_with_zone.cpp:745:32: error: 'v' may be used uninitialized in this function [-Werror=maybe-uninitialized]

using this toolchain

Kernel: Linux 4.6.0-1-sparc64-smp sparc64 (sparc64)
Toolchain package versions: binutils_2.26-9+sparc64.2 dpkg-dev_1.18.7 g++-6_6.1.1-6 gcc-6_6.1.1-6

[redboltz]

@jamessan , thank you for reporting the issue. I've checked the code. In this case, the warning "maybe-uninitialized" is no problem.

So

#pragma GCC diagnostic ignored "-Wmaybe-uninitialized"

should be added like as follows:

msgpack-c/test/user_class.cpp

Line 154 in 3c997ed

#if (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6)) && !defined(__clang__)

[redboltz]

I created the PR #495 to suppress the warning. @jamessan , could you test it on your environment?

Kernel: Linux 4.6.0-1-sparc64-smp sparc64 (sparc64)
Toolchain package versions: binutils_2.26-9+sparc64.2 dpkg-dev_1.18.7 g++-6_6.1.1-6 gcc-6_6.1.1-6

I believe that the warning "maybe-uninitialized" is false positive in this case.

[jamessan]

👍

[redboltz]

Thank you for checking. I merged #495.

I recently updated Debian's packaging to the 1.4.2 release and ran into this build failure on sparc64. All other architectures built fine.

Is the issue the last problem to add msgpack-c to Debian? If it is, I will release the version 1.4.3 soon. If there are other tests, I will wait the result. And then I will release 1.4.3 that includes #495 and other fixes for the problems that might happen in the future.

[jamessan]

msgpack-c is in Debian's experimental release. I need to finish trying to get the rest of the packages which depend on libmsgpackc ready for the transition to the post-1.0 version of the library, before uploading it to sid. You can see the discussion/status in the transition tracker.

[redboltz]

Thank you for the comment. I understand the situation.

[jamessan]

The same change needs to be made to user_class.cpp.

[redboltz]

I sent #506 to fix the issue. Could you test it?

[redboltz]

merged.

[redboltz]

Fixed by #506